In 2026 business got a fashionable new thing: the company brain. It's an internal AI you feed everything your company has piled up — documents, chat threads, CRM, process docs. In return it answers your team's questions: what's the wholesale discount, how did we fix this breakdown last time, who owns this account.
A useful thing, no argument. But everyone deploying one shares the same fear: what if the AI answers wrong? The fear is justified. It's just pointed in the wrong direction.
Because your company already has a second brain like that, and it's getting things wrong right now. It's what ChatGPT and the other AI services know about you from the web. Nobody built it, filled it or checked it. And it isn't talking to your team. It's talking to your buyers — when they ask who to buy from.
The short answer
Your company has two brains. The internal one is the company brain: an AI knowledge base for your team. You're still planning to build it. The external one is what ChatGPT, Perplexity and Gemini tell anyone who asks about you. It's already built, without you, and it's already answering buyers.
Both break the same way: the AI pulls a wrong fact. When the internal brain is wrong, it costs you a wrong strategy and bad tactical calls. When the external one is wrong, it costs you a customer you'll never see. And without customers there's no strategy to get wrong. So start with the external one. Not instead. First.
What is a company brain, and why is everyone talking about it?
A company brain is a layer that takes everything your company knows (docs, CRM, issue trackers, chat threads, people's heads) and turns it into answers for people and memory for AI agents. The term was set loose in spring 2026, when Y Combinator — the world's best-known startup incubator — announced it was looking for teams to build companies this kind of brain. Then it snowballed: explainer videos, vendors, consultants.
Before advising anyone on this, I checked what's behind the fashion. I keep a set of scout scripts for exactly that: they measure how much a topic gets searched on Google, how many YouTube videos come out about it, what jobs are open, and who's discussing it on forums. I also measured the neighboring term, context engineering — that's what engineers call the same work. Here's the reading from September 24, 2026:
| What I measured | Result |
|---|---|
| "company brain" on Google | +86% year over year, −10% last quarter |
| YouTube views, last 12 months | 658k, 47% of videos are "build it yourself" |
| Context engineering videos per month | 572 (20 a year ago) |
| "Context engineer" jobs, US | 4,039, median salary $138,500 |
| Jobs titled "company brain" | 2 in the US, 0 in the UK and Germany |
| Reddit forum posts on "company brain", 30 days | 23, buyers among them: 1 |
How to read it. The fashion peaked in May and June: interest is alive but cooling. What keeps growing is the engineering name for the job, context engineering. There is money in the topic, but at the top end: the audit giant BDO, Dell and Databricks hire these engineers in-house at $138k a year. And on the forums, the company brain conversation runs between people building one and people selling one. Buyers barely appear: one in 23 posts.
The second source is a book, "The Company Brain," published in September 2026. It was written by Slite, a company that sells knowledge-base software itself, so I read it with that interest in mind. But its numbers are inconveniently honest for a seller. Slite surveyed 149 people who use such a brain or plan to get one:
- 71% expect a knowledge base that keeps itself accurate;
- only 17% have a company brain that actually works;
- the top worry about adopting one is accuracy, at 77%. Not cost, not setup;
- 79% keep knowledge in two or more systems at once.
Look what comes out: the core pain of the internal brain isn't "find the information," it's "don't lie." The book has a story about exactly that, told by a product manager at a fintech company. A junior, one month in, wrote a long report on a subject he didn't yet understand and saved it to the shared base. Before AI that was harmless: nobody ever found the page anyway.
"But all of a sudden, AI finds this single dump page and draws very strange conclusions from it."
Now the same thing, just outside the company. An old catalog page with 2022 prices. A review of a service you stopped offering. A directory with your old address. ChatGPT finds that dump and confidently retells it to your buyer.
