Self-presentation is the work: context collapse and the imagined audience

Set the profile beside a paper diary and the argument comes into focus: what looks like identity expression online is identity management under a handicap. A profile and a feed address an audience whose membership and size cannot be read off the screen. Users respond not by expressing a settled self but by inventing strategies to guess and steer who will see them (Marwick and boyd, 2011). The sixth of ten units argues three linked claims. The audience online is necessarily imagined. The difficulty of imagining it is a property of platform affordances rather than of user naivety. The elaborate strategies people devise are evidence for the second claim rather than against it (boyd, 2010; Litt, 2012). The seventh unit takes the same problem of visibility into work the platform assigns and pays for.

An audience you cannot see is an audience you have to imagine

Goffman (1959) argued that social interaction is a performance in which people manage the impression they give. They stage it in the front region and prepare it in the back, so the audience's reading is what the performance is organised around. What a network site changes is not the need to perform but the audience. Marwick and boyd (2011) interviewed Twitter users with very different followings. They found that flattening many audiences into one timeline makes audience-specific self-presentation impossible. Users therefore work from an imagined audience, often the most sensitive reader they can picture, using targeting, concealment and maintained authenticity. Litt (2012) turns that finding into a construct with a theory behind it. She draws on Giddens's account of structuration. What people picture as their audience depends on available norms and routines. The imagined and the actual audience are in tension, and this imagined composition mediates between a platform's reach and what is posted. The performance is aimed at a guess.

Collapse is a property of the platform, not a lapse in the user

boyd (2010) characterised these environments as networked publics and named four affordances that distinguish them from earlier publics: persistence, replicability, scalability and searchability. Those affordances generate three dynamics users must manage: invisible audiences, collapsed contexts, and a blurring of the public and the private. Context collapse is not an accident of one site but a consequence of default properties, content that persists and can be copied, searched and spread far beyond its moment of creation. boyd and Ellison (2007) supply the definition beneath the account. A social network site lets people construct a profile, articulate a list of connections, and view and traverse those lists. The defining practice is not meeting strangers but making an existing network visible. Because profile and connection list are the site's basic furniture, that visibility is a default rather than a considered choice, and the sites differed in how much they exposed without asking.

Bernstein and colleagues (2011) studied 4chan's /b/, a board with over 7 million users where almost nothing persists. Most threads left the first page within five seconds and expired from the site in under five minutes. More than 90 per cent of posts were fully anonymous. The community worked anyway, through the opposite of the Facebook regime: status came from in-jokes and references re-earned in each thread. The case bounds the claim: persistence, replicability, scalability and searchability are defaults of the dominant sites, not of the medium as such.

Hogan (2010) sharpens what is new. Goffman's dramaturgy assumes a co-present audience; social media work largely through exhibitions, asynchronous sets of updates and photographs assembled over time and viewed when the viewer chooses. The exhibition has a curator as well as a performer, because the platform or the ranking decides what is shown. A feed read through a screen reader stages a different exhibition. This curatorship pushes display towards a lowest common denominator safe for every audience at once.

Display is an exhibition, and the imagined audience can be measured

If the audience is imagined, the gap between the imagined and the real one should be measurable, and Bernstein and colleagues (2013) measured it. Combining a survey with log data covering posts by 222,000 Facebook users over a month, they found users systematically underestimated audience size, guessing it was about 27 per cent of its true size. Folk theories for estimating reach did not work. Counting friends or reading likes varied widely without indicating who saw a given post. Users still typically did reach about 61 per cent of their friends each month. The imagined audience is wrong in a direction that makes people post as if fewer were watching. Vitak (2012) shows the measured consequences. She surveyed Facebook users and modelled network composition, disclosure, privacy strategies and social capital together. Audience size and diversity shaped both what people disclosed and whether they used advanced privacy settings. Audience and disclosure characteristics also predicted bridging social capital.

The anchored profile shapes what can be shown, and it distributes its costs

The performance is staged on a site that ties the performer to a name. Zhao and colleagues (2008) analysed 63 Facebook accounts. They found users in this nonymous setting claimed identities implicitly rather than explicitly, showing rather than telling and stressing group and consumer identities over personally narrated ones. Those identities were hoped-for possible selves projected past the gating offline life imposed. The compensation for that constraint is measurable. Ellison, Steinfield and Lampe (2007) surveyed 286 undergraduates. They found Facebook use associated with three forms of social capital, most strongly bridging social capital, the loose ties that supply information and opportunity. The association was strongest for those low in self-esteem and life satisfaction. The profile work this unit describes returns network resources to the people who do it.

