70.1 February 2023

Elements of an Emergent Style Guide for Kickstarter

doi.org/10.55177/tc679601

By Stephen Carradini and Eric Nystrom

ABSTRACT

Purpose: Crowdfunding campaigns are a way to secure capital for organizations, entrepreneurs, artists, and more. Little research has focused on stylistic aspects of text associated with successful campaigns.

Method: We used corpus analysis to analyze the text of 312,529 Kickstarter campaigns. We used a novel scoring method to compute how often verbs and words surrounding verbs were associated with success or failure. We then identified prominent stylistic aspects of text that were often included in successful and unsuccessful campaigns.

Results: Stylistic elements strongly associated with success included using we instead of I, using contractions instead of full forms of verbs, inviting the reader to join the project and receive rewards, and projecting confidence via will instead of the more uncertain would.

Conclusion: Stylistic findings interact. Specifically, using we and contractions together indicates outcomes strongly associated with success. Broadly, each of the findings point toward creators of campaigns attempting to build trust in the readers. The elements of this emergent style work together toward a goal of producing campaign text that describes a campaign readers accept and trust as likely to succeed.

Keywords: Style, Style guide, Kickstarter, Crowdfunding, Verbs

Practitioner’s Takeaways

  • Using we as the subject of a sentence was often more strongly associated with successful campaigns than using I.
  • Using a contraction was often more strongly associated with successful campaigns than the full forms of the same words.
  • Pairing we with a contraction (e.g., we’re) was often strongly associated with successful campaigns.
  • Telling readers what to expect when the campaign succeeds was often associated with success for writers using we and writers using I as the subject of sentences.
  • Inviting readers to join the campaign was a strong tactic.

INTRODUCTION

Technical communicators often work closely with style guides because the style guide is the “rule-driven document that sets the parameters for consistency and acceptability of all written materials produced by an individual or group” (Adhya, 2015, p. 184, emphasis ours). Organizational style guides ensure that employees’ written materials are deemed acceptable by the organization before publication. Yet social media platforms, one outlet of publication for technical communicators’ work, offer limited guidance regarding what they deem acceptable word choice, grammar, and punctuation on their platforms. Furthermore, users may have expectations of what acceptable text should look like from an organization on social media. Given this context of ambiguity in which social media content is published, understanding what acceptable text looks like for organizations in specific social media platforms would give the technical communicator much more guidance than is currently offered for many platforms.

Kickstarter is a social media platform that supports crowdfunding, an activity that allows creators of new products, services, or creative activities to seek venture funding from the public via online donations (Belleflamme, Lambert, & Schwienbacher, 2014; Kickstarter, PBC, 2021). Like most social media platforms, Kickstarter offers limited stylistic advice to those who would crowdfund an idea. Furthermore, the success of a Kickstarter campaign depends on the creator’s ability to convince the audience to give money to the campaign through the platform. Understanding what types of stylistic markers are accepted by users in campaigns would allow technical communicators to employ text accepted by readers and increase their chances of successful campaigns.

Thus, this study uses computer-aided corpus analysis to identify users’ expectations for Kickstarter campaign text. We identify stylistic elements frequently featured in campaigns that users accepted (successful campaigns) and rejected (unsuccessful campaigns). The binary nature of Kickstarter makes this method possible—with few exceptions, a campaign either succeeds or fails to reach its financial goals. As a result, a corpus analysis can identify elements of style included in successful or unsuccessful campaigns. Given the context of campaign creators trying to convince readers to fund a speculative venture, we found that readers accept certain stylistic conventions that function as indicators of trustworthiness. Creators build trust in campaign readers by:

  • using we instead of I,
  • speaking at a level of formality acceptable to Kickstarter readers' expectations,
  • inviting readers to join the campaign, and
  • projecting confidence.

These stylistic elements work together to further the goal of building trust in the reader and encouraging the reader to accept the creator as competent to complete the campaign (via delivering the project and fulfilling promised rewards). We close by collecting conventions into an emergent style guide for the Kickstarter platform.

Entrepreneurship

Crowdfunding is prominently (but not exclusively) associated with entrepreneurship, a topic of interest in the last decade of technical communication research (Spinuzzi 2016; 2017). From a practitioner standpoint, entrepreneurship—in the form of independent work/contracting, small business owning, and other types of labor arrangements—has been a viable, at times desirable, and sometimes necessary option for technical communicators since at least the early 2000s. Lauren and Pigg (2016) argued that “despite growing labor statistics and narratives that describe practices that enable independent work, TC still needs to systematically investigate the phenomenon to better understand it’’ (p. 343). From a research standpoint, Spartz and Weber (2014) note that technical communicators and entrepreneurs conduct similar rhetorical work. For example, entrepreneurs create and revise genres such as the oral pitch and the pitch deck in response to the rhetorical situation and audience input (Spinuzzi et al., 2015), just as with technical documents and presentations. Furthermore, entrepreneurship and technical communication can converge in a single company, such as in the technical marketing communication of the aviation startup studied by Mara (2008). Beyond an interest in the genres of entrepreneurial activity, technical communication scholars have investigated the experiences of Black entrepreneurs (Jones, 2017) and entrepreneurs beyond North American borders (Fraiberg, 2021). Yet new technologies are changing the work and experiences of entrepreneurship:

The context of 21st-century entrepreneurship is rapidly changing due to the increasing popularity of new methods such as crowdfunding. Traditional genres, like the elevator pitch or business plan, … have increasingly become subsumed into emergent methods. Elevator pitches, for instance, have been replaced by viral videos; while the basic structure might be familiar, the kinds of writing necessary to invent and circulate such videos are drastically different (Vealey & Gerding, 2016, p. 414).

Online genres and technologies, as demonstrated generally in social media and specifically in crowdfunding, require entrepreneurs to enact new types of writing to be successful. Writing, whether in new forms or old, requires content and style. In this article, we focus on style, so we now turn our attention toward what is known about style in social media.

Social Media Style

Social media platforms have become valuable resources for organizations to communicate to their constituencies and for people to communicate back to the organization (Fan & Gordon, 2014). Yet in-house organizational style guides offer much more detail to employees of organizations than social media platforms do.

Social media platforms govern acceptable content[1] more often than acceptable style. The most prominent way platforms enforce acceptable content is daily content moderation, where moderators take action against content that violates the platform’s content guidelines (Gallagher et al., 2020). More subtle forms of governance exist. Wang and Gu (2016) demonstrated how platforms signal acceptable types of content through design decisions, explaining that WeChat’s communicative features shape and protect the possible content of users in ways that reinforce the designers’ goals of “free dissemination of information and public involvement” (p. 23). Design can also shape content in negative ways: elements of Facebook’s design encourage content that reflects “sensationalism, controversy, drama, intrigue, as well as feelings of amusement, anxiety, fear, and suspicion over curiosity, empathy, understanding, or kindness” with the overarching goal of monetizing the platform (Sano-Franchini, 2018, p. 401). Platforms can also suggest acceptable content as examples: the DIY repair wiki iFixIt “clearly lays out the content models for a variety of content types” when users begin to create content for the wiki (Getto & Labriola, 2019, p. 392).

Compared to published content guidelines, social media platforms offer limited governance of platform-appropriate word usage, punctuation use, grammar, and text formatting. Twitter famously began by only allowing posts of 140 characters due to technical concerns (King, 2009), but later raised the cap to 280 characters in a stylistic choice made to help people avoid “cramming” text into the small character count (Rosen, 2017). Further examples of style imposed via interface include Instagram’s character limits on user bios (Meta, 2022) and caption requirements for certain types of YouTube videos (Google, 2022). These guidelines apply across the whole platform for professional and personal users.

Professional users of social media must contend with user expectations as well as platform expectations because social media “texts are partially constrained by the platform guidelines as well as community expectations” (Pope, 2018, p. 494). Users of social media platforms have expectations for what acceptable content by professional organizations looks like on social media (Martin, Greiling, & Wetzelhütter, 2018). For example, West found that students have “their own personal ideas about how social media could and should be used, and they often attempted to map those uses onto businesses and organizations” (2017, pp. 413–414).

Understanding these often-unwritten social expectations of a platform can be a challenge for professionals and organizations. Much ineffective social media use and many social media crises come from organizations producing text and content that users find unacceptable, such as when users rebuffed Qantas Airlines’ seemingly innocuous tweet about a giveaway contest as insensitive toward an ongoing, unrelated corporate crisis (Bowdon, 2014). Furthermore, users may feel that acceptable language may differ for professional content and personal content (Liederman & Perloff, 2022). Consider Figure 1, which displays a prominent case: Twitter users perceived Chase Bank’s professional use of a sarcastic joke format popular on social media as poor-shaming, with users as prominent as United States Senator Elizabeth Warren tweeting back disapproval (Molina, 2019). While expectations for the differences between professional and personal content could be discussed in organizational style guides, technical communicators working without a social media section in an organizational style guide must investigate and understand users’ social expectations for professional content in individual platforms.

Crowdfunding

Kickstarter is one of the most prominent examples of a crowdfunding platform. While models of funding vary among platforms, Kickstarter follows the All-or-Nothing model (Cumming, Leboeuf, & Schwienbacher, 2015). Kickstarter campaigns have a financial goal that the campaign creator designates in advance (Mollick, 2014, p. 5). If funding comes in from “‘family,’ ‘friends,’ ‘professional acquaintances,’ ‘previously known fans,’ and ‘previously unknown backers’” to reach or exceed the goal, the Kickstarter campaign is successful, and the creator gets the money to make the project (Davidson & Poor, 2015, p. 296). If the funding does not come in to reach the goal, the campaign fails, and the creator does not get any of the pledged money. This binary result makes Kickstarter campaigns high-stakes efforts; failed campaigns can result in wasted time and effort.

