AI visibility
How Does AI Choose Which Businesses to Name?
When a customer asks ChatGPT to recommend a local accountant, a plumber, or a marketing consultant, the mental model most people bring is “search engine” — a ranked list ordered by relevance, recency, or some kind of quality score.
That’s not what’s happening.
AI engines don’t rank businesses. They recite businesses they’ve already encountered in their training data. If your business has been mentioned accurately, specifically, and repeatedly in sources the model treats as authoritative, it gets named. If it hasn’t — regardless of how good your website is, how many years you’ve been in operation, or what your Google ranking looks like — it doesn’t.
Understanding that distinction changes what “being visible to AI” actually requires.
Why the training corpus is the answer
AI language models are trained on large snapshots of the web — text pulled from websites, publications, directories, reviews, and other sources, captured up to a knowledge cutoff date. The model’s picture of your business is the sum of what those sources said about you, weighted by how authoritative the model considers each source.
This is meaningfully different from traditional search in two ways.
The engine isn’t checking your site in real time. When someone asks ChatGPT a question, the model isn’t crawling the web. It’s synthesizing an answer from what it learned during training. Your recently-updated service page may not be in the corpus at all, or may be less established in the model’s knowledge than a trade-publication mention from three years ago.
(Search-augmented AI tools — ChatGPT with web search enabled, Perplexity — partially close this gap by pulling live results. But the underlying model still shapes which sources it trusts and how it weights them when synthesizing an answer.)
The signal that predicts citation isn’t the one SEO taught you to track. Traditional search uses backlinks as the primary authority signal. Ahrefs studied 75,000 brands and found something different for AI citation share: backlinks show weak correlations. What does correlate — 0.656 to 0.709 across ChatGPT, AI Mode, and AI Overviews — is brand mentions in trusted web sources: directories, review platforms, industry publications, and news.
The strongest single-platform correlation in their data: YouTube brand mentions, at 0.737.
What “trusted sources” looks like in practice
The specific sources that matter vary by vertical, but the pattern is consistent: the engine names businesses it has “read about” in contexts it treats as credible, not businesses that have simply been telling their own story on their own website.
For a local service business, that tends to mean: a Google Business Profile, one or two category-specific directories (HomeAdvisor, Yelp, Angi, or whatever is authoritative in the vertical), a Nextdoor or local news mention, and reviews on a platform the engine indexes. For a B2B service, it’s more likely to be LinkedIn, industry association listings, and coverage in vertical trade publications.
The YouTube finding is worth sitting with. The likely reason a video platform correlates most strongly with AI citation share isn’t that AI engines watch video — it’s that YouTube is a platform where businesses appear by name alongside structured, specific, indexable information: business name in the description, service category in the title, location in metadata. It’s the combination of credible source + specific attributable facts that does the work.
That combination is the real pattern: the engine needs to have seen your business named, with consistent identifying facts, in sources it treats as authoritative for your vertical. Once, in one place, isn’t enough. Multiple times, in multiple relevant sources, with the same facts — that’s the profile an engine cites confidently.
Why inconsistency matters more than most people expect
Ahrefs’s tracking data shows AI Overview citations have become concentrated in a narrower pool of sources over time. Engines are citing fewer sources per answer, which means the businesses that get named are the ones the engine is most confident about.
Inconsistency creates exactly the kind of signal the engine needs to skip you and cite someone it’s more certain about. If your business name appears differently on your website than in Google Business Profile — a slight variation in the legal name, a DBA, an abbreviation — that’s a mismatch. If your phone number changed last year and some directories still list the old one, that’s a contradiction. If your service area description varies from one platform to the next, that’s a conflict the engine can’t confidently resolve.
The response to conflicting signals is usually not to name you. It’s to cite a business whose information is consistent and complete across the sources the engine has read.
What the audit actually looks at
There’s no shortcut to this — but the path is more specific than “do more SEO” or “get more reviews.” The question isn’t how many backlinks you have. It’s what the specific sources AI engines read say about your business: whether you’re present in the right ones for your vertical, whether the information they carry is consistent, and whether it’s specific enough for an engine to cite you confidently.
That’s a diagnostic question, not a general best-practices question. The answer is different for a licensed contractor in a mid-size metro than for a B2B consultant serving national clients. The AI Visibility Library has more on each of those angles as pieces land.
→ Here’s what we found when we ran this audit on ourselves — and what we looked at.
→ Get a free AI Visibility Snapshot — we’ll run the diagnostic for your business.
Brand-mention correlation data from Ahrefs’s study of 75,000 brands; AI Overview citation-concentration data from Ahrefs tracking. AI citation behavior changes as engines update training and detection methodology — if you’re reading this more than six months after the date above, check for more recent findings.