Duguay (2016) interviewed 27 LGBTQ young people in the United Kingdom and walked through their Facebook accounts with them. Context collapse forced a continuous and unequal management of disclosure. Participants reinstated the contexts the platform had flattened by tailoring performances to separate audiences and keeping audiences apart. The burden fell hardest on those whose identity was not assumed in the default audience. Were collapse a matter of user skill, the burden would not fall this way.

Contested question: is context collapse a condition of the technology or a skill users can acquire?

Three positions are argued about context collapse. boyd (2010) and Marwick and boyd (2011) hold the first position, which is structural. The affordances of networked publics produce invisible audiences and collapsed contexts. Users can only manage the consequences, so the strategies they invent are coping rather than solutions. The second, held by DeVito and colleagues (2017), is that self-presentation is platform-conditional and skilled. From an exploratory study of 193 users they built an affordance-based taxonomy. They found significant variation in how people perceived the same affordances across platforms, experience and personality. That is what one would expect if users learn to read each platform and adjust. On this reading the strategies are competence rather than evidence of a design fault.

Bernstein and colleagues (2013) and Vitak (2012) hold the third position. It recasts the problem as measurement and tooling. The gap is real but quantifiable, and privacy controls do change disclosure. The difficulty is that sites do not show users who is present or offer control fine enough to restore a context. What would settle it is a manipulation none of these studies performs: vary audience transparency inside one platform, show a random half of users their true recent audience, and measure whether strategies and disclosures change. A within-person design across platforms would separate the structural claim from the skill claim. It holds the person constant while the affordances vary. The taxonomy of DeVito and colleagues (2017) is the instrument it needs.

Rate the three positions rather than repeating them. The structural position rests on interviews with Twitter users and on a conceptual account of affordances, so it infers the constraint from what users describe. The skill position rests on an exploratory study of 193 users, with perception measured rather than behaviour (DeVito et al., 2017). The measurement position carries the strongest numbers, a survey joined to logs covering 222,000 users (Bernstein et al., 2013). While the dispute is open, a designer should show people who actually saw a post before offering a control over the next one. The strongest objection is that the same affordances carry different costs for different users (Duguay, 2016).

Where the argument is heading

The frontier is authorship. Jakesch and colleagues (2019) ran three experiments in which participants judged Airbnb hosts whose profile text was labelled, or suspected, to be written by artificial intelligence. They found a Replicant Effect: mistrust arose specifically when participants believed they were seeing a mixed set of human and machine writing. The identity signal now extends from content to authorship. Hohenstein and colleagues (2023) push the question into the interaction. In two randomised experiments on algorithmic response suggestions, the replies changed the language people used, sped up communication and made partners rate each other as closer. Suspicion drew negative evaluation. Self-presentation is being partly delegated to a model, and the delegation is now itself a signal.

The sources this unit rests on

Each note states what the document did and found. Identifiers were resolved on 2 November 2026.

