“ChatGPT doesn’t know us” is a sentence we hear from clients regularly, usually right after someone on the leadership team typed a question about companies in their space and didn’t like the answer. The catch is that ChatGPT doesn’t “know” companies the way Google does. It recommends based on an entirely different mechanism — and until that mechanism is understood, there’s not much you can do to change it.
Where ChatGPT’s recommendations actually come from
In practice, three layers operate at once. The first is the model’s training data — large public corpora: Wikipedia, Reddit, GitHub, Stack Overflow, industry media. Brands with a solid mention footprint there for years had a head start long before anyone was talking about “AI visibility.” The second layer is live web retrieval — for many queries, ChatGPT actually searches the web through its own systems, crawlers such as OAI-SearchBot, and search partners, which means current web presence genuinely matters today. The third layer is a preference for “trusted” sources within a given category — for B2B software that means places like G2, Capterra, or TrustRadius; for general topics, Reddit or YouTube.
This explains why a perfectly optimized company website isn’t enough on its own. The model doesn’t read it in isolation — it weighs it against what everyone else says about the company, and that second part more often decides the recommendation.
Where to actually start
Analyses cited in 2026 industry reports point to something that’s surprising at first: brand mentions across the web — in forums, reviews, and media — predict AI visibility far more reliably than inbound links, domain authority, or any other classic SEO metric. That flips the priorities a lot of companies are used to.
The practical sequence looks different from classic SEO:
Start with presence elsewhere. Active, genuine mentions on Reddit, YouTube, and industry review platforms (G2, Capterra, and similar), plus coverage in the go-to publications in your niche. This is where the model most often picks up the signal about who the credible players in a category actually are.
Then technical accessibility. Check whether robots.txt blocks crawlers like OAI-SearchBot, whether key pages are fully indexable, and whether they load fast enough to even be picked up during live retrieval.
Only then, owned content. Comparison pages, definitions of industry terms, concrete use-case write-ups — formatted so a model can easily extract them, and only after credibility has been built externally. Content on your own site, without outside confirmation, rarely does the job by itself.
It’s not just whether you’re mentioned, but how
Mention frequency is only half the picture — the other half is tone. A brand mentioned in half of responses with neutral or negative framing converts worse than one mentioned less often but described in clearly positive terms. So it’s worth regularly checking not just “are we showing up,” but also the language a model uses to describe your company — and acting when that description is inaccurate or out of date.
In practice, testing 30-50 buyer-relevant prompts weekly, across several models at once (ChatGPT, Perplexity, Gemini, Google AI Mode), works well — not to chase a single result, but to spot a trend before it turns into a real problem.
What this looks like in practice
At Unomage, we built platform.unomage.com precisely so this process doesn’t mean manually typing questions into a handful of chat windows once a quarter. The platform is publicly testable — you can check for yourself how specific brands are represented today in the answers of the leading models, before deciding where it’s worth investing time.
If you want to see where your company currently stands in ChatGPT and similar tools — and where to realistically start changing that — our team in Warsaw is happy to walk you through a first audit.
This article was created with the help of the Unomage AI platform.