When ChatGPT recommends a local business, it cites that business's own website in 31% of answers, and never as the only source

Ask ChatGPT or Perplexity "what's the best med spa in Dallas?" or "who's a good personal-injury lawyer near me?" and you get a confident, specific shortlist of named businesses. Where does that list come from? Every business owner assumes: my website. Fix the copy, add a services page, and the AI will find me. So we checked what the engines actually read — and the answer overturns that assumption.

Finding 1: The business's own website is cited in 31% of answers about it, and never alone

Across the 541 ChatGPT web-search answers we analyzed, each one a real recommendation about a specific local business, the engine cited that business's own website in 147 of the 472 answers where the business's domain was on record: 31%. In the other 69%, the recommendation was assembled entirely from third-party sources. Not once did the engine build an answer from the business's website alone: every one of the 147 answers that cited the homepage also cited outside sources, and most answers skipped the homepage completely.

The engines are not source-shy. On average ChatGPT cited 5.5 distinct sources per answer, drawing on 1,141 different domains across the 541 answers. It reads widely. Your website is one small, and usually skipped, input among them.

Finding 2: What it reads instead — review aggregators and "best-of" directories

The sources that actually shape the answer are third-party platforms that pool reviews and rank businesses. Here is what ChatGPT cited most, as a share of the 472 answers with a known business domain:

Read that list as a to-do. Almost every source that outranks your own website is a place you can appear — a Birdeye or BBB profile, a listing on a "best [service] in [city]" roundup, a Reddit thread where your name comes up. The engines converge on the businesses that show up across many of these at once. AI visibility is a consensus built off-site, and it is a list of sources you're either on or missing from.

Finding 3: It's true in every vertical — even the ones that lean on their site most

The own-website citation rate never gets close to a majority in any industry we measured:

Med spas and personal-injury firms lean on their own sites the most and still top out under 37%. For dentists and home-services companies it is under 24%: three of every four AI answers about such a business are built without its website in the citation set at all. If your entire AI-visibility strategy is "we redesigned the site," you are optimizing the one input the engine reaches for least.

What this means if you're a business — or the agency they pay

The old SEO reflex — rewrite the homepage, add pages, tune the meta tags — moves the surface the engines barely read. Getting recommended by AI is a different job: seeding and cleaning the third-party corpus the engines do read — review platforms, local directories, "best-of" listicles, the community threads where your category gets discussed. That is ongoing work, not a one-time fix, which is exactly why it's a retainer and not a checkbox. For agencies, this data is the argument: the client's site can be flawless and still be invisible to AI, because the engine is reading Birdeye and BBB, not the site you built.

But you can't fix a corpus you haven't measured. The first step is knowing which sources the engines are reading for your business, and which competitor they're naming instead — which is what an AskedAbout audit reports, per engine, per question.

Methodology

183 completed AskedAbout free AI-visibility checks across four local-service verticals: 65 med spas, 48 dentists, 42 personal-injury law firms, 28 home-services companies (June 12 to September 9, 2026), 1,084 parsed answers. Each check submits buyer-intent, category-and-location questions to the two engines in the free check, OpenAI's ChatGPT search model and Perplexity, and parses the responses. Only ChatGPT's web-search mode exposes the URLs it actually read: each answer carries its sources as links tagged utm_source=openai. Perplexity cites without surfacing source domains in a parseable form. So the source analysis here is computed over the 541 ChatGPT answers (of 1,084 total), every one of which carried at least one source, covering 182 distinct businesses. A cited host counts as the business's own website when it equals, or sits under, the domain the business entered at check time; 23 businesses (69 answers) entered no domain and are excluded from every share on this page, so the shares run over 472 answers about 159 businesses. Recounted 2026-10-03 on a fresh database export: the July 16 version read 28% over 511 answers because it counted the answers about businesses with no recorded domain as not cited; on that basis today's corpus reads 27% (147 of 541). Excluding the answers where the question cannot be decided is the correct measure and is what every figure above now uses; the personal-injury rate moves most (26% to 36%) because 10 of 42 firms entered no domain. Corrected 2026-07-21: an earlier version of this page computed these shares over a corpus that included three engineering QA rows — one of which, a concurrency stress test, wrote 274 answers by itself and skewed every answer-level share. The audit_cut filter now excludes them by exact name. Synthetic smoke-test rows are excluded by business name via our canonical audit_cut script — never by test flags, which our own prospect audits also carry. n and per-vertical splits are stated so you can weight them; we re-publish as the corpus grows.

More from the same corpus: the full AI-visibility benchmark across 152 local businesses, why AI recommends national chains instead of your shop, and why AI recommends your competitors. Reselling this to clients? The white-label Agency 5-pack puts five of these audits under your logo for $249.