You are always talking to the wrong audience
In a face-to-face conversation, you know who you are talking to. You see their expression shift. You hear their tone. You share a physical space, a social history, a set of unspoken rules about what this conversation is. You adjust in real time — a raised eyebrow makes you clarify, a laugh confirms the joke landed, a crossed arm tells you to change approach.
Now open Slack. Write a message. Hit send.
Who read it? Your manager who takes everything literally. The new hire who doesn't know the team's inside jokes. The colleague in another timezone who woke up to your message stripped of the three hours of conversation that preceded it. A future employee who will search the channel archive six months from now.
You wrote one message. It landed in five different contexts simultaneously. And you controlled none of them.
This is context collapse — and it is the defining communication failure of digital life.
Goffman's stages and the audiences you used to separate
To understand what digital communication breaks, you need to understand what face-to-face communication gives you for free.
In 1959, sociologist Erving Goffman published The Presentation of Self in Everyday Life, arguing that social life operates like a theater. You have a front stage — the performance you put on for a specific audience — and a back stage — where you relax, rehearse, and drop the act. Crucially, Goffman identified audience segregation as the mechanism that makes this work: you present different versions of yourself to different groups, and those groups don't overlap.
You talk to your manager differently than you talk to your best friend. You talk to your children differently than you talk to your therapist. Not because you are being fake — because context-appropriate communication is how meaning works. The same sentence, "I don't care," means something fundamentally different when said to a close friend during a relaxed conversation (genuine indifference to a trivial choice) versus said to a partner during an argument (emotional withdrawal). The words are the same. The audience, the relationship, the tone, the history — all of that is context, and context is what determines meaning.
Goffman's audience segregation kept your performances separate. Your boss never heard how you talked in the locker room. Your parents never read what you wrote to your college friends. The boundaries were physical — different places, different people, different conversations. And those physical boundaries did the context work automatically.
Digital communication demolished those boundaries.
Marwick and boyd: the collapse
In 2011, Alice Marwick and danah boyd published "I Tweet Honestly, I Tweet Passionately," the study that gave context collapse its name. They studied how Twitter users navigate the problem of having radically different audiences — colleagues, friends, family, strangers — all collapsed into a single follower list, all receiving the same broadcast.
Their core finding: social media technologies flatten multiple distinct audiences into one. The technical term is context collapse — "the collapsing of many different audiences into a single context" — and it breaks Goffman's audience segregation at a structural level. On Twitter (or any social platform), you cannot maintain different front stages for different audiences. You have one stage. Everyone watches.
Marwick and boyd found that users develop coping strategies. Some imagine a specific audience and write for them, ignoring everyone else. Some adopt the voice of a personal brand — a carefully managed, lowest-common-denominator self that offends no one and connects deeply with no one. Some use irony and plausible deniability. Some just stop posting.
But none of these strategies fully solve the problem, because context collapse is architectural. It's not a bug in how you use the platform. It's a feature of how the platform works.
The richness you lose: Daft and Lengel's hierarchy
Context collapse isn't just about audiences. It's also about what communication channels physically cannot carry.
In 1986, Richard Daft and Robert Lengel published their theory of media richness, which ranks communication media by their ability to convey complex, ambiguous information. Their hierarchy, from richest to leanest:
- Face-to-face: Immediate feedback, full body language, vocal tone, facial expression, shared physical context
- Video call: Facial expression and tone preserved, but body language truncated, physical context eliminated
- Phone call: Vocal tone preserved, all visual cues eliminated
- Personal written (email, DM): All paralinguistic cues eliminated, asynchronous feedback
- Formal written (memo, report): Stripped to content, no personal cues, no immediate feedback
Each step down the hierarchy permanently discards contextual information. A face-to-face conversation carries tone, timing, volume, pace, eye contact, posture, gesture, spatial proximity, and the full history of your relationship with that person as ambient context. A Slack message carries words.