Yet the stakes are not just high for the creator. The people who fund campaigns (which Kickstarter calls backers) contribute money based on their interest in the project and estimation of its success. If a campaign meets its funding goal but the creator fails to complete the project, the backers will not get the promised outcome of the project, may not get the rewards offered, and likely will not get their money back. Kickstarter frames this relationship between creators and backers on their Trust and Accountability page:

  • “It’s the creator’s responsibility to bring a project to life and deliver the rewards they’ve promised to backers.
  • It’s the backer’s responsibility to determine whether to support a project. When they do, it’s their responsibility to deliver that support, regardless of the project’s outcome.
  • Rewards aren’t guaranteed by Kickstarter or the creator” (Kickstarter, PBC, 2022a).

The role of backers in Kickstarter campaigns—according to Kickstarter—is to determine whether to support a project in a speculative context where a poor decision based on misplaced trust may mean lost money, no project outcomes, and no rewards. In this context, creators must effectively convince the reader to trust that the creator can “bring a project to life” (Kickstarter, PBC, 2022a). The stakes are high for the creators and the backers.

As a result of these stakes, understanding the characteristics of successful campaigns has become a critical concern for researchers (Hu & Yang, 2020). This type of analysis reached a high point in Ryoba et al. (2021), who suggested that a collection of only nine elements can be used to predict the success or failure of a (completed) Kickstarter campaign: country, number of images, number of videos, number of words in a project, number of updates to the project, number of comments on a project, presence of FAQs in a project, cost level of rewards for users (also known as pledge levels), and number of people backing the project. The campaign creator cannot control some of these backward-looking indicators of success, such as the number of people backing the project and the number of comments on a project. Creators can control some indicators, such as the number of videos (Cudmore & Slattery, 2019), the number of images (Carradini & Fleischmann, 2022), and the number of words in a project (Thapa, 2020). For technical communicators, these findings form technical guidance regarding multimedia inclusion but do not give much guidance on the writing style.

Even though the quantity of multimedia elements in a campaign can help predict success, Mitra and Gilbert (2014) found elements of text to be an even stronger predictor of success. The authors analyzed 59 control variables and 20,391 phrases to find that “the top 100 predictors of funded and not funded are solely comprised of phrases” such as “mention your,”pledgers will,” and “we can afford” instead of variables such as the presence of a video in the campaign, the number of updates provided to backers, whether the project was connected to a Facebook account, and how long the campaign ran (pp. 54, 57, italics original throughout). Thus, creators should be interested in what textual elements associate with success in Kickstarter campaigns.

Textual Analysis of Kickstarter Campaigns

Researchers have identified various textual elements associated with successful and unsuccessful campaigns (Sandouka, 2019). Tirdatov (2014) analyzed appeals in the 13 most-funded Kickstarter campaigns at that time, revealing 12 appeal subtypes divided among ethos, logos, and pathos appeals. The author found that “all of the thirteen most-funded projects contained all three types of Aristotle’s rhetorical appeals (with varying combinations of appeal subtypes), a situation not necessarily common to all kinds of rhetorical discourse” (p. 21). Ishizaki (2016) used computer-aided rhetorical analysis to identify rhetorical strategies of successful and unsuccessful campaigns from 901 campaigns in the technology category: “The rhetorical strategy that emerges… is to use a confident voice and provide significant expert details with clear explanations of what the product enables. On the other hand, failed pitches seem to address more general non-expert audience with less specialized details” (p. 3; cf. Koch & Siering [2015] for a concurring assessment on a detailed proposal).

Other advice to writers drawn from research includes:

  • Convey reciprocity, scarcity, social proof, social identity, likeability, and authority (Mitra & Gilbert, 2014);
  • Highlight that the product is novel or useful, but not both useful and novel (Mukherjee et al., 2017);
  • Avoid self-disclosure, raising questions, and apologizing (Grebelsky-Lichtman & Avnimelech, 2018); and
  • Employ language in the text reflective of hope, optimism, resilience, and confidence (Anglin et al., 2018; cf. Ishizaki [2016] regarding confidence).

These texts point to argumentation strategies, concepts, and word types that authors can use. While Anglin et al. (2018) come close to style when suggesting effective categories of words, several studies have looked at grammar, punctuation, and specific word use directly. Ishizaki (2016) offered a grammar suggestion (“many successful pitches also use the second person … to help the reader envision the use of their product”) as well as a suggestion that quotation marks are “used by many successful pitches as well, including direct quotes and special terms that are double-quoted” (p. 3). Butticè and Rovelli (2020) suggested avoiding first-person singular pronouns, while Grebelsky-Lichtman and Avnimelech (2018) suggested using first-person plural pronouns like us and we instead. These studies identify elements of style but do not bring the findings together as a whole and suggest how these findings might work together to advance the goals of the creator in creating trust.

These findings, spread across many studies, suggest potential aspects of style that often appear in successful Kickstarter campaigns. Due to the high-stakes nature of crowdfunding, a large body of work has grown around determining the successful elements of Kickstarter campaign writing. These individual findings provide guidance on a range of specific textual issues. Yet questions about crowdfunding persist: “as crowdfunding continues to evolve at a rapid pace, more research is needed to better understand how different permutations of the genre make use of writing and multimodal elements to persuade potential backers” (Pope, 2018, p. 496).

Research Gap and Research Question

Questions of nuance, corpus size, generalizability, and interaction among findings remain regarding textual analysis of Kickstarter campaigns. Some analyses leave open questions of nuance: Butticè and Rovelli (2020) questioned whether the correlation of singular first-person pronouns with failed campaigns reflects creators’ narcissism or something else. Some studies considered small numbers of campaigns: Lins, Fietkiewicz, and Lutz (2018) considered 195,217 words from 221 campaigns, while Koch and Siering (2015) analyzed 762 campaigns. Concerns about the generalizability of findings appear in two ways. Grebelsky-Lichtman and Avnimelech (2018) selected campaigns regarding the topics of 3-D printers, mobile applications, iPhone stands, and organic food; positive or negative findings from these categories may not be representative of trends across all topics on the platform. Tirdatov (2014) approached the problem of topic breadth by analyzing the most-funded campaigns from 13 project categories. While solving the problem of scope, text from the most-funded studies may not be representative of the text from average campaigns. Studies that identify nuanced patterns of text from a large corpus would help technical communicators write in broadly stylistically acceptable ways on the Kickstarter platform. Finally, few studies show how stylistic elements interact in text or how creators can use stylistic elements together to work toward a goal.

Given a lack of style guides for social media platforms, the existence of users’ expectations for professional content online, the high stakes of crowdfunding campaigns, and prior research on the subject, we developed a research question: What stylistic cues, as instantiated in patterns of text, are prominently associated with success in a broad sample of Kickstarter writing? To investigate this question, we used corpus analysis to identify textual patterns associated with successful or unsuccessful campaigns from 312,529 Kickstarter campaigns.

METHOD

This project used a multi-step, computer-aided process to develop and analyze a corpus of text representing the body text of Kickstarter campaigns. Technical communication scholars have used corpus analysis to analyze Kickstarter campaigns before. For example, Ishizaki (2016) investigated 901 campaigns; we expand on Ishizaki’s approach by investigating 312,529 campaigns.

To begin the project, we scraped data from Kickstarter by modifying the software tool Quickscrape to gather data from as many campaigns as possible (Shuttleworth Foundation, 2014). Scraping is a common and viable method for researchers to assemble a significant corpus (Gallagher & Beveridge, 2022; Ryoba et al., 2021) and has been deemed permissible by American courts (Tse & Bryan, 2022). A hired coder used Quickscrape to systematically search through the pages of Kickstarter’s discovery feature for campaign URLs in category/location pairs that Kickstarter allows (such as music projects in Montana or technology projects in Europe). While rare, type/region pairs containing more than 200 pages of campaign listings could not be scraped in their entirety. After collecting campaign URLs from each listing page, the coder used a scraping tool to harvest text, HTML tags, and embedded URLs of the campaign. Some pages presented atypically and could not be scraped effectively. The scraper ran for 24 days (May 31, 2018 to June 23, 2018). This scraper is available in a public GitHub repository accessible from the first author’s website (Carradini, 2022).

In 2019, the authors scraped the URLs collected in 2018 a second time to fill in missing metadata. The second scrape iterated over the earlier list of URLs, using curl (a common scriptable website downloading tool) to save the entire HTML output of each page and a JSON object containing campaign metadata embedded in each page. A JSON object is a specially-formatted data record composed of regular text that can be read semantically by computer programs but also manipulated and transmitted as text, i.e., without special network or database connections. In each campaign’s page, Kickstarter embedded a JSON object containing extensive metadata about that campaign. We subsequently extracted these JSON objects from the saved HTML pages using text-processing tools, forming the basis of our second dataset. We could not scrape the text of the campaigns during the second scrape because of changes in Kickstarter’s server configuration. Thus, we combined the full text of each campaign captured by the first scrape with metadata from the second scrape (e.g. amount of money raised, location of campaign, project category) for our complete analysis.

Data Cleaning

connect each campaign’s data across the two scraping efforts. A primary key, as understood in data science, consists of a unique identifier, which would identify a single campaign and no other. Our identifier used a combination of User ID and the campaign name, combined with an underscore (a character not permitted in either field and thus safe as a data separator). This information existed in both scrapes, and did not require the construction of an outside reference or tracking list (as would have been the case if we invented our own numbering scheme). An example primary key is cooper_sushi-cop, representing the campaign titled Sushi Cop by user Cooper. Despite their simplicity, primary keys are powerful and flexible data structures for text-based data, because a data table might hold many dozens of records associated with each individual primary key. A record can thus represent only a small fragment of data from the campaign, but can support extensive analysis in combination with other records from the same campaign using the same primary key.

Using the primary keys as a foundation, we built custom text processing tools (that we describe further below) to manipulate these datasets as streams of data, with the intent that each tool’s output could provide a platform for the next data exploration. We only manipulated text using scripted tools, rather than interactive programs, so that our data transformations could be reproduced at any time with precisely the same results.

To clean the data, we eliminated broken or empty records. We also removed data from campaigns that were active at the time of the first scrape, because these did not reflect the conditions of successful/unsuccessful that we planned to analyze. For scrape 1, we generated primary keys as described above, then lightly cleaned the full text, including removing a common warning code about HTML5 that was an artifact of the scraping process and not part of the actual campaign text. For scrape 2, little cleaning was needed due to the structured nature of the data: we extracted the campaign metadata from the JSON object embedded in each page (checking for malformed JSON and fixing if necessary), generated the primary key, and saved output data files containing the primary keys and fields of specific interest.

We ultimately used text and metadata from 312,529 campaigns. Of those, 129,413 campaigns succeeded (41.4%); 155,438 failed (49.7%); and 27,678 (8.9%) reported some other status, such as canceled or suspended. Our corpus is a larger corpus than previous corpora used for textual analysis. Early prominent research studied less data: Mollick’s (2014) study analyzed 48,526 campaigns, while Mitra and Gilbert’s (2014) analyzed 45,815 campaigns. Ryoba et al. (2021) originally used a dataset that included 424,980 campaigns, but they ultimately analyzed the word count in 21,885 campaigns instead of alphanumeric text. We contend that a larger corpus increases the ability to draw widely applicable conclusions based on the text of Kickstarter campaigns.

Data Analysis

Our analysis relies on short combinations of words (called tokens) to enable analysis at scale. In contrast to previous work, we formed these tokens from words connected to each other by grammatical relationships instead of simple proximity (cf. Mitra & Gilbert, 2014). Instead of these three-word patterns, our approach sought to use verbs as the central item of our tokens because “verbs are the most central elements of clauses. They can express actions, events, states of affairs, and the like” (Aarts, 2011, p. 65).

To identify verbs, we ran a probabilistic grammar tagger library named Lingua::EN::Tagger (Coburn, 2019) on the text of each Kickstarter campaign. Lingua::EN:Tagger labels each word with a part of speech, using a modified version of the commonly-used Penn Treebank schema (Marcinkiewicz, 1994) for its tag set (See Appendix A for the list of tagged features). The tagger did not stem words but did separate possessive endings and contractions and then tagged those as separate words. The tagger distinguishes several categories of verbs: infinitive verbs, past tense verbs, gerund verbs, past/passive participle verbs, base present form verbs, present third person singular +s verbs, and modal verbs (See Appendix A for full description and examples). We included all words the tagger identified with these categories when constructing our verb pseudo-phrases.

Because the tagger only tags individual words, we constructed verb-focused pseudo-phrases for analysis by combining adjacent tagged verbs. In constructing verb pseudo-phrases, we looked for words with any of these seven verb tags. We also included the preposition to as part of the pseudo-phrases representative of the infinitive form. We built the tokens from the inside out, beginning with a word tagged as a verb that moves backward and forward in the sentence. We discarded non-verb words adjacent to the verb, such as conjunctions, adjectives, and determiners. If we found additional verbs (or to), we added them to the verb pseudo-phrase.

Once we completed the verb pseudo-phrase, we identified the nouns on each side of the verb pseudo-phrase as the next step in token construction. For the purposes of token construction, we considered nouns to be anything tagged as such, including proper and plural nouns, plus possessive and second possessive determiners such as mine, yours, their, and your (tags NN, NNP, NNPS, NNS, PRP, and PRPS; see Appendix A). In this practice, we follow Aarts’ (2011) classification of pronouns as a subclass of nouns: “Pronouns belong to the class of nouns because they can head noun phrases that can function as Subject, Direct Object, Indirect Object, Complement of a preposition…, and Predicative Complement” (Aarts, 2011, p. 44, capitalization original). We excluded the DET tag in our list, representing determiners (e.g., this, each, some). We felt these determiners failed to give us enough information to know what this or each represented in the context of the sentence. Thus, these words did not function in the same way as words we saw as subject Ns.

Hunting for components of the verb pseudo-phrase stopped when we encountered a noun, determiner, sentence beginning, or sentence end. We made no attempt to construct the tokens with noun phrases, as an experiment along these lines showed that the number of unique tokens would have ballooned. The number of shared tokens necessary for our discussion of shared language would have decreased substantially.

When we combine the words together, we create a modestly-sized token consisting of a noun, one or more verbs as a pseudo-phrase, and another noun (henceforth: NVVNs). This token is a rough approximation of a basic sentence structure, necessitated by the importance of keeping tokens short to permit computation and comparability. Still, tokens depict a far more powerful representation of the intent and construction of any particular sentence than could be captured by a bag-of-words approach characterized by any-three-consecutive-word tokens utilized in previous work.

This approach resulted in 19,208,610 NVVN uses, each connected to the campaign from which it originated by the primary key. We then removed NVVNs that appeared in only a single campaign, because we wanted to study language shared across campaigns. The final data contained 1,371,258 distinctive NVVNs appearing in two or more campaigns. See Table 1 for a list of most common NVVNs.

We constructed NVVNs using underscores to separate the part-of-speech positions within the token and hyphens to connect multiple words (if present) in the verb pseudo-phrase: we_are-asking_you. Generation of NVVNs respected sentence boundaries, so, not every NVVN actually includes two nouns: _thank_you (note leading underscore) indicates the NVVN features no opening, because thank likely opened the sentence in the campaign; similarly, it_happen_ (as in let’s make it happen!) has a trailing underscore indicating no trailing noun exists. NVVNs which had no identifiable tagged noun prior to the beginning of the verb pseudo-phrase (such as a sentence with a determiner in the subject) have an underscore as the opening symbol: _is_my. The tagger identified verbs wherever they were in the sentence, and we built NVVNs from each of the verbs tagged; for example, the it_happen_ NVVN could be built from let’s make it happen! As a result, not all NVVNs represent nouns surrounding the sentence’s primary verb. NVVNs also captured all verbs between nouns, which means the verb pseudo-phrases we analyzed could range in size from one to many words: go or will-be-used-to-help.

One distinctive methodological element concerns cross-referencing text with success or failure, then mapping those values. To do this, we pulled the campaign status information from our second scrape and assigned the status of each campaign to every NVVN associated with it. We coded campaign success as +1, failure as -1, and other status (canceled, purged, or suspended) as 0. We totaled the uses of each textual object (any NVVN, or a component of one, such as a verb pseudo-phrase or noun) and generated an average success score (a number we call score throughout) that reflects its lean. If we found more instances of a textual object such as an NVVN in unsuccessful campaigns, the overall score for that NVVN would be negative; if we found more instances of the NVVN in successful campaigns, the score would be positive. We truncated calculated scores after the third number for clarity (for example, -.178 or +.233 instead of +.233857364859). Table 2 includes a list of highest and lowest scores.

Our sample included an unequal number of successful and failed campaigns. On average, failed campaigns contained only about two-thirds as much text as successful ones, a finding noted in more detail by Ryoba et al. (2021). Because successful campaigns included more words than failed campaigns on average, we found an average of 70.18 NVVNs per successful campaign and 51.46 per failed campaign. Ultimately, successful campaigns used 48.5% of the 19.2 million NVVNs in our corpus, while failed campaigns used only 42.7% of the NVVNs. Due to this disparity, the baseline for negative and positive scores should be considered via separate scales.

Examining the 1,371,258 shared-language NVVNs, 561,795 of them had a positive overall score, representing 3,984,405 total uses. The average scores of these positive shared-language NVVNs ranged from +.001 to +1.000, with Quartile 1 (the value below which 25% of the data lies, also known as Q1) at +.153, the median at +.310, and Quartile 3 (the value below which 75% of the data lies, aka Q3) at +.555, with a mean of +.402. Examining only the 466,828 negative-scoring NVVNs, representing 2,519,752 uses, revealed very nearly a mirror image. For the negative shared-language NVVNs, the minimum was -.001 and maximum -1.000, Q1 was -.136, the median -.333, Q3 was -.551, and the mean was -.403. There were also 342,635 shared-language NVVNs, representing 926,368 uses, which had a score of exactly zero. The overall mean number of uses per unique NVVN is 5.41. The mean uses per NVVN for positive-scoring NVVNs is somewhat higher (7.09), while the mean uses for negative-scoring NVVNs is 5.39. NVVNs scoring exactly zero had 2.70 uses on average.

We can also extract the verb pseudo-phrases from the NVVNs to inspect them independently. The mean score for all verb pseudo-phrases is slightly positive (+.078). Examining the 5.6 million uses of verb pseudo-phrases that ended up with a positive overall average score, the non-zero positive scores ranged from +.001 to +1.000, with Q1 at +.052, median at +.081, Q3 at +.166, and the mean at +.135. Examining the 1.67 million uses of verb pseudo-phrases that scored negatively returns scores ranging from -.001 to -1.000, with Q1 at -.016, the median at -.044, Q3 at -.120, and the mean at -.106. We found 98,665 uses of verb pseudo-phrases which scored exactly zero on the scale (consider Table 3 for a list of highly positive and highly negative verb pseudo-phrase scores.) These two sets of scores offer a set of comparative statistics to keep in mind while evaluating the scores of the NVVNs and verbs below.

