The Founder Context Session: How to Get Tribal Knowledge Out of Your CEO's Head and Into Something Your AI Can Read

Tribal knowledge is the operating knowledge a company runs on that nobody has written down — it lives in people's heads and moves by conversation. In a startup, most of it is go-to-market truth: why you win, who you're really for, what you'd never claim, and what your best customers say. It sits in the founder's head, it has never been forced onto paper, and it is the largest piece of context your team's AI cannot read. This playbook is the forcing function: a structured interview, run by someone other than the founder, that turns tribal knowledge into claims a person can review and a machine can load.

What is tribal knowledge — and why is your founder's version the expensive kind?

Every company has three sources of go-to-market context. The customer layer — what buyers actually experienced, in their words. The market layer — what the category, the competitors, and the research say. And the founder layer — the perspective of the people who have been solving this problem longer than the company has had documentation. The first two can be gathered. The third is the one that does not exist anywhere outside a few people's heads.

That is what makes it expensive. The founder layer is where the provenance lives: not "we're not for insurance companies," but "we worked with two of them, the compliance layer made every deployment a six-month project, and the deals weren't worth it." The first version is a rule. The second is a reason — and a reason is what gives a junior rep conviction on a discovery call, what lets a new marketer position without guessing, and what lets an AI loaded with your context operate the way the founder would instead of the way a generic assistant does. Everyone has the same AI now; the founder layer is the part of your context no one else has (how it works).

The founder layer is where the "why" lives — and the "why" is what gives everyone else conviction.

Why can't the founder just write it down?

Because the people who know a problem best are the worst at explaining it. After enough years inside a space, the founder has compressed everything: the hard-won distinctions, the exceptions, the reasons behind the rules. All of it has become second nature — and second nature does not get broken down when it is spoken. The founder says the thing, moves on, and assumes the listener followed. To a buyer, or to a model, the compressed version is just an assertion.

The pitch deck makes this worse, because it looks like the record. It isn't. A deck is the founder's claims in presentation form — unchecked, undated, and stripped of the stories that made them true. A new hire who reads the deck has read what the founder asserts, not what the company has learned.

The fix is a second person. Nobody can read the label from inside the bottle, and a founder trying to document their own thinking will skip exactly the parts that matter most, because those are the parts that feel obvious. A neutral interviewer — the first marketer, a fractional partner, an advisor — can stop at every "obviously" and ask for the long version. People say sharper things to someone who isn't their account manager; the same is true pointed inward.

One way to see the stakes: at founder-led direct sales, a misread word costs nothing. The founder is on the call and corrects it in real time. The moment the company scales past the founder's calendar — ads, landing pages, a rep working a script, an AI drafting the outbound — that same ambiguity is the difference between a click and nothing. The founder context session is the cheapest point at which to catch it.

How do you run a founder context session?

Sixty to ninety minutes, recorded, on the record, with the founder as subject and someone else as interviewer. The output is not a transcript. It is a set of claims — each one decompressed, tagged with where it came from, and marked for what still has to be checked. The runbook below is the same one we use when a canon build starts with a client's founder.

Ground rules

• Stories over slogans. "Tell me about the last time that happened" beats "what's our positioning?" every time. The slogan is what the founder has already decided to say; the story is where the reason lives.
• Two question modes. Direct questions have an answerable fact behind them ("who are we definitely not for?"). Talk-about prompts are open riffs ("walk me through the deal we lost") where the founder's real language surfaces. Use both; the direct question gets you the claim, the talk-about prompt gets you the words.
• Follow the tangent. Some founders talk. Let them ride, then ask why that point was important enough to raise — the answer is usually a claim they didn't know they held.
• Index by frame. A sales-side CEO and a delivery-side CTO describe the same product from different seats. Note the seat with every claim; it changes what the claim means and where it belongs.
• Scope it first. If the company is working one segment, keep every question inside that segment. If you're feeding one content pillar, aim at that pillar. A bounded session produces claims; an unbounded one produces a long recording.