Daft and Lengel identified four factors that determine media richness: (1) capacity for immediate feedback, (2) number of available cues and channels, (3) language variety, and (4) degree of personal focus. Text-based digital communication scores low on all four. You lose feedback speed (asynchronous), cues (no tone, no face, no body), language variety (text only), and personal focus (broadcast to many).
This means that the most ambiguous, emotionally complex, and relationship-dependent messages — exactly the ones that need the most context — are the ones most likely to fail in digital channels. And yet those are the messages people send via text every day.
The egocentrism trap: you think you're clearer than you are
Here is where it gets worse. Not only does digital communication strip context — you systematically fail to notice that it does.
In 2005, Justin Kruger, Nicholas Epley, Jason Parker, and Zhi-Wen Ng published a study in the Journal of Personality and Social Psychology titled "Egocentrism Over E-mail: Can We Communicate as Well as We Think?" Their finding: people dramatically overestimate their ability to convey tone and intent via email.
Across five experiments, participants believed they could communicate sarcasm, humor, and seriousness via email with about 78% accuracy. Their actual accuracy was 56% — barely better than chance. The mechanism is egocentrism: when you write a sarcastic email, you hear the sarcasm in your head. You read it back with the tone you intended. So it sounds clear to you. But your reader doesn't have your internal voice. They have words on a screen.
Kruger et al. demonstrated that this overconfidence is specifically egocentric — people fail to detach from their own perspective when evaluating how others will interpret the message. You know what you mean. You cannot un-know it. And that knowledge makes you blind to how the same words read without your context.
This is the L-0168 principle in action: the same words mean different things to different people. But digital communication amplifies this by removing every non-verbal cue that normally helps bridge the gap. Face-to-face, your tone of voice signals sarcasm. Over email, you're relying on your reader to infer it from text alone — and they infer it correctly only about half the time.
The 93% myth and the real cost of lost cues
At this point, someone usually cites Albert Mehrabian's famous "7-38-55 rule" — the claim that communication is 7% words, 38% tone, and 55% body language. If true, text-based communication would lose 93% of its meaning.
The rule is a misapplication. Mehrabian's 1967 studies were narrowly designed: participants judged the emotional attitude behind single spoken words like "dear" or "terrible" when tone and facial expression were manipulated to be inconsistent with the word. Mehrabian himself has repeatedly clarified that the formula applies only to the communication of feelings and attitudes through single words, not to communication in general. As he wrote: "Unless a communicator is talking about their feelings or attitudes, these equations are not applicable."
But the misapplication contains a useful signal. While you don't lose 93% of meaning in every text message, you do lose all paralinguistic cues — the vocal tone, facial expressions, body language, and physical co-presence that disambiguate words in real time. And for messages that carry emotional content, attitude, or relational complexity, the loss is severe. The exact percentage doesn't matter. What matters is that you systematically underestimate the loss.
Context collapse at scale: when audiences collide
The individual consequences are frustrating — a misread email, a joke that falls flat, a terse message that creates unnecessary anxiety. The collective consequences reshape public life.
When Marwick and boyd described context collapse, they were describing what happens when your boss, your mother, your ex, your professional peers, and ten thousand strangers all receive the same message simultaneously. A statement calibrated for one audience gets judged by all of them.
Jessica Vitak's 2012 research on Facebook disclosures found that as users' networks grew larger and more heterogeneous, they didn't post less — but they shifted toward a safer, more generic register. Positive emotional language increased. Negative emotional language decreased. Language style variability dropped. People didn't stop performing. They started performing the safest possible version of themselves — what Vitak describes as a "lowest common denominator" self.
This dynamic drives the phenomenon of public figures being held accountable for statements made in one context when those statements surface in another. A joke made to a small, trusted audience — where shared history provides the interpretive frame — gets screenshot and broadcast to millions who lack that frame. Without the original context, the words are reinterpreted by each new audience using their own context. The statement didn't change. The audience did. And meaning collapsed along with it.