This rough sketch of the distribution of positive and negative NVVNs and verb pseudo-phrases above suggests some of the insights the data may yield. Note that the NVVN distributions, in both positive and negative directions, reflected a more even distribution across a broader range of scores than was the pattern with verb pseudo-phrases, where scores were clustered in a smaller range closer to zero. Every verb pseudo-phrase emerged from an NVVN, so this finding might be puzzling at first glance. We argue that this finding shows the extent to which the surrounding noun context matters for success. While some verb pseudo-phrases are strongly associated with success or failure on their own, many others may depend on what other nouns accompany them to be strongly associated with successful or failed campaigns. The same verb pseudo-phrase may appear in a strongly-positive NVVN and a strongly-negative NVVN; these scores would tend to cancel each other out (given roughly similar scores), and drive the overall verb pseudo-phrase score closer toward zero. We argue that this dynamic reflects the value of examining verbs in the context of their surrounding nouns.

This method of identifying verb pseudo-phrases and NVVNs allows us to identify word patterns arranged around a central predicate and provide a nuanced understanding of the ways in which a verb pseudo-phrase’s association with Kickstarter’s success or failure might be a function of its context. This method permits transparency in both aggregation and disaggregation: we can build verb pseudo-phrase scores on the basis of thousands of uses, or conversely drill down from a verb pseudo-phrase to reveal nuances of use shaped by nearby nouns, verbs, and sentence boundaries. We can disaggregate a composite verb pseudo-phrase score to see which NVVNs contributed positively and negatively to the overall lean of a verb pseudo-phrase. Thus, we get a finer-grained understanding of what types of successful and unsuccessful word patterns occur together.

Limitations

This method relies on correlation. While verb pseudo-phrases are strongly or weakly associated with success or failure, correlation does not equal causation. None of the verb pseudo-phrase patterns we report in this article have scores of -1 or +1, which shows that every use of the verb pseudo-phrase existed in a failed campaign or a successful campaign (respectively). Even if we did include those, the correlation of each instance of a verb pseudo-phrase with failure does not suggest the causation that the use of any particular verb pseudo-phrase caused the failure of a campaign.

Our tools were not always ideal for the desired work, which is a limitation of many data-driven projects. We employed the probabilistic tagger Lingua::EN::Tagger, because manually tagging all of the words of the corpus would be unfeasible. This tagger produces strongly likely, but not guaranteed, outcomes. Thus, we agree with Itchuaqiyaq, Ranade, and Walton’s (2021) finding that their tools were strongly assistive but could not get every case correct. The authors manually corrected tool output; while we could not manually correct every incorrect grammar tag in a corpus of 199 million total words, we manually checked our reported findings to make sure that the examples we used in this article reflected actual verb use and the broad intent of the phrases that included the verbs.

Some noise exists in the text data. The noise occasionally appears to result from strange characters (especially letters from odd fonts used for decoration), as well as non-sentence constructions such as some bullet points and headlines. We filtered out this noise during the data-cleaning process. The tagger does not perfectly identify all terms, especially regarding vernacular English and non-English words. The tagger applies the tag for common noun if the particular grammatical function of a word cannot be determined. Some noise artifacts ended up smashing together adjacent words or eliminating sentence boundaries. These kinds of noisy artifacts have the side-effect of generating long, unique NVVNs. Since we sought common uses rather than rare ones, these noise-impacted tokens made little difference in our data routines. If future researchers pursued unusual words, these deficiencies might be worth addressing with manual correction of input text or removing with additional filters.

Note too, that the constructed verb pseudo-phrases hyphenated within the NVVN tokens are composed only of words identified by the tagger as one of several kinds of verbs (as well as the word to). As mentioned above, the verb pseudo-phrases omit any other word (such as a, the, or not) between the two N’s. This process helps enhance comparability, but caution should be exercised when reading verb pseudo-phrases in a way where negation (not) or article specification would be significant. We did not see this concern as a problem for our analysis because each use of every NVVN can be traced back directly to the full text of the specific campaign in which it appeared. We checked our reported results to ensure that we did not report a verb pseudo-phrase as projecting positively when it included (in context) a negation.

Example

We can demonstrate this method with several forms of the verb thank. A few frequent verb pseudo-phrases regarding the verb thank are thank, ca-thank (explained below), want-to-thank, and would-like-to-thank. If we consider all thank variations together, verb pseudo-phrases including thank have an overall positive valence of +.150 with 86,938 NVVN uses. The overall score of the verb thank is positive, as revealed across the scores of all the pseudo-phrases incorporating this word. Yet these four verb pseudo-phrases have varied average scores.

The verb pseudo-phrase thank (i.e., the word thank used without any other supporting verbs) is ultimately associated more often with success than failure (+.156, 74545 uses). The verb pseudo-phrase thank appeared most commonly in the NVVN _thank_you, representing the short text phrase Thank you. We would conclude that thanking people with Thank you is a common, slightly positive tactic.

The verb pseudo-phrase ca-thank returned a +.455 score (593 uses). This verb pseudo-phrase commonly represents the phrase can’t thank you enough. (The verb pseudo-phrase looks unusual because the grammar tagger split the contraction of cannot into ca and n’t, then excluded n’t from the verb pseudo-phrase because n’t is not a verb. In this case, we manually checked the verb pseudo-phrase and NVVNs to confirm what the output meant.) This very strong score means that creators say can’t thank you [enough] in successful campaigns more often than in unsuccessful ones. Despite its strong relationship with success, many fewer ca-thank tokens exist than thank tokens. Relative relationships between scores must always be remembered when dealing with disparate numbers of use cases.

The verb pseudo-phrase want-to-thank scored less positively than thank or ca-thank, coming in close to 0.0 (only +0.075). Sentences that use want-to-thank appear with almost equal frequency in successful and unsuccessful campaigns. Therefore, future creators may not be helped or harmed by using want-to-thank. Another variation, would-like-to-thank (-0.072, 1,251 uses) features a slightly negative overall score. Unsuccessful campaigns used this language only slightly more often than successful ones. The implication seems clear to creators contemplating what language to use in a new campaign: thanking patrons is often associated with success, but some ways of thanking are more effective than others.

These distinctions between types of constructions of thank indicate that trying to assess individual verbs—much less individual words—obscures diverse work that goes on within verb pseudo-phrases. Assessing individual NVVNs provides a level of nuance not currently applied to this problem.

RESULTS AND DISCUSSION

We identified four findings that reflect distinctive stylistic elements associated with success: we vs. I, informal language, invitational language, and confidence language.

We vs. I

In Kickstarter, using we in campaign language proves to be more effective than using I. We discovered this finding in several ways.

First, we discovered that verb pseudo-phrases starting with am are often somewhat-to-highly negatively associated with success: am (-.144, 46804 uses), am-asking (-.125, 2812 uses), am-raising (-.469, 2419 uses), am-working (-.192, 1557 uses), and am-looking (-.378, 983 uses), among others. We found that NVVNs with an am in the leading verb position are primarily associated with the noun I as the leading N in our corpus, suggesting that the two-word phrase I am is more likely to be present in unsuccessful campaigns than successful ones.

Looking specifically at am reveals a tendency for authors to invoke personal narrative using I am. NVVNs such as i_am_mother (-.536 average score, with 177 uses), i_am_father (-0.375, 112), and i_am_veteran (-0.690, 84) point toward strong negative relationships with success. The professional narrative also does not bode well: terms such as photographer (i_am_photographer, -0.330, 312), doctor (i_am_doctor, -0.320, 25), and executive (i_am_executive,-0.296, 27) are also negatively associated with success. Given that pitches often begin with explanations of who the entrepreneur is, this finding suggests that readers may not expect or appreciate certain types of personal and professional narratives.

Building on this preference against I, we found a direct preference for we. Consider NVVNs related to the verb assure. I_assure_you scored fairly negatively: -.217 (331 uses). Yet the score of we_assure_you almost exactly inverted i_assure_you: +.214 (121 uses). we_assure_you is more likely to occur in successful campaigns, even though creators mention we_assure_you less often than i_assure_you. This trend holds for many other verb patterns:

  • we_know_you (+.220, 1916 uses) vs i_know_you (-.029, 1515 uses);
  • we_need_help (+.143, 4099 uses) vs i_need_help (-.132, 2519 uses);
  • we_believe_it (+.092, 1167 uses) vs i_believe_it (-.156, 1361 uses);
  • we_ask_your (+.199, 491 uses) vs i_ask_your (-.097, 268 uses);
  • we_can-raise_money (+.262, 282 uses) vs i_can-raise_money (-.239, 146 uses);
  • we_have-created_ (+.258, 147 uses) vs i_have-created_ (-.047, 170 uses);
  • we_have_music (+.227, 158 uses) vs i_have_music (-.179, 106 uses).

These patterns illustrate a trend that we found in NVVNs associated with many different verbs. Phrases using we as the primary noun in an NVVN were more prominently associated with successful campaigns, while the same phrases using I more commonly appeared in unsuccessful campaigns. We tested this finding an additional way by applying our scoring system to the leading nouns of all NVVNs and found that we and I were the two most common leading nouns. The pronoun we scored +.155 with 1,102,236 total uses, while I scored -.064 with 889,060 total uses. We was much more likely to be successful than I, suggesting that Kickstarter readers have a preference for campaigns that write from a plural perspective.

We found a few areas where solo language was associated with success. The verb pseudo-phrases am-recording and ‘m-recording scored +.557 with 131 uses and +.670 with 100 uses, respectively; the verbs am-releasing and ‘m-releasing fared nearly as well, with scores of +.404 with 57 uses, and +.580 with 50 uses, respectively. Recording and releasing suggest artistic ventures (recording an album, releasing a comic book); as we mention below, solo artists and musicians create credibility and trust in the reader in other ways.

Ultimately, readers seem to prefer we language to I language, with some exceptions. We argue that the context of the speculative venture may favorably influence readers’ opinions toward teams running campaigns (as instantiated in we) instead of individuals running campaigns (as instantiated in I). Given Kickstarter’s mandate to backers to assess whether a campaign can “bring a project to life,” teams of people may seem more trustable (even if only through multiple people believing in and working on the idea) to complete a project than an individual (Kickstarter, PBC, 2022a).