The engine: double-click every weight-bearing word

The one technique the whole session runs on. Whenever the founder uses a word that is carrying weight — a noun, a verb, a label for the problem — stop and ask: "Can I double-click on that? What does it mean to you, in this context?" Then do it again on the answer. The surface answer is always compressed; the value is one or two levels down.

Here is what that looks like in practice. On a positioning call with a vendor of AI product-management tooling, the founders were all former enterprise product leaders — and the product was finding traction with startups that have small engineering teams and none of the enterprise process. The word verify was everywhere: on the website, in sales calls, in the deck. "Our tool helps verify what was built." Asked what verify meant, the technical co-founder's answer was infrastructure and bug testing. Asked the same question from the business side, it meant user acceptance testing. From the product-strategy seat it meant delivered-as-intended. From the lifecycle seat it meant adopted. One word, four meanings, and a startup buyer who would hear whichever one they already knew.

Keep double-clicking and a family tree of terminology grows out of a single word. That tree is exactly what is missing from most positioning guides. A human listener eventually reconstructs it from context — after enough calls, they work out that the company means something close to user acceptance testing when it says verify, and adjust. An AI will not. Unless the tree is captured on purpose, every model drafting from your context will pick a meaning at random and ship it under your logo.

A human infers the meaning of your words from context over time. An AI won't — unless you captured it.

When to stop

This is the art-form part, so here is the honest rule: drill until the next level down no longer changes your positioning. You are not completing the tree. You are narrowing to the branch that maps to your ideal customer — the one meaning of verify that your buyer holds — and leaving the other branches for customer interviews and market research to finish. The founder's time is the scarce input; the founder session points the flashlight, it doesn't have to light the whole room.

Asked how many sessions it takes, the true answer is: as many as you can get. The real constraint is not the founder's patience with questions; it is whether they see the payoff. Read a processed claim back in the room — "here's what that story turns into" — and the founder will keep talking. Make them wait a week for a document and the second session never gets scheduled.

What do you ask? The question bank for a subject matter expert interview

A founder context session is a subject matter expert interview pointed at the company's own go-to-market. The question bank below is organized by the canon layer each answer fills, so the output lands somewhere instead of in a drive. It is deliberately not a script: the anchor questions are the familiar ones any positioning exercise uses, and the real work is the double-click under each. The full bank, with direct and talk-about tags, is in the downloadable guide further down.

• Positioning. What did you set out to solve, and what do customers actually buy you for now? Who are we definitely not for — and give me the story of how we learned that. Who are we up against, and why them? Walk me through the last deal we lost.
• Messaging. Describe the state the first ten customers were in before they found us. What's the moment it clicks for a buyer? What's the claim you'd bet the company on — and what makes you sure?
• Language. What do we call things, and why those words? What words do customers use that we don't? What have we stopped saying?
• Objections. What almost kills deals? What do you say back that works? Which objection turned out to be right?
• Proof. Which customer's story do you tell most often, and why that one? What's the number you wish you could cite? Who would go on the record tomorrow?

And the standing follow-up, laced through all of them: "You said that like it's obvious. Walk me through why it's true, as if I'm a buyer hearing it cold." The same bank works for any internal expert — a head of sales, a founding engineer, a long-tenured CS lead — which is why it doubles as a thought leadership interview guide: the material that makes a founder worth quoting is the decompressed version of what they think is obvious.

How do you turn the session into context your AI can load?

The mistake is to treat the transcript as the deliverable — parse it, tag it, call it the canon. It isn't. The founder session produces theses: the instincts and distinctions that take twenty years in a space to form, surfaced as claims about the product and who it serves. Each one is a starting point for research, not a finished fact.

Processing runs in four steps:

• Pull the claims and decompress them. Every assertion about the product, the buyer, the alternatives, or the outcome — written with the founder's own definition of any word that fans out. Keep the branch of the tree your ideal customer holds.
• Assign a proof state. Founder-asserted is a state, not proof. It sits alongside customer-confirmed, sourced, and retired — and it is the most common state a session produces. That is fine; it is honest, and it tells the next person what to go check.
• Write the follow-up question. An unproven founder claim is the next customer interview question. "We're not for enterprise" becomes "what did your security review involve, and how long did it take?" This is the hand-off from the founder layer to the customer layer, and it is the step that turns an opinion into context worth loading. The practice on the other side of that hand-off — voice-of-customer research as the product marketer's core job — is its own piece in this cluster.
• Route each claim to its file. A working canon is five plain files — positioning, messaging, language, proof, objections — and every processed claim has exactly one home (how to start a GTM canon). The never-say list fills fastest from a founder session: the retired terms, the competitor-owned words, the claims you walked back. Expect it, and treat it as the first win.