This is not a defense of any particular statement. It's a structural observation: when context collapses, meaning collapses with it. Every audience fills in the missing context with their own assumptions, and those assumptions diverge.
AI and the new context collapse
Large language models process text. That's it. They don't hear tone. They don't see faces. They don't share your biographical history or cultural background. When you type a prompt into an LLM, you're communicating through the leanest possible channel — pure text, zero shared context, no relationship history.
This makes every interaction with an AI a context collapse event. The model doesn't know if you're being sarcastic. It doesn't know your professional background. It doesn't know that when you say "fix this," you mean "adjust the formatting" rather than "rewrite the logic." Every ambiguity in your prompt is a context gap that the model fills with statistical patterns rather than shared understanding.
The emerging discipline of context engineering — which has rapidly superseded "prompt engineering" as the operative concept — is fundamentally an attempt to restore the context that text strips away. When you write a detailed system prompt that specifies your role, expertise, desired tone, and output format, you are manually reconstructing the context that a face-to-face conversation provides automatically. You are compensating for context collapse by front-loading the contextual cues that text alone cannot carry.
This reveals something important: the problem isn't the AI. The problem is the channel. Every text-based communication channel collapses context. AI just makes the collapse visible because the model has zero ability to fill in what's missing from shared experience — it can only work with what you explicitly provide. Your human colleagues do the same thing when they misread your email. They just do it silently.
Protocol: communicate as if context doesn't exist
You cannot prevent context collapse. It's structural to digital communication. But you can compensate for it — if you stop assuming your reader shares your context.
Step 1: Assume zero shared context. Before sending any consequential message, re-read it as if you are a stranger encountering it for the first time. No tone of voice. No knowledge of the preceding conversation. No understanding of your relationship with the sender. If it could be misread, it will be misread.
Step 2: Make the implicit explicit. When your message carries emotional content, state the emotion directly. "I'm frustrated about the timeline" is unambiguous. "This timeline is interesting" is a context collapse grenade — it could be genuine curiosity, sarcasm, passive aggression, or deadpan humor, and the reader has to guess which.
Step 3: Match the channel to the complexity. Use Daft and Lengel's hierarchy as a decision tool. Routine, factual, low-ambiguity messages: text is fine. Complex, emotionally charged, relationship-dependent messages: move up the richness ladder. Have the difficult conversation on a call. Have the really difficult conversation face-to-face. The five minutes of scheduling save the five hours of damage control.
Step 4: Write for the unintended audience. Anything you write digitally can be forwarded, screenshot, or searched. Write every message as if it will be read by the person most likely to misinterpret it — because eventually, it will be. This isn't paranoia. It's structural awareness.
Step 5: Restore context when receiving. When a message lands wrong, your first question should not be "why did they say that?" It should be "what context am I missing?" The most productive response to an ambiguous message is not interpretation — it's inquiry. Ask what they meant. The three seconds it takes to clarify save the three days of resentment that misinterpretation produces.
The medium is the message — and the message is incomplete
Marshall McLuhan's famous observation gains precision through the lens of context collapse. The medium doesn't just shape the message — it amputates parts of it. Every communication channel is a filter that passes certain contextual signals and blocks others. Face-to-face passes almost everything. Text passes almost nothing except the words themselves.
You live in an era where most of your consequential communication happens through the leanest channels available. Your relationships, your reputation, your work, your ideas — all mediated through text that carries a fraction of what you intend.
The previous lesson showed that shared words don't guarantee shared meaning. This lesson shows why digital communication makes that gap structural and systematic. The words arrive. The tone doesn't. The intent doesn't. The relationship context doesn't. The audience you wrote for and the audience that reads are rarely the same.
The next lesson, L-0170, takes this further. Context collapse isn't just a digital communication problem — organizations impose their own context on every message, every decision, every behavior. The context that shapes you is bigger than any single conversation.