Furthermore, using we is a successful convention, as our large numbers of cases show: writing from a stance of we accords with, and may invoke, previous or similar campaigns that have succeeded to the reader. Meeting the audience’s stylistic expectations shows a level of fluency with the platform that reflects well on not only being able to “bring a project to life” but also perhaps also “deliver the rewards they’ve promised to backers” (Kickstarter, PBC, 2022a). Speaking in conventional ways (as shown below) may inspire trust in readers.

Informal Language

Informal language played a complex role in our corpus. Contractions were often associated with successful campaigns, while phrases associated with the term bless you were strongly associated with unsuccessful campaigns. These two findings suggest that Kickstarter readers have complex expectations for the conventions of informality.

Contractions offered a specific case. In many verbs, the contraction form of a verb strongly associated with success, while the full version associated with failure. Verb pseudo-phrases containing contractions are routinely associated strongly with success. For example, the mean success score for the top 500 most frequent verb pseudo-phrases was +.091 (Min: -.469, 1stQ: +.008, Med: +.072, 3rdQ: +.158, Max: +.633). Twenty-six of those 500 verb pseudo-phrases included a contraction of some kind (e.g., ‘re, ‘m, ‘ve-been; our grammar tagger split contractions, so our output contains partial contractions functioning as their own verb). Those common contraction verb pseudo-phrases showed a mean success score of +.275. The highest score was +.442 for ‘ll-get (5754 uses), with ‘ll-send (+.431, 2650 uses) and ‘re-asking (+.401, 1866 uses) close behind. The worst-scoring popular verb pseudo-phrases involving a contraction were ‘m-asking (+.062, 1880 uses) and ‘m (+.064, 29077 uses), but these two were still associated slightly more with success rather than failure. Moreover, of the 26 common verb pseudo-phrases including a contraction, these were the only two which scored below the median for the 500 most common verbs.

Examining a contraction with its full equivalent highlights the success skew in favor of contractions. Examples of verb pseudo-phrases involving am or ‘m are a useful way to highlight this because they would not show the effects of the preference for we vs. I explored in the previous section. Consider the base verb: am has a somewhat negative average success score of -.144 (46804 uses), while its contraction form, ‘m, has a mildly positive association with success: +.064 with 29077 uses. More complicated phrases show the same pattern, where the full form has a much more negative average success score. For example, ‘m-asking (+.062, 1880 uses) associates positively with success, while am-asking associates negatively with success (-.125, 2812 uses). Other examples include:

  • ‘m-making (+.277, 593 uses) vs. am-making (-.033, 769 uses),
  • ‘m-excited (+.228, 224 uses) vs. am-excited (-.030, 197 uses).
  • ‘m-doing (+.091, 1034 uses) vs. am-doing (-.190, 963 uses), and
  • ‘m-creating (-.013, 223 uses) vs. am-creating (-.189, 603 uses), and
  • ‘m-hoping (exactly even, .000, with 807 uses) vs am-hoping (-.246, 678 uses).

The pattern of the contraction scoring recognizably higher seems to have been operative in nearly all cases. This finding held true whether the phrases themselves were relatively positive, such as ‘m-offering (+.352, 699 uses) vs am-offering (+.137, 1079 uses), or both relatively negative, such as ‘m-raising (-.307, 994 uses) vs am-raising (-.469, 2419 uses) or ‘m-seeking (-.196, 204 uses) vs am-seeking (-.382, 912 uses).

The positive association with contractions seems to hold true in the case of other words. For example, consider ‘ll-add (+.488, 1121 uses) and will-add (+.249, 3097); ‘ll-ask (+.511, 315 uses) and will-ask (+.216, 945 uses); and ‘ll-be (+.336, 6719 uses) and its full equivalent will-be (+.042, 82448 uses). The number of uses of these three verb pairs is somewhat-to-greatly unbalanced in favor of the full version of the verb; these numbers contrast with the numbers for am vs ‘m, which often had similar or even exactly the same number of uses for both forms. With due caution in interpretation for more unbalanced pairs, the pattern established with ‘m and am seems to hold for other contractions as well.

We also found that the we vs. I finding interacts with the contractions finding to produce strong effects regarding the verbs ‘m, am, ‘re, and are. This interaction can be seen in verb pseudo-phrases, including the verb asking: ‘re-asking (+.401, 1866 uses), are-asking (+.113, 4934), ‘m-asking (+.062, 1880), and am-asking (-.125, 2812). We plus a contraction resulted in better scores than I plus a full verb. Yet the strength of the we-plus-contraction finding is reflected by the we-plus-full-verb version (we are asking), also scoring higher than the I-plus-contraction version (I’m asking). Verbs surrounding seeking perform similarly: ‘re-seeking (+.240, 154 uses); are-seeking (-.076, 1690), ‘m-seeking (-.196, 204), and am-seeking (-.382, 912). In many examples concerning these four forms of am-plus-verb-type, the we-plus-contraction language scores the best. This pattern includes verbs that are:

  • Very positive overall: ‘re-offering (+.517, 1632 uses), are-offering (+.278, 3976 uses), ‘m-offering (+.352, 699 uses), and am-offering (+.137, 1079 uses)
  • Mostly negative overall: ‘re-looking (+.251, 1273 uses) vs are-looking (-.107, 3397 uses) and ‘m-looking (-.109, 676 uses) vs am-looking (-.378, 983 uses)
  • Varied: ‘re-working (+.373, 831 uses), are-working (+.028, 3488 uses), ‘m-working (+.062, 930 uses), am-working (-.192, 1557 uses)

Using we and contractions together often indicates strong outcomes; using I and full forms of verbs together often indicates poor outcomes.

Yet we also found some informal language strongly associated with unsuccessful campaigns. One prominent category of verb pseudo-phrases strongly associated with unsuccessful campaigns can be perceived as interpersonal statements instead of professional statements. For example, the verb bless has a distinctly negative score (-.404, 1310 uses), as does blessed (-.349, 86 uses). Unsurprisingly, NVVNs using this term feature similarly negative scores: god_bless_ (-.470, 586 uses; the trailing underscore shows that this verb pseudo-phrase appeared at the end of a sentence) and god_bless_you (-.460, 419 uses) exist much more commonly in unsuccessful campaigns than successful ones. Bless you is a phrase spoken often in interpersonal contexts. In the situation of being spoken after a person sneezes, Arnovick (2000) argued that “some speakers intend Bless you as a religious blessing. They mean literally what they say. Some remember the superstition and its call for a quick ‘Bless you’ or ‘Gesundheit.’ Some speakers recognize both forms as a vague wish that the sneezer stay well or regain good health. Still, others are motivated solely by politeness” (p. 122, italics original). While the asynchronous, digital mode of Kickstarter campaigns differs from the synchronous, co-located experience of a sneeze, the uses of bless you and its variants seem similar. Whether a Kickstarter creator offers a literal blessing or speaks only from politeness, neither of these motivations point toward professional activity. Bless you may be a level of interpersonal familiarity too informal for readers to support. Further research could consider other statements of politeness or idioms to determine whether bless you is a specific case or indicative of a wider rejection of interpersonal communication styles.

Ultimately, the conventions we discovered surrounding informal language speak to how well creators understand the platform’s conventions. The level of informality suggested by contractions is generally accepted positively by readers of the platform. The level of informality that bless you brings to the project is not generally accepted positively by readers of the platform. We argue that understanding the appropriate level of formality on the platform shows that the creator understands the conventions of the platform. A creator’s understanding of the platform’s conventions may encourage readers to trust that the creators’ (perceived) familiarity with the platform can translate into an effectively completed campaign.

Appropriate levels of informality are not the only way that writers build the reader’s trust in Kickstarter campaigns, though. Writers also build reader trust via invitational language.

Invitational Language

Creators successfully used language inviting the reader into the campaign. We found successful correlations surrounding verbs that invite the reader to join the campaign, make the reader feel helpful to the campaign, and remind the readers of what items and experiences they will get from the campaign.

The positive scores of the verb pseudo-phrase ‘ll-join (+.435, 526 uses, including NVVN phrases such as you_’ll-join_me [+.396, 116 uses] and you_’ll-join_us [+.454, 337 uses]) and please-join (+.136, 1417 uses, including _please-join_us [+.171, 770 uses], _please-join_our [+.343, 131 uses], and _please-join_me [-.021, 326 uses]) suggest that asking the reader to join the project is a good move. Please join us reflects a mildly positive score, while Please join our shows a more strongly positive score. However, please join me returns an almost neutral score, showing that the positive associations with join may be outweighed by the negative aspects of being a solo campaign runner. While please join is a single example, the finding that us and our score more positively than me in this construction suggests that the we vs. I finding from above may hold for various pronoun types, such as object pronouns and possessive pronouns. Further research on types of pronouns and their effectiveness would be warranted.

Inviting the readers to participate in marketing the campaign is often associated with success: us_get_word (as in, help us get the word out) is effective, scoring +.161 with 482 uses. Other verb pseudo-phrases that suggest user involvement include ca-do (+.215, 2736 uses, as in can’t do this without you) and will-help (+.105, 18831 uses, as in you will help us and your support will help us). These findings point toward an emotional reward for being part of the project: the feeling of being helpful or necessary.

