To show up in ChatGPT recommendations, do three things: build a strong presence on the third-party sources ChatGPT cites (review platforms, directories, and communities), make your own website machine-readable (server-rendered, clearly structured, with answers stated up front), and earn corroboration — quotes, statistics, and business facts that stay consistent everywhere you appear online. ChatGPT does not rank websites the way Google does; it assembles answers from sources it trusts, so the real goal is to become the business those trusted sources already agree on.
In 2026, 45% of consumers ask AI assistants for local business recommendations, up from just 6% a year earlier, according to BrightLocal. For a small business, ChatGPT has quietly become a referral channel on par with the map pack — and most owners have never checked what ChatGPT says about them.
The pattern holds beyond consumers. G2's 2026 research found that 51% of B2B buyers now start research with AI chatbots more often than with Google, and 69% have switched vendors based on AI guidance. Adobe data shows visitors referred by AI convert 54% better than average: fewer clicks, warmer buyers.
Roughly 60% of searches now end without a click, per Bain. When a customer asks ChatGPT "who is the best plumber near me," the recommendation itself is the entire result. A business that is not in the answer has no page two to fall back on.
ChatGPT rarely recommends a business because of that business's own website alone. ChatGPT leans on third-party sources it already trusts: review platforms, established directories, community threads on Reddit, YouTube videos, and "best of" listicle articles. The first job is to exist — accurately and positively — in those places.
Reviews come first, because review volume and recency are signals every AI engine reads. Then work down the list:
Skip the obscure paid directories. AI engines pull from a fairly narrow set of trusted sources for any given question, so ten complete, accurate, review-rich profiles will do more for your ChatGPT visibility than a hundred thin listings on sites no engine ever cites.
Most AI crawlers do not execute JavaScript. If a website only renders its content in the browser — common with modern site builders and single-page frameworks — an AI crawler may see a nearly blank page. Server-rendered HTML, where the words exist in the raw page source, is the baseline requirement.
Structure every important page so a machine can lift an answer from it. Put the direct answer in the first two sentences, use descriptive headings, and write short self-contained paragraphs that make sense out of context. AI engines quote passages, not pages, so write passages worth quoting.
Keep pages current: content updated within the last three months earns an average of 6 AI citations versus 3.9 for content more than two years old (SE Ranking, 2025). And skip the gimmicks — in one Ahrefs study of 137,000 domains, 97% of llms.txt files received zero AI-bot traffic.
AI engines recommend businesses whose facts they can verify across multiple sources. The landmark generative-engine-optimization study (Aggarwal et al., KDD 2024) measured what actually moves AI visibility: adding quotations lifted it 41%, statistics 33%, and cited sources 28%. Concrete, verifiable specifics beat adjectives every time.
Consistency is the quiet half of corroboration. Your business name, address, phone number, hours, services, and pricing claims should match exactly across your website, Google Business Profile, directories, and social profiles. Conflicting facts give an AI engine a reason to drop your business and recommend a competitor it can verify.
Give other sites something quotable. Publish one specific number ("we answer service calls within two hours"), a named customer quote, or a local data point, and repeat it everywhere your business appears. When several independent sources echo the same fact, ChatGPT treats that fact — and your business — as reliable.
Most advice on this subject is written for software companies and then applied to everyone else. It does not transfer. We run a measurement engine that asks web-grounded AI assistants the questions customers actually ask, many times over, and records every source each answer draws on. Below is what came back, by industry. The pattern is not the same in any two of them.
| Industry | Directories | Competitors' own sites | Editorial & forums | Sample |
|---|---|---|---|---|
| Landscaping | 48% | 35% | 17% | 285 citations |
| Funeral homes | 30% | 67% | 3% | 4,903 citations · 655 runs |
| Forestry mulching | 24% | 74% | 2% | 404 citations |
| Land clearing | 17% | 80% | 3% | 981 citations · 160 runs |
| Marketing agencies | 1% | 82% | 17% | 2,304 citations · 405 runs |
Read the first and last rows together. A landscaper gets nearly half their citations from directories; a marketing agency gets one percent. Advice that says “get listed everywhere” is close to useless for the agency and close to essential for the landscaper. There is no general answer to this question — only an answer for your industry.
The second column is the one most people miss. In four of those five industries, competing businesses' own websites are the single largest source of citations. AI engines are not only reading directories and roundups; they are reading your competitors' service pages and quoting them. That is why the website work below matters as much as the listings work.
Two shifts, both recent enough that most published advice predates them.
ChatGPT started reading your own site directly. Around 8 August 2026, ChatGPT began running site-scoped background searches against the business's own domain during local queries. Measured across 1,487 matched queries between 30 July and 20 August, the share of local citations coming from businesses' own websites rose from 15.9% to 42.8%. Your own pages went from a minor source to the largest single one in three weeks.
The structured business card is a separate system from citations. In a study of 2,879 product-surface runs across 12 categories and 12 metros (21 August 2026), Yelp supplied the structured business card ChatGPT displays in 95.83% of runs — while appearing as a visible citation only 1.04% of the time. Both numbers are correct and they measure different things. It means your Yelp record can be doing most of the work of getting you named while never showing up as a source. If you are measuring only visible citations, you will conclude Yelp does not matter. It matters enormously; it is simply invisible in that column.
This is a gate, not an optimisation. Nothing below it works if it fails. In an analysis of more than 500 million crawler fetches, the crawlers used by OpenAI, Anthropic, Meta, ByteDance and Perplexity executed no JavaScript at all. Gemini and Applebