Why doesn't ChatGPT recommend my Atlanta business?
Usually not because your website is bad. Three specific mechanics decide whether an AI answer names you, and only one of them is on your own site.
Most owners who ask this have already done the obvious things — the Google Business Profile is filled in, the reviews are decent, somebody added schema markup last year. And AI still names a competitor. Here is what is actually happening, with the research behind each part.
Most of the citations are not on your website
When an AI assistant answers a local question, it cites sources. Whitespark's study of 540 local queries found that about 60% of those citations point at third-party sites — directories, "best of" roundups, local press, forum threads — rather than the business's own domain.
That single fact reorders the whole job. If most of what the model reads about your category lives on other people's pages, then improving your own pages moves a minority of the signal. The work that matters is getting accurately represented on the sources the answers actually pull from.
The queries you're being judged on aren't the ones you optimised for
The same study split queries by intent and found AI Overviews appear on:
- only about 15% of classic "near me" searches
- but 92–97% of informational and hybrid questions — "how much does X cost in Atlanta", "what should I ask before booking X", "is X worth it"
Local SEO has spent a decade optimising the first group. The AI answers are overwhelmingly happening in the second. If your content answers no questions and only announces that you exist, there is very little for a model to quote.
The answer changes week to week, so one check tells you nothing
Orbit Media re-ran 72 identical prompts two days apart. Claude kept only 30% of the URLs it had cited before; Gemini kept 38%; Perplexity 65%. Across 1,792 query-and-domain combinations, all four engines agreed on the same cited source 1.7% of the time.
Two things follow. First, asking ChatGPT about your business once and concluding anything is a coin flip — you have sampled one draw from a distribution. Second, any vendor selling you a one-time "AI visibility score" is selling you that same coin flip with a certificate attached.
What about llms.txt?
Skip it. A study across 300,000 domains found no measurable relationship between having an
llms.txt file and being cited more often. Ahrefs looked at 137,000 sites: 97% of llms.txt
files received zero traffic, and of the 3% that got any, 96% of it was non-AI bots. Google has
said it isn't needed.
It is not harmful. It is just not the thing standing between you and being recommended, and if somebody sold it to you as a package, that tells you something about the rest of the package.
So what actually moves it
In order of how much they matter, based on where the citations are:
- Be accurately present on the third-party sources in your category — the directories and roundups the answers already cite. This is where the 60% is.
- Answer the informational questions in public, on your own site, in plain language. Those are the queries where AI answers appear almost always.
- Keep your entity data consistent — name, address, category, hours — so the model has no reason to doubt which business you are.
- Measure it repeatedly, because a single reading is noise.
That last one is why we do this as tracking rather than as an audit. A number you can watch move is the only version of this that tells you whether anything worked.
Sources: Whitespark, "AI Overviews in Local Search" (540 queries); Orbit Media, "AI Citation Sources" (13,184 citations); Ahrefs analysis of 137,000 sites and the 300,000-domain llms.txt study.