Internal vs external brain: how do they compare?
| Internal brain (company brain) | External brain (ChatGPT, Perplexity, Gemini) | |
|---|---|---|
| Who asks | Employee, AI agent | Buyer |
| Status | 83% have no working one | Already live |
| Who built it | You, a months-long project | AI, from everything online about you |
| Cost of an error | Wrong strategy and bad tactical calls | A customer goes to a competitor |
| Do you see errors? | Eventually, in the results | No, the buyer just never shows up |
| Time to check | Weeks | One evening |
The first thing this table shows: the external brain is already running. Nobody ordered it, approved it or signed off on it. It assembled itself from your site, directories, reviews, old press releases and the forum threads where someone once asked about you. And right now it's answering the question "who should I pick."
The second: you don't see its mistakes. If the internal AI feeds you bad data, it eventually surfaces in the results of your decisions, and you go looking for the cause. But if ChatGPT quotes a buyer your three-year-old price, or doesn't name you at all among the four companies it lists, nobody comes to complain. The customer quietly goes to the ones it named.
Why start with the external brain?
It's closer to revenue
The internal brain shapes how you spend money. The external one decides whether there's money to spend. No customers, no money. No money, no company brain.
It's cheaper and faster
An internal knowledge-base pilot runs $20–50k and 4–8 weeks at typical agency rates. Meanwhile the research firm Gartner predicts that by the end of 2026 companies will abandon 60% of AI projects that lack properly prepared data. The external brain checks out in one evening, and the quick fixes on your own site take a week or two.
Something is already broken there, usually something dumb
This summer I did free teardowns of 34 websites — their owners sent me the links themselves in my Reddit thread. Six of the 34 were blocking AI crawlers, most often through a default Cloudflare protection setting the owner didn't know existed. ChatGPT simply couldn't read those sites. No company brain cures that.
The work carries over
For outside AI to tell the truth about you, you need one verified source of facts: what you do, for whom, at what price, where, and how you differ. Machine-readable markup, a facts page, consistent listings. That same layer later feeds your internal brain. You clean up once and use it twice.
And the catch, right away: technical markup alone won't save you. One of my clients, a manufacturer with a large catalog, had Product markup — machine-readable product descriptions for search engines — on 86% of their product pages. AI still mixed the products up: title from one, image from another, description from a third. Markup describes a page. It doesn't tell a machine how two nearly identical pages differ. That's cured with meaning, not tags.
The full catalog breakdownHow do you check what AI says about your company?
One evening is enough for a first reading:
- 1
Write down 15–20 questions your buyer actually asks. Not "Company X reviews" but "who's best at Y in Z," "how much does Y cost," "how is X different from Competitor." Anyone can be found by name. The question is whether you get found by the problem.
- 2
Ask them to ChatGPT, Perplexity and Gemini.
- 3
Count how many answers name you.
- 4
Write down every fact the AI states about you: prices, services, city, timelines, founding year. Check each one. Every wrong fact is that junior's report, except a buyer is reading it.
- 5
Ask about yourself by name, separately: prices, location, what you do. Don't be surprised by namesakes.
- 6
Check that your site lets AI crawlers in: open yoursite.com/robots.txt and, if you use Cloudflare, its settings. The bots GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended — that's ChatGPT, Perplexity, Claude and Google — should get through.
When should the internal brain come first?
To avoid "when you have a hammer, everything looks like a nail":
- you have 200+ people and onboarding takes months;
- your main pain is people drowning in process docs, not a shortage of customers.
Then build the internal one. Just remember that 77% fear exactly the same thing in it: a wrong answer.
Slite asked 149 people what worries them most about an internal company brain. 77% said wrong answers.
Nobody asked them what the external one is saying.
Want me to look?
I'll look at your site through the eyes of an AI crawler and send you a short read: whether the bots get in, which facts about you AI gets wrong, and where to start. Free, and no call for the sake of a call. After that, we'll see whether you actually need a second brain, and how I can help you build it.
Sources
- Slite, "The Company Brain" (survey of 149 people, 2026)
- Y Combinator, Requests for Startups
- Gartner: lack of AI-ready data puts AI projects at risk (Feb 26, 2025)
- Search, YouTube, job and Reddit data: author's demand radar, September 24, 2026
