Founder media
Why AI-generated founder content sounds generic, and how to use AI without that problem
A language model writing from a short prompt produces the most likely version of a post on that topic, which is close to what everyone else's model produces.
It has no access to the founder's examples, judgments or doubts unless they are supplied. Readers notice, and LinkedIn now reduces reach for content members flag as AI-written. Use AI for research, transcription and checking drafts, and keep the substance and the final wording grounded in the founder's own words.
Why the output converges
A language model predicts likely text. Given "write a LinkedIn post about why onboarding matters for AI products", it produces a fluent average of the many posts like that it has seen. Ask several founders' models the same thing and you get similar posts with different names attached.
What makes a founder's writing distinct is missing from the prompt: a specific customer who churned, a decision that went wrong, a number from their own product, a view their peers would argue with. The model cannot invent those truthfully.
How readers respond
Technical readers are sensitive to this. In September 2026, Colin Breck's essay "I don't want to read what you didn't write" reached the front page of Hacker News, arguing that writing produced by a model loses the author's reasoning and judgment. Thomas Ptacek's guide to writing with an LLM, published the same month, sets a strict first rule: "You may not use a single word an LLM suggests to you." He recommends using the model as a copyeditor, not a ghostwriter.
Platforms are responding as well. LinkedIn added a way for members to flag posts that seem like AI slop; Fortune reported more than a million clicks in the first two weeks and 40% fewer views for flagged content.
For a technical founder, the cost is reputational as well as reach. A generic post signals that the founder did not think about it, which is the opposite of what the post was meant to show.
Patterns that make writing read as AI-generated
- Contrast setups that reject one idea to introduce another
- Lists of three abstract nouns
- Dramatic one-line paragraphs and quotable closing lines
- Confident claims about "most founders" with no evidence
- Vague words such as "unlock", "leverage", "game-changing"
- Invented anecdotes ("a founder told me last week...")
- The same structure in every post
A workflow that keeps the founder's voice
| Step | AI is useful for | The founder or editor does |
|---|---|---|
| Research | Summarising sources, finding discussions, checking facts | Choosing what matters |
| Interview | Transcription | Asking and answering the questions |
| Selection | Searching the transcript | Picking the passages with a real view |
| Drafting | Little, if any | Writing from the founder's words |
| Editing | Flagging repetition, filler, unclear sentences | Deciding what to change |
| Approval | Nothing | The founder approves |
Learning the voice
A founder's voice is learned from corrections. Keep a file of phrases the founder uses, phrases they reject, claims they want qualified and topics they avoid. Each round of edits makes the next draft closer. After a few months the founder should be correcting details, not rewriting.
How I use AI at OwnedSignal
For research support, transcription and checking drafts against a list of patterns like the one above. Not for inventing opinions, and not to replace the interview. Every published piece starts from something the founder said or a source that can be cited, and the founder approves the final version.
Sources
- I don't want to read what you didn't write, Colin Breck, 2026-09-20.
- How to write with an LLM, Thomas Ptacek, sockpuppet.org, 2026-09-17.
- 1 million people clicked LinkedIn's AI slop button, Fortune, 2026-08-25.