While writers should invite the reader to attain the emotional reward of being part of the project, Kickstarter requires the goal of each crowdfunding campaign to be something created: “every project needs a plan for creating something and sharing it with the world” (Kickstarter, PBC, 2022b). Thus, suggesting things the user will get represents another way of inviting people into the campaign. The verb ‘ll-get is strongly positively associated with success (+.442, 6367 uses). Examining the trailing nouns of NVVNs beginning with you_’ll-get_(x) shows a clear and strongly positive inclination toward concrete rewards: copy, access, everything, rewards, it, kinds, chance, pdf, download, book, set, rewards, something, and game, among others. The only trailing noun with a weak positive score is chance, from the NVVN you_’ll-get_chance (+.026, 75 uses), which suggests backers may avoid uncertainty. We discuss uncertainty language more below.

Another set of strongly positive words relates to getting rewards: NVVNs you_get_your (+.190, 1896 uses), you_get_ (+.185, 1483 uses), _get_your (+.237, 1271 uses), you_get_you (+.122, 939 uses), and you_can-get_your (+.370, 853 uses), as well as the verb pseudo-phrase will-receive (+.286, 21750 uses) are all positive correlations. The verb pseudo-phrase ‘ll-send (+.431, 2650 uses) includes we_’ll-send_you (+.394, 1283) and i_’ll-send_you (+.434, 469). These verb pseudo-phrases speak confidently about the rewards that the supporters of the campaign will receive. This focus on rewards associates so strongly with success that the finding represents a rare case where an I-focused term (I’ll send you) is more positively associated with success than a we-focused version of the same term (we’ll send you), albeit with far fewer uses. We would argue, then, that showing facility with the conventions of running a campaign (as instantiated in knowing how to fulfill rewards) may help readers trust that creators can “deliver the rewards they’ve promised to backers” (Kickstarter, PBC, 2022a). In other words, explaining the delivery process for rewards may be a way for solo writers to distinguish themselves as unusually trustable solo creators.

The verb will-contact (+.322, 1253 uses) also refers to the rewards process. Campaign creators must contact the backers after the conclusion of the successful campaign to acquire details about the rewards fulfillment (such as mailing address, reward options, and other technical concerns). The most commonly used NVVNs here are we_will-contact_you (+.370, 789 uses) and i_will-contact_you (+.445, 184 uses). These NVVNs, as well as NVVNs that describe the contact process in further detail (we_’ll-send_survey, +.578, 171 uses; i_’ll-send_survey, +.666, 48 uses), hold the same pattern from We/I’ll send you of being strongly positive for the we-focused term and even more strongly positive for the less-used I-focused term. This language may be successful because less successful campaigns omit specific references to rewards. Ultimately, being detailed and positive about rewards is associated strongly with success, especially for individual creators.

In a speculative context where success is not guaranteed, inviting readers to join the campaign (and take on a small measure of the speculative risk through funding) may seem an unusual way to garner support for a campaign. Yet the ways the invitations function in the campaign suggest a familiarity with the experience of Kickstarter campaigns: Kickstarter campaigns have both emotional and practical benefits for the readers. The creator who reflects awareness of both benefits in prose may demonstrate a familiarity with the platform’s conventions that builds trust.

Confidence Language

Several verbs looking forward to the success of the campaign score strongly positively: hope (+.122, 29358 uses) and hope-to-see (+.193, 1099 uses). The verb pseudo-phrase ca-wait-to (as in can’t wait to, +.520, 1042 uses) is also a surprisingly positive way of confidently looking forward.

Looking forward is good, but looking forward confidently is better. The verb reach is often used to describe future and already achieved fundraising goals and strongly associated with success (+.296, 21557 uses). NVVNs such as we_reach_goal (+.380, 1319 uses), we_reach_stretch (+.498, 882 uses), and we_reach_level (+.605, 233 uses) reveal what will happen at specific intervals: the word goal is often used to describe the financial goals of the campaign, while stretch and level often describe reaching a goal beyond the original financial goal (stretch goals or extra levels of funding to unlock new rewards). Other campaigns offer more detail, telling readers about what can be expected when a specific number of dollars is raised or number of people contributing is reached: we_reach_10,000 (+.369, 195 uses) and we_reach_100 (+.404, 99 uses). The value of specific goals can be seen when aggregating the average score for we_reach NVVNs by whether a number or a letter opens the second noun position. Both scores are positive, but we_reach_NUMBER NVVNs score more positively (+.509, 3468 uses) than we_reach_WORD (+.302, 10476 uses) NVVNs. Looking forward confidently with detailed expectations of what will occur at certain checkpoints is a strong way forward.

Some writers inject unnecessary uncertainty by overly qualifying their verbs. Consider the case of the modal verb would, which can be used to describe the future with an implication of greater uncertainty than a modal verb like will. Overall, all verb pseudo-phrases beginning with would collectively scored slightly negatively (-.041, 83125 uses) compared to all verb pseudo-phrases beginning with will, which collectively scored positively (+.065, 502705 uses). Comparing some of the most common forms of this verb shows an association between projecting certainty by using the word will and project success. For example, will-be (+.042, 80379 uses) scores a bit higher than would-be (-.001, 14177 uses). The pattern holds true with the verb pseudo-phrase will (+.050, 33300 uses) vs. pseudo-phrase would (-.063, 9573 uses) and will-help (+.105, 18831 uses) vs would-help (-.034, 1278 uses). The tilt toward a more positive score for will over would remains, even if both verbs’ pseudo-phrases score negatively: will-have scores slightly negatively (-.044, 22143 uses) but would-have scores worse (-.110, 1496 uses). The difference remains when the score of the would form is associated with success, as in would-receive (+.200, 80 uses) vs will-receive (+.286, 21750) and would-get (+.005, 372 uses) vs will-get (+.230, 12695 uses). Creators use the word would in conjunction with other words that suggest uncertainty, such as like. These constructions score negatively on average: All verb pseudo-phrases beginning with would-like have an average score of -.040 across 19781 uses.

Ultimately, writers should project confidence and avoid uncertainty. Using a modal verb that suggests certainty (will) demonstrates that the creator is confident in claims, which may translate to overall confidence in their own ability to complete the campaign effectively. Using a modal verb reflecting uncertainty (would) demonstrates that the creator is not confident in claims, and is perhaps less confident that the campaign will succeed. We argue that readers can better trust a creator who is confident in their own claims (and, by extension, abilities) rather than one who is not.

CONCLUSION

Style guides allow technical communicators to write text deemed acceptable by an organization. In this paper, we identify types of textual patterns that users of Kickstarter find acceptable, as evidenced by the textual patterns’ inclusions in successful campaigns. These patterns represent elements of an emergent style guide for Kickstarter that can help entrepreneurs communicate effectively on the platform. While our findings on their own offer guidelines, together they form evidence of a set of reader expectations for Kickstarter campaigns, drawn from the decisions of groups of readers to fund or not fund campaigns that include certain types of arguments and phrases (as determined by the prominent NVVN word patterns we identified). See Table 5 for a list of findings with guidance for entrepreneurs, technical communicators, and entrepreneurial technical communicators.

Beyond the individual findings in Table 5, we argue that these findings interact with each other in distinctive ways. Texts are not mere collections of words; words reflect attempts by writers to convey meaning through a text. Style contributes to that making of meaning. The goal of a style guide is not simply to ensure consistency across texts but also to ensure the acceptability of texts (Adhya, 2015, p. 184). Each element of style contributes to the acceptability of a document as perceived by the organization that made the style guide; in this instance, we have developed an emergent style guide that reflects what users across the platform broadly see as acceptable (as evidenced by what elements of style more often appear in successful campaigns than unsuccessful campaigns). Yet even with acceptability (as instantiated in a successful campaign) as a goal, the writer should not dogmatically reproduce the findings of the style guide. Instead, writers should understand how these findings work together to produce acceptable writing. Thus, we argue that some findings directly interact with and mediate each other, while all of the findings work together toward a specific goal: getting the reader to trust that the creator can bring the project to completion.

First, some findings work together directly. We found that we vs. I and contractions often work together in successful campaigns: phrases such as we’re offering and we’re looking are much more commonly associated with success than we-plus-full-verb constructions, I-plus-contraction constructions, and I-plus-full-verb constructions (the last of which almost always scored the worst in the set). Two effective stylistic practices working in tandem often point toward an even more successful stylistic element.

Team language and invitational language also intersect in distinctive ways. While we found an overall preference for we-focused language across many campaigns, we also found that invitational language intersected with I-focused language to produce an unusual result. One form of invitational language concerned telling readers about the campaign process as a way of inviting them into the process. I-focused language that indicated the solo creator’s preparation to fulfill rewards (such as I’ll send you or I’ll contact you) represented the rare place where I-focused language scored better than the we-focused version of the same verb pseudo-phrase. We argue that the preference for we-language is due to readers trusting teams to complete a project more often than trusting individuals. This level of trust may reflect the complexity of running a successful campaign and completing the project that campaign funds. Creators must create and deliver the product or experience that the Kickstarter campaign represents, then create and deliver the rewards promised to the backer (if the rewards for funding the project differ from the product or experience being created, which is often the case). Subject matter expertise and fulfillment expertise are both necessary. Some types of projects (such as technology or theater projects) may require more expertise than a single person is perceived to have; teams may have more subject matter expertise and would be potentially more trustworthy. Teams of people may also be more trusted to complete the fulfillment of rewards, as people can be perceived to take on specialized roles within the team or at least divide the work of fulfillment over more than one person.

Conversely, I-focused invitational language surrounding the process of rewards may be a convincing way for the individual creator to show their competence and understanding of how to complete a campaign. Demonstrating that the single creator has considered the process of fulfillment may differentiate the solo creator from creators that have not written about fulfillment. Solo creators who do not mention fulfillment may be perceived to have not thought about fulfillment deeply. The audience may also perceive these solo creators to be less prepared to fulfill rewards than solo creators who did mention fulfillment. In these ways, we found that stylistic elements interacted with each other to produce effects that would not be expected via a single stylistic element alone.

