Why AI doesn't know your company

AI doesn't know your company because the knowledge that makes it specific was never written down. The workaround everyone uses, the judgment call behind the rule, which client never gets a Friday release — none of it sits in a document, so no model, and no search index over your files, can read it.

A language model holds two things: public text it was trained on, and whatever you paste in or connect. Your company's operating knowledge is in neither pile. It lives in people. Every real fix starts there.

A general model knows the published world: every framework, every postmortem that reached a blog, the averaged voice of every company that ever wrote about itself. Ask it how to run a retro or price a SaaS tier and it answers instantly, because thousands of companies wrote that down.

About your company it knows only what you hand it: the prompt, the files you attach, the systems you connect. That is why its advice sounds generic: it answers for the average company, the only company it has ever met. The specifics that would change the answer sit nowhere it can read.

The missing knowledge has a name: dark context — what an organization knows but never wrote down, the situated context a general AI model cannot see. The workaround that keeps invoicing alive. The reason the approval rule exists. The client you chased for a year and why it fell through. Who to ask before touching the deploy script.

None of it was ever recorded: too obvious, too awkward, or too busy a week. So it lives in people, in habits, and in the space between meetings. And it walks out the door when they leave.

You should connect your data. The AI still won't know your company. Retrieval over the wiki, the drive, the tickets, and the warehouse surfaces the knowledge you did record. Most of it sits unused (Gartner calls that dark data), and turning the lights on it is worth the work.

But the ceiling of every connector, pipeline, and integration is your written corpus. A perfect index over every document you ever produced still cannot retrieve what no one ever produced. The gap that keeps AI generic at your company was never a data gap.

No. Both work on text that already exists. Fine-tuning adjusts a model's weights on your written corpus; RAG retrieves passages from it. Each sharpens how the model uses what you wrote, and each inherits the same limit: words never written can be neither trained on nor retrieved. A custom GPT with your brand guidelines attached knows your fonts, not your judgment.

Bigger models don't close the gap either. They get more fluent about the average company. No technique manufactures input.

Capture here means structured conversation: the people who know the work, talking in their own words, one piece at a time, with the result kept as current, queryable memory rather than a one-off documentation sprint. Do it before the person who holds the knowledge resigns, retires, or forgets.

Why doesn't ChatGPT know my company?

ChatGPT was trained on public text and sees only what you type or connect to it. Your company's specific knowledge — the workarounds, the judgment calls, the reasons behind the rules — was never written down, so it exists nowhere ChatGPT can read.

Why doesn't AI understand my business context?

Because most business context is unwritten. It lives in people and in the space between meetings, not in documents. A model can only work with recorded text, so the knowledge that makes your business specific stays invisible to it.

Can I fix it by connecting my data or documents to AI?

Partly. Connecting your systems surfaces the knowledge you did record — worth doing, since most of it sits unused as dark data. But knowledge that was never recorded cannot be connected. It has to be captured from people first.

Is this solved by fine-tuning or RAG?

No. Both work on text that already exists: fine-tuning adjusts a model on your written corpus, RAG retrieves passages from it. Neither can supply knowledge no one ever wrote down. The input is missing, not the technique.

What is dark context?

Dark context is what an organization knows but never wrote down — the situated context a general AI model cannot see. The term was coined by Ola Möller and Andriy Zhukov at Dark Context, by analogy to dark matter and as a nod to Gartner's dark data.