How AI Decides What to Say About Your Brand — and How to Change It

When ChatGPT recommends you, ignores you, or gets you wrong, it's not random. There's a mechanism behind it. Understand the four forces that shape every AI answer, and you can start steering them.
Key Takeaways
AI answers aren't opinions — they're assembled. Models build a response from training data, live web retrieval, and a confidence judgment about what's safe to say.
Four forces decide your fate: what the model learned in training, what it retrieves live, how much the web agrees about you, and how machine-readable your facts are.
The black box is more influenceable than it looks. Each force has a lever you can pull — fresh content, consensus, citeable sources, clean structure.
You're already in the system, like it or not. The model is answering questions about your brand today, from whatever it can find. The only choice is whether you shape that input.
Stop treating the AI like a mind, start treating it like a machine
It's tempting to think of an AI assistant as having an opinion of your brand — that it "likes" your competitor or "thinks" you're expensive. That framing is comforting and useless. An assistant doesn't hold opinions; it assembles answers from inputs, every single time, according to a mechanism. And a mechanism, unlike an opinion, can be understood and influenced.

This matters because the stakes are no longer small. AI's share of web traffic nearly doubled year-over-year (SE Ranking, 101,574-site dataset), and a growing share of buyers accept an AI's answer as the verdict. So when the machine describes your brand, that description is doing real work in real deals. The good news: once you see the four forces feeding every answer, "what the AI says about us" stops being a mystery and becomes a set of levers.

Force 1 — Training knowledge: what the model already "knows"
Every model carries a vast store of patterns learned from the web up to a cutoff date. This is the baseline it reasons from when you aren't using live search — and it's why a model can describe your brand confidently without looking anything up. If your category was well-documented when the model trained, it has a rich picture; if not, it improvises from fragments.

The catch: training knowledge is a snapshot, and it can be stale. A model may "remember" your old pricing, a discontinued product, or a founder who left — and state it with total confidence. You can't retrain the model, but you can shape what future versions learn by building a strong, consistent presence across the web now, so the next snapshot captures you accurately. Training is the slow lever, but it compounds.

Force 2 — Live retrieval: what it fetches in the moment
Many assistants — Perplexity, ChatGPT with search, Gemini — don't rely on memory alone. They retrieve live web pages at the moment of the question and build the answer from what they find, citing sources as they go. This is the fast lever, and the one you can move this week.

Here's the practical consequence: for these engines, your current, crawlable, well-structured pages directly feed the answer. If your site clearly states the facts and the crawler can read them, you get pulled in and cited. If your facts are stale, buried, or locked in images, the model retrieves someone else's version of you instead. Live retrieval is where fresh, extractable content pays off fastest.

Force 3 — Consensus: how much the web agrees about you
This is the force most people miss. An AI doesn't trust a single page; it triangulates. When many independent, credible sources describe you the same way, the model treats it as established fact and states it confidently. When sources disagree — or there's only a lonely claim on your own site — it hedges, caveats, or leaves you out.

So a lot of what the AI "says" is really an echo of what the web agrees on. That's why consistency is a strategy, not a nicety: your category, your strengths, your key facts should read the same on your site, in directories, in reviews, and in third-party coverage. You're not just publishing claims; you're building the corroboration that turns a claim into something the model will repeat. Consensus is what converts "your marketing says so" into "the AI says so."

Force 4 — Machine-readability: whether it can use you at all
The fourth force underpins the other three: none of your content can influence an answer if the machine can't parse it. An AI extracts text, structure, and signals — not visuals or vibes. Facts trapped in images, content gated behind JavaScript, and pages with no structured data are effectively invisible, no matter how good they are.

This is the unglamorous, high-leverage lever. Clean, crawlable, well-structured pages with schema markup and explicit canonical facts (down to an llms.txt pointing to your sources of truth) make you usable to the model. Get this wrong and your best content never reaches the answer; get it right and everything else you do starts to count.

Put the four forces together
Now the picture resolves. When a buyer asks about your category, the assistant blends what it learned (training), what it can fetch (retrieval), what the web agrees on (consensus), and what it can actually read (machine-readability) — and produces an answer that recommends you, ignores you, or describes you wrongly. Each of those four inputs is something you can influence:

Training → build a strong, consistent web presence so future models learn you correctly.
Retrieval → keep fresh, crawlable, extractable pages that state your facts plainly.
Consensus → make the web agree about you with consistent facts everywhere.
Machine-readability → ship clean structure, schema, and explicit canonical facts.
Pull all four and you stop being at the mercy of the black box. You become an input to it.

You can't steer what you can't see
There's one thing the four forces can't give you on their own: visibility into the output. You can pull every lever and still not know what the assistant is actually telling buyers today — which is exactly the blind spot Sourceable removes. It shows you how AI currently describes, cites, and recommends your brand across ChatGPT, Gemini, Perplexity, and Claude, so you can see which forces are working for you and which are working against you — and watch the answer change as you act.

The AI is already deciding what to say about you. Now you know how it decides — and that you have more say in it than you thought.

See what AI says about your brand with Sourceable.