These two intersections of stylistic elements—we vs. I and informal language—point to a concern with professionality, and we argue that many of the elements in this emergent style guide point toward a need in Kickstarter writing: development of the reader’s trust. The preference for team language instead of solo language suggests that teams may be perceived to be more competent or capable than solo creators. Kickstarter readers seem to expect a certain level of informality as a marker of confidence (and therefore competence), while rejecting too much informality (such as bless you) as a sign of a too much interpersonal language and a lack of professionality (and therefore potential lack of competence). Invitational language regarding rewards also suggests that writers understand the conventions of the platform: writers who understand that the reader may desire both a physical reward and an emotional benefit of being part of something bigger than themselves can demonstrate this knowledge and potentially be perceived as competent. Confidence language attempts to directly display competence through demonstration of assurance that the project will succeed. Each of these types of language makes an effort to show the reader that the reader can trust the creator to “bring the project to life” (Kickstarter, PBC, 2022a). With instances of each of these types of language being individually positively associated with success, we would expect that the interaction effect of all of these working together would be positive as well. But we again argue that the individual words of the campaign do not represent the reason campaigns succeed. Instead, each of these findings help signal a critical aspect that readers appear to look for in Kickstarter campaigns: the ability to successfully complete a campaign.

In addition to these new findings, we offer validation to previous findings with a larger data set. Previous studies have noted that we language associates with success (Grebelsky-Lichtman & Avnimelech, 2018). We found that, for the same verb, the verb pseudo-phrase beginning with we almost always associated more positively with success than the verb pseudo-phrase beginning with I. This finding may reflect a trust issue: a group of professionals saying we assure you may look more trustworthy than a single person saying I assure you. We would appreciate support may look like a professional organization giving a pitch, while I would appreciate support may look unprofessional and rely too much on pathos. Given the low bar to creating a Kickstarter campaign, several of the cues that people look to when determining to support a campaign or not appear to be markers of prior competence or professionalization. Thus, we would add “signals capacity to complete tasks of the campaign” to the list of things that Grebelsky-Lichtman and Avnimelech (2018) noted that we can do in Kickstarter campaigns: “Using plural pronouns such as us and we demonstrates immediacy in communication, brings people together, and highlights commonalities” (p. 4182, italics original).

Similarly, our study does not negate the idea that I-focused language may be associated with narcissism (Butticè & Rovelli, 2020), but we would suggest that I-focused language may be associated with a lack of capacity or ability to complete a campaign. Our study and that of Butticè and Rovelli (2020) found an exception in the use of I for comics and music. Butticè and Rovelli (2020) argued that “in industries where the product value is foremost associated to the product functionality (e.g., the content of a comic book), backers pay less attention to the entrepreneur and the penalization for narcissism is reduced or even disappears” (p. 5). While this idea may be true, we would argue that readers have no reason to believe that an individual would be less successful than a team in subject matter expertise needed to complete a music or comics project, because solo music and comic campaigns are common. However, fulfillment expertise would still be needed for a solo music or comics creator, which suggests why I plus fulfillment language is very strongly associated with success. A display of fulfillment expertise in a context where readers more readily accept an individual’s subject matter expertise creates a strong indicator that a creator can bring a project to completion.

Our study validates several findings from Mitra and Gilbert (2014). The authors argue that terms reflecting authority, such as project will be and we can afford, are positively associated with success (p. 58). We agree that words of this type are positively associated with success. We would argue specifically that project will be and we can afford demonstrate confidence language. Similarly, Mitra and Gilbert (2014) identified forward-looking phrases (“next step is, in the upcoming, will be published, to announce”) as particularly helpful (p. 59, italics original); our findings agree that looking forward using confident verbs like hope and can’t wait to is associated with success. Confidence matters because confidence demonstrates the ability to complete a project in the eyes of Kickstarter readers. Finally, our finding that invitational language is strongly associated with success connects with Mitra and Gilbert’s (2014) finding that terms concerning reciprocity are positively associated with success (e.g., mention your, also receive two, pledged will) (p. 57).

Grebelsky-Lichtman and Avnimelech (2018) and Mitra and Gilbert (2014) indicated that creators should avoid language of uncertainty: “phrases which exude negativism (not been able), or lack assurance (later i, hope to get) are predictors of not funded” (Mitra & Gilbert, 2014, p. 57, italics and capitalization original). Anglin et al. (2018) also suggest using confident language. Our findings regarding confidence language validate each of these findings with a large data set.

Finally, we extend the finding of Koch and Siering (2015) that details (which they operationalized as increasing word amounts) are valuable in Kickstarter campaigns. Instead of word amount, we found specific words that were useful for describing details. We found that explaining what specific things the reader will get from the campaign upon its success is a strongly positive move that demonstrates confidence and the creator’s awareness of fulfillment. This move may build trust in the reader that the creator has paid attention to details and can thus successfully complete the project.

Other prior findings about text in Kickstarter studies are yet to be validated, such as those of Mukherjee et al. (2017) regarding innovation language and Ishizaki (2016) about use of you and quotation marks. These could be researched in further studies. Expanding this work beyond crowdfunding, much less beyond Kickstarter, poses a distinct set of challenges. Kickstarter specifically and crowdfunding, in general, offer a somewhat unusual situation among writing studies in that the success or failure of the piece of writing is public. Yet each crowdfunding platform contains its own conventions (Dushnitsky & Fitza, 2018). Discovering emergent style guides for other online genres would require tools to be built that reflect the success and failure conditions of varied online genres. For social media posts from a specific organization, one might be able to create a threshold of views, likes, shares, and/or comments that approximates success.

Ultimately, readers of Kickstarter campaigns have expectations of what Kickstarter campaigns should say in their text that have become conventions in successful campaigns. We suggest that these conventions, as discovered via verb pseudo-phrases associated with successful campaigns, are derived from the reader’s goal of finding a campaign with a creator or creators whom the reader trusts to “bring the campaign to life” via delivering the physical and emotional rewards promised (Kickstarter, PBC, 2022a). These findings comprise elements of an emerging Kickstarter style guide (as displayed in Table 5), offering suggestions to the technical communicator who is involved in writing a Kickstarter campaign text. Understanding Kickstarter style expectations can help entrepreneurial technical communicators respond with acceptable style for an audience of readers who expect certain stylistic moves that signal the ability to complete a campaign successfully.

REFERENCES

Aarts, B. (2011). Oxford modern English grammar. Oxford University Press.

Adhya, E. (2015). Key elements of an effective style guide in the new age. Technical Communication, 62(3), 183–192.

Anglin, A. H., Short, J. C., Drover, W., Stevenson, R. M., McKenny, A. F., & Allison, T. H. (2018). The power of positivity? The influence of positive psychological capital language on crowdfunding performance. Journal of Business Venturing, 33(4), 470–492. https://doi.org/10.1016/j.jbusvent.2018.03.003

Arnovick, L. K. (2000). Diachronic pragmatics: Seven case studies in English illocutionary development. John Benjamins Publishing Company.

Belleflamme, P., Lambert, T., & Schwienbacher, A. (2014). Crowdfunding: Tapping the right crowd. Journal of Business Venturing, 29(5), 585–609. https://doi.org/10.1016/j.jbusvent.2013.07.003

Bowdon, M. A. (2014). Tweeting an ethos: Emergency messaging, social media, and teaching technical communication. Technical Communication Quarterly, 23(1), 35–54. https://doi.org/10.1080/10572252.2014.850853

Butticè, V., & Rovelli, P. (2020). “Fund me, I am fabulous!” Do narcissistic entrepreneurs succeed or fail in crowdfunding? Personality and Individual Differences, 162, 110037, https://doi.org/10.1016/j.paid.2020.110037.

Carradini, S. (2022). Data sets and tools. https://stephencarradini.com/data-sets/

Carradini, S., & Fleischmann, C. (2022). The effects of multimodal elements on success in Kickstarter crowdfunding campaigns. Journal of Business and Technical Communication. https://doi.org/10.1177/10506519221121699

Coburn, A. (2019). Lingua::EN::Tagger: Part-of-speech tagger for English natural language processing [Computer software]. CPAN: https://metacpan.org/release/ACOBURN/Lingua-EN-Tagger-0.31.

Cudmore, A., & Slattery, D. M. (2019). An analysis of physical and rhetorical characteristics of videos used to promote technology projects, on the Kickstarter crowdfunding platform. Technical Communication, 66(4), 319–346.