Then run the two gap checks the session exists to enable. First, internal alignment: does the current website, the last year of content, and the deck the reps actually present say what the founder just said — or has a year of production quietly contradicted it? Second, customer confirmation: have customers said this too, in their words, on the record? The gaps between those three layers — founder, marketing, customer — are the actual findings. The founder session generates the claims; the customer and market layers verify them. Founder judgment is the procedural half of your context; the data half still has to be captured from the people who bought (context engineering vs. building agents).

The founder session generates the claims. The customer layer verifies them. The gaps between the two are the findings.

The session guide

Everything above, as a document you can hand to the interviewer: the founder pre-read, a 75-minute agenda, ground rules, the question bank with direct and talk-about tags, the post-session processing checklist, and a worked example of three processed claims with their lineage visible. Copy it, adapt it, run it.

How often do you re-run it?

On the cadence the decisions move. Positioning changes quarterly at most, so the positioning block gets re-run quarterly. Language drifts continuously — re-run the language block whenever the words on the website stop matching the words in the sales calls. And re-run the whole session after any lost deal that surprised the founder, because surprise is the signal that a claim in the canon has quietly stopped being true. The canon's change ledger records what moved and why; the founder session is where most of those entries originate at a company under $10M ARR, because the founder is still where the judgment lives.

The honest limit

Everything on this page is free to run, and a founder session that lands as five plain files is a real improvement over the deck-and-Notion state most companies are in: the founder layer is captured instead of latent, the never-say list exists, and the team's AI has something to load that no competitor has. What the guide cannot do is finish the job. Founder-asserted claims stay founder-asserted until a customer says the same thing on the record, with consent to use it — and that capture, plus keeping every claim alive as the numbers age, is a motion rather than a document. It is the part Proofmap runs: the internal-expert interview is the first step of a canon build, the customer interviews that verify it follow, and Proofcanon keeps the messaging tied to what real people actually said.

Founder-asserted is a state, not proof. The session gets you to the state; customers get you to the proof.

Frequently asked questions

What if the founder is the one running marketing?

Then the founder cannot be the interviewer, because self-interviewing skips exactly the parts that feel obvious. Get a peer founder, an advisor, or a fractional partner to run it. The questions matter less than having a second person in the room who is allowed to say "what do you mean by that?"

How is this different from a positioning workshop?

A positioning workshop decides. A founder context session captures. The workshop produces a canvas the team has agreed to; the session produces the provenance behind it — the stories, the retired words, the reasons for every no. Do both, and do this one first, so the workshop argues from what the founder has actually learned rather than from what the deck already says.

How long before the AI output changes?

As soon as the processed claims are loaded as context. The first visible change is usually in what the AI stops saying — the never-say list takes effect immediately — followed by sharper answers on who the product is and isn't for. Claims that are still founder-asserted will read as confident but unproven, which is the correct behavior until a customer confirms them.

Isn't this the product marketer's job?

Yes — the product marketer is the natural interviewer, and the canon is where product-marketing judgment gets encoded so the whole team's AI can apply it (where a GTM canon lives, and who maintains it). Below the size where a dedicated PMM makes sense, the interviewer is whoever runs marketing, plus a third party when that person is the founder.

Does this work for organizational knowledge beyond go-to-market?

The method does — decompression and the double-click work on any subject matter expert. The question bank doesn't: it is built to fill the five go-to-market layers. For engineering or operations knowledge, keep the method and rewrite the bank around the files that function actually runs on.

Related: What is a GTM canon? · Context engineering vs. building agents · The AI-ready messaging framework · Out-prove, don't outspend · Why verified proof is the currency of AI citation