• Bernstein, M. S., Bakshy, E., Burke, M. and Karrer, B. (2013). Quantifying the invisible audience in social networks. doi:10.1145/2470654.2470658 Combined a survey with log data on posts by 222,000 Facebook users and found users underestimated their audience, guessing 27 per cent of its true size. It makes the imagined audience a measured gap.
• boyd, D. (2010). Social network sites as networked publics: affordances, dynamics and implications. https://doi.org/10.4324/9780203876527 Named four affordances of networked publics, persistence, replicability, scalability and searchability, and three dynamics, invisible audiences, collapsed contexts and a blurring of public and private. It makes the unit's claim structural.
• boyd, D. and Ellison, N. B. (2007). Social network sites: definition, history, and scholarship. doi:10.1111/j.1083-6101.2007.00393.x Defined social network sites by profile, articulated connection list and traversal, arguing the defining practice is making an existing network visible. It grounds the visible network as a default of the form.
• DeVito, M. A., Birnholtz, J. P. and Hancock, J. T. (2017). Platforms, people, and perception: using affordances to understand self-presentation on social media. doi:10.1145/2998181.2998192 Built an affordance-based taxonomy of self-presentation and tested it on 193 users, finding the same affordances perceived differently across platforms, experience and personality. It is the position that self-presentation is skilled.
• Duguay, S. (2016). He has a way gayer Facebook than I do: investigating sexual identity disclosure and context collapse on a social networking site. doi:10.1177/1461444814549930 Interviewed 27 LGBTQ young people in the United Kingdom and found context collapse forced unequal, continuous management of disclosure through tailored performances and separated audiences. It shows the cost falls on those the default audience would not accept.
• Ellison, N. B., Steinfield, C. W. and Lampe, C. A. (2007). The benefits of Facebook friends: social capital and college students' use of online social network sites. doi:10.1111/j.1083-6101.2007.00367.x Surveyed 286 undergraduates and found Facebook use most strongly associated with bridging social capital, with the association strongest for those low in self-esteem. It shows profile work produces measurable returns.
• Goffman, E. (1959). The presentation of self in everyday life. https://monoskop.org/images/1/19/Goffman_Erving_The_Presentation_of_Self_in_Everyday_Life.pdf Argued that social interaction is a performance in which people manage the impression they give, staged in a front region and prepared in a back region. It is the dramaturgical foundation for reading a profile as managed performance.
• Hogan, B. (2010). The presentation of self in the age of social media: distinguishing performances and exhibitions online. doi:10.1177/0270467610385893 Distinguished synchronous performances from asynchronous exhibitions and named the virtual curator that gathers display, pushing it towards a lowest common denominator. It explains what is new beyond many audiences.
• Hohenstein, J., Kizilcec, R. F., DiFranzo, D. J., Aghajari, Z., Mieczkowski, H., Levy, K. E. C., Naaman, M., Hancock, J. T. and Jung, M. F. (2023). Artificial intelligence in communication impacts language and social relationships. doi:10.1038/s41598-023-30938-9 Ran two randomised experiments on algorithmic response suggestions and found the replies changed language, sped up communication and made partners rate each other as closer, while suspicion drew negative evaluation. It shows self-presentation is now partly delegated.
• Jakesch, M., French, M., Ma, X., Hancock, J. T. and Naaman, M. (2019). AI-mediated communication: how the perception that profile text was written by AI affects trustworthiness. doi:10.1145/3290605.3300469 Ran three experiments on Airbnb host profiles and found a Replicant Effect: mistrust arose only when a mixed set of human and machine writing was believed to be in view. It extends the identity signal to authorship.
• Litt, E. (2012). Knock, knock. Who's there? The imagined audience. doi:10.1080/08838151.2012.705195 Turned the imagined audience into a theoretical construct, drawing on Giddens's structuration, whose composition depends on available norms and mediates between reach and what is posted. It supplies the construct the argument turns on.
• Marwick, A. E. and boyd, D. (2011). I tweet honestly, I tweet passionately: Twitter users, context collapse, and the imagined audience. doi:10.1177/1461444810365313 Interviewed Twitter users with different followings and found a flattened audience makes audience-specific self-presentation impossible, so users work from an imagined audience using targeting, concealment and maintained authenticity. It is the paper the thesis is built on.
• Vitak, J. (2012). The impact of context collapse and privacy on social network site disclosures. doi:10.1080/08838151.2012.732140 Surveyed Facebook users and found audience size and diversity shaped disclosures and the use of advanced privacy settings, and that audience and disclosure predicted bridging social capital. It gives collapse a behavioural signature.
• Zhao, S., Grasmuck, S. and Martin, J. (2008). Identity construction on Facebook: digital empowerment in anchored relationships. doi:10.1016/j.chb.2008.02.012 Analysed 63 Facebook accounts and found users claimed identities implicitly, showing rather than telling and stressing group and consumer identities over narrated ones. It shows the anchoring regime shapes what can be expressed.
• Bernstein, M. S., Monroy-Hernandez, A., Harry, D., Andre, P., Panovich, K. and Vargas, G. (2011). 4chan and /b/: an analysis of anonymity and ephemerality in a large online community. doi:10.1609/icwsm.v5i1.14134 Studied 4chan's /b/, where most threads left the first page in five seconds and expired within five minutes and over 90 per cent of posts were anonymous, yet the community thrived. It shows the network-site affordances are defaults rather than necessities.