Cumming, D. J., Leboeuf, G., & Schwienbacher, A. (2015). Crowdfunding models: Keep-it-all vs. all-or-nothing. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2447567

Davidson, R., & Poor, N. (2015). The barriers facing artists’ use of crowdfunding platforms: Personality, emotional labor, and going to the well one too many times. New Media & Society, 17(2), 289–307. https://doi.org/10.1177/1461444814558916

Dushnitsky, G., & Fitza, M. A. (2018). Are we missing the platforms for the crowd? Comparing investment drivers across multiple crowdfunding platforms. Journal of Business Venturing Insights, 10. https://doi.org/10.1016/j.jbvi.2018.e00100

Fan, W. & Gordon, M. D. (2014). The power of social media analytics. Communications of the ACM, 57(6) , 74–81. https://cacm.acm.org/magazines/2014/6/175163-the-power-of-social-media-analytics

Fraiberg, S. (2021). Introduction to special issue on innovation and entrepreneurship communication in the context of globalization. Journal of Business and Technical Communication, 35(2), 175–184. https://doi.org/10.1177/1050651920979947

Gallagher, J. R., & Beveridge, A. (2022). Project-oriented web scraping in technical communication research. Journal of Business and Technical Communication, 36(2), 231–250. https://doi.org/10.1177/10506519211064619

Gallagher, J. R., Chen, Y., Wagner, K., Wang, X., Zeng, J., & Kong, A. L. (2020). Peering into the Internet abyss: Using big data audience analysis to understand online comments. Technical Communication Quarterly, 29(2), 155–173. https://doi.org/10.1080/10572252.2019.1634766

Getto, G., & Labriola, J. T. (2019). “Hey, such-and-such on the Internet has suggested…”: How to create content models that invite user participation. IEEE Transactions on Professional Communication, 62(4), 385–397. https://doi.org/10.1109/TPC.2019.2946996

Google. (2022). Live Caption Requirements. YouTube Help. https://web.archive.org/web/20220601010846/https://support.google.com/youtube/answer/3068031?hl=en

Grebelsky-Lichtman, T., & Avnimelech, G. (2018). Immediacy communication and success in crowdfunding campaigns: A multimodal communication approach. International Journal of Communication, 12(27). https://ijoc.org/index.php/ijoc/article/view/7691

Hu, W., & Yang, R. (2020). Predicting the success of Kickstarter projects in the US at launch time. In Intelligent Systems and Applications (pp. 497–506). https://doi.org/10.1007/978-3-030-29516-5_39

Ishizaki, S. (2016). Computer-aided rhetorical analysis of crowdfunding pitches. In 2016 IEEE International Professional Communication Conference (IPCC) (pp. 1–4). IEEE. https://doi.org/10.1109/IPCC.2016.7740540

Itchuaqiyaq, C. U., Ranade, N., & Walton, R. (2021). Theory-to-query: Developing a corpus-analysis method using computer programming and human analysis. Technical Communication, 68(3), 7–28.

Jones, N. N. (2017). Rhetorical narratives of black entrepreneurs: The business of race, agency, and cultural empowerment. Journal of Business and Technical Communication, 31(3), 319–349. https://doi.org/10.1177/1050651917695540

Kickstarter, PBC (2021). Bring your creative project to life. Kickstarter. https://web.archive.org/web/20210505075032/https://www.kickstarter.com/learn

Kickstarter, PBC (2022a). Trust and accountability. Kickstarter. https://web.archive.org/web/20220104052031/https://www.kickstarter.com/trust

Kickstarter, PBC (2022b). Our rules. Kickstarter. https://web.archive.org/web/20220510195233/https://www.kickstarter.com/rules

King, C. (2009). Why 140 characters. ChrisKing.Info. https://www.chrisking.info/140-characters

Koch, J-A., & Siering, M. (2015). Crowdfunding success factors: The characteristics of successfully funded projects on crowdfunding platforms. Proceedings of the 23rd European Conference on Information Systems (ECIS 2015), Muenster, Germany, https://ssrn.com/abstract=2808424

Lauren, B., & Pigg, S. (2016). Networking in a field of introverts: The egonets, networking practices, and networking technologies of technical communication entrepreneurs. IEEE Transactions on Professional Communication, 59(4), 342–362. https://doi.org/http://doi.org/10.1109/Tpc.2016.2614744

Liederman, E., & Perloff, C. (2022). “Silence, Brand”: How to avoid corporate cringe on social media. AdWeek. https://www.adweek.com/media/silence-brand-how-to-avoid-corporate-cringe-on-social-media/

Lins, E., Fietkiewicz, K. J., & Lutz, E. (2018). Effects of impression management tactics on crowdfunding success. International Journal of Entrepreneurial Venturing, 10(5), 534–557. http://dx.doi.org/10.1504/IJEV.2017.10006928

Mara, A. (2008). Ethos as market maker: The creative role of technical marketing communication in an aviation start-up. Journal of Business and Technical Communication, 22(4), 429–453. https://doi.org/10.1177/1050651908320379

Marcus, M. P., Santorini, B., and Marcinkiewicz, M. A. (1993). Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2):313–330.

Martin, S., Greiling, D., & Wetzelhütter, D. (2018). Expectations of Facebook users towards a virtual dialogue with their public utility. International Journal of Energy Sector Management, 12(3), 408–425. https://doi.org/10.1108/IJESM-11-2017-0013

Meta. (2022). [Query landing page for search term “Characters”]. Instagram Help Center. https://web.archive.org/web/20220707172559/https://help.instagram.com/search/?helpref=search&query=characters

Mitra, T., & Gilbert, E. (2014). The language that gets people to give: Phrases that predict success on Kickstarter. CSCW 2014 Crowdfunding: “Show Me the Money!” Baltimore, Maryland. https://doi.org/10.1145/2531602.2531656

Molina, B. (April 30, 2019). Chase tried to motivate customers with lower bank balances. It backfired badly. USA Today. https://web.archive.org/web/20211113051224/https://www.usatoday.com/story/money/business/2019/04/30/chase-bank-monday-motivation-tweet-advising-customers-how-save-money-backfires/3624733002/

Mollick, E. (2014). The dynamics of crowdfunding: An exploratory study. Journal of Business Venturing, 29(1), 1–16. https://doi.org/10.1016/j.jbusvent.2013.06.005

Mukherjee, A., Yang, C. L., Xiao, P., & Chattopadhyay, A. (2017). Does the crowd support innovation? Innovation claims and success on Kickstarter. HEC Paris Research Paper No. MKG-2017-1220. http://dx.doi.org/10.2139/ssrn.3003283

Pope, A. R. (2018). Understanding the writing demands of crowdfunding campaigns with the genre-mapping report. Business and Professional Communication Quarterly, 81(4), 485–505. https://doi.org/10.1177/2329490618795935

Rosen, A. (2017). Tweeting made easier. Twitter Blog. https://blog.twitter.com/official/en_us/topics/product/2017/tweetingmadeeasier.html

Ryoba, M. J., Qu, S., & Zhou, Y. (2021). Feature subset selection for predicting the success of crowdfunding project campaigns. Electronic Markets, 31, 671–684 https://doi.org/10.1007/s12525-020-00398-4

Sandouka, K. (2019). Text analysis of crowdfunding: A literature review. AMCIS 2019, Cancún. https://aisel.aisnet.org/amcis2019/social_computing/social_computing/6/

Sano-Franchini, J. (2018). Designing outrage, programming discord: A critical interface analysis of Facebook as a campaign technology. Technical Communication, 65(4), 387–410.

Shuttleworth Foundation. (2014). quickscrape. In github.com/ContentMine/quickscrape

Spartz, J. M., & Weber, R. P. (2015). Writing entrepreneurs: A survey of attitudes, habits, skills, and genres. Journal of Business and Technical Communication, 29(4), 428–455. https://doi.org/10.1177%2F1050651915588145

Spinuzzi, C. (2016). Introduction to the special issue on entrepreneurship communication. IEEE Transactions on Professional Communication, 59(4), 316–322. https://doi.org/10.1109/TPC.2016.2607803

Spinuzzi, C. (2017). Introduction to special issue on the rhetoric of entrepreneurship: Theories, methodologies, and practices. Journal of Business and Technical Communication, 31(3), 275–289. https://doi.org/10.1177/1050651917695537

Spinuzzi, C., Pogue, G., Nelson, R. S., Thomson, K. S., Lorenzini, F., French, R. A., … & Momberger, J. (2015). How do entrepreneurs hone their pitches? Analyzing how pitch presentations develop in a technology commercialization competition. In Proceedings of the 33rd Annual International Conference on the Design of Communication (pp. 1–11). https://doi.org/10.1145/2775441.2775455

Thapa, N. (2020). Being cognizant of the amount of information: Curvilinear relationship between total-information and funding-success of crowdfunding campaigns. Journal of Business Venturing Insights, 14, e00195. https://doi.org/10.1016/j.jbvi.2020.e00195.

Tirdatov, I. (2014). Web-based crowd funding: Rhetoric of success. Technical Communication, 61(1), 3–24.

Tse, S. & Bryan, K. L. (2022). hiQ Labs v. LinkedIn. National Law Review, XII(109). https://www.natlawreview.com/article/hiq-labs-v-linkedin

Vealey, K. P., & Gerding, J. M. (2016). Rhetorical work in crowd-based entrepreneurship: Lessons learned from teaching crowdfunding as an emerging site of professional and technical communication. IEEE Transactions on Professional Communication, 59(4), 407–427. https://doi.org/10.1109/TPC.2016.2614742

Wang, X., & Gu, B. (2016). The communication design of WeChat: Ideological as well as technical aspects of social media. Communication Design Quarterly, 4(1), 23–35. https://doi.org/10.1145/2875501.2875503

West, S. (2017). Confronting negative narratives: The challenges of teaching professional social media use. Business and Professional Communication Quarterly, 80(4), 1–17. https://doi.org/10.1177/2329490617723118


Stephen Carradini is an Assistant Professor of Technical Communication at Arizona State University – Polytechnic campus. He teaches and researches social media in the workplace and digital ethics. His work has appeared in Journal of Business and Technical Communication, IEEE Transactions on Professional Communication, and Journal of Technical Writing and Communication, among others. He can be reached at Stephen.Carradini@asu.edu.

Eric Nystrom is an Associate Professor of History at Arizona State University's Polytechnic campus, where he studies the history of technology and digital history. His digital history research has appeared in venues including Law and History Review, American Journal of Legal History, Journal of Appellate Practice and Process, Mining History Journal, and Ordinary Lives, a forthcoming monograph from the University of Massachusetts Press.

[1]     Here and throughout the paper, we define “content” as all of the multimodal aspects that go into social media posts: text, emojis, images, audio, video, et al. We refer to alphanumeric text as “text.” We call the process of developing all aspects of a Kickstarter campaign “creating” a campaign. Producing the text of the campaign is called “writing” throughout.

APPENDIX A: TAG SET

The Readme of the Lingua::EN:Tagger includes a description of its tagset, which is the Penn Treebank tagset lightly modified to replace tags with numbers and those which would conflict with standard HTML tags. We reproduce the table below from Coburn (2019).