Why AI engines disagree about your brand

The afternoon I asked three engines the same question

I ran a simple test on our own brand a few weeks ago, the kind of thing I now do for every client before we touch a single page. I opened three tabs and asked ChatGPT, Perplexity, and Gemini the same question: "What does Acromatico do?"

One described us correctly as a photography studio that also runs AI visibility and GEO consulting. One led with the photography and never mentioned the consulting side at all — the same gap that used to exist on an old bio page we hadn't touched in over a year. One got the city wrong, pulling from an outdated directory listing that still showed an address we'd moved away from.

None of the three were "lying." Each one gave an accurate answer to the question of what it had actually read. That's the part most business owners miss when they get frustrated that "AI doesn't know us." The problem usually isn't the AI. It's that we, the business, have told three or four slightly different versions of our own story across the pages an AI engine can reach, and each engine picked up a different version.

Three reasons the same brand gets three answers

This isn't random. There are specific, documented mechanical reasons AI engines disagree with each other, and understanding them changes what you actually fix.

1. Frozen training data vs. live retrieval

Large language models are trained on a snapshot of the web up to a cutoff date, then some are paired with live web retrieval for certain queries and some are not. A model answering purely from training weights is describing your brand as it existed whenever that snapshot was taken — which could be a year or more out of date. A retrieval-augmented engine like Perplexity is built to fetch current pages on nearly every query, so it's more likely to reflect what your site says today. Two engines asked the identical question can be looking at genuinely different points in time.

2. Different crawlers, different access

Each AI company runs its own crawler with its own name, its own respect for robots.txt, and its own crawl schedule: Google uses Google-Extended for Gemini and AI features, OpenAI uses GPTBot, Anthropic runs ClaudeBot, and Perplexity operates PerplexityBot. If any one of those is blocked on a page, or that page returns an error, or it simply hasn't been crawled recently, that engine is working from whatever it saw last — while another engine that crawled you yesterday has a completely different picture.

3. Different source weighting

When an engine does retrieve live, it still has to choose which sources to trust. Your own site, your Google Business Profile, a directory listing, a press mention, and a review site can each say something slightly different about you — a founding date, a service list, a location. Every engine resolves that conflict differently, which is exactly why one engine can favor your own words and another can favor a stale directory entry it happens to weight more heavily.

Why this matters more than a ranking drop ever did

A ranking drop is visible — you check Search Console and the number is lower. A disagreement between AI engines is invisible unless you go looking for it, and it's arguably worse, because it isn't costing you a rank, it's costing you the story a prospect hears before they ever visit your site. If someone asks ChatGPT what your company does and gets the wrong answer, you don't get a chance to correct them in a sales call. The wrong version already landed.

This is different from the "fix your facts" problem of pricing or founding dates being wrong in one place. Facts are binary — right or wrong. Story consistency is softer and easier to let drift: your homepage says one positioning, your About page says another, your last press mention describes a service you deprioritized two years ago, and none of it is technically false. It's just no longer the same story, and an AI engine has no way to know which version is current unless you make one version unmistakably dominant.

How to actually audit the disagreement

Before you touch a page, find out how bad the drift actually is. This takes about twenty minutes and costs nothing:

• Ask the same question three ways. "What does [your company] do?", "What does [your company] specialize in?", and "Who is [your company] for?" — across ChatGPT, Perplexity, and Gemini at minimum.
• Write down the differences, not just the errors. Note where the emphasis shifts, not only where a fact is flat wrong. Emphasis drift is the early warning sign of a story that's splitting.
• Trace each wrong or dated detail back to its source. An outdated service description often traces to an old blog post or a directory listing you forgot existed, not to the AI "hallucinating."
• Check whether your own pages agree with each other. Open your homepage, About page, and most recent press mention side by side. If they don't tell the same story, no engine can be expected to either.

This is the same method behind a full DIY AI visibility audit — this version is scoped specifically to catch disagreement rather than absence.

Making the story consistent, on purpose

You cannot email an AI engine and ask it to update its answer. What you can do is make the accurate, current version of your story so dominant across every crawlable surface that it becomes the path of least resistance for any engine, regardless of when it last crawled you or which sources it trusts.

Pick one canonical description and reuse it everywhere

Write a single, precise sentence describing what your business does and who it's for. Use that exact sentence — not a rephrased version — in your site's meta description, your About page, your Google Business Profile, your LinkedIn company page, and any directory you control. Repetition of the same wording across independent surfaces is one of the clearest consistency signals a retrieval system can pick up.

Mark it up so machines don't have to guess

Add Organization schema to your homepage with your name, description, and service categories filled in explicitly. Structured data removes the ambiguity of an engine having to infer your positioning from prose — it just reads the field. Our schema audit guide walks through checking whether yours is actually complete.

Retire the pages that tell the old story

Old blog posts, outdated bios, and abandoned directory listings don't disappear just because you stopped updating them — they stay crawlable and keep feeding the old version of your story to any engine that reaches them. Update or redirect anything that describes a service you no longer offer, a location you've left, or a positioning you've moved on from.

Keep your About page doing the heavy lifting

Of every page on a site, the About page tends to carry the most weight for exactly this kind of query, because it's where a business is expected to state plainly who it is. We cover why in your About page is doing more SEO work than your blog — the same reasoning applies directly to which page an AI engine leans on to answer "what does this company do."

Re-check on a schedule, not once

Crawlers don't refresh on your calendar. Re-run the three-question audit above every time you materially change your positioning, and again on a quarterly cadence even if nothing changed, since a directory or third-party page can drift out from under you without any action on your part.

What this looked like for our own bio page

Going back to that afternoon: the fix wasn't complicated. The old bio page that undersold the consulting side got rewritten to match the same sentence we now use on the homepage and in our Organization schema. The outdated directory listing got updated to the current address. Neither of those took more than an hour combined — the hard part was noticing the drift existed in the first place, which is the entire reason the audit step above comes before anything else.

Italo Campilii

Founder, Acromatico · runs AI visibility & brand systems across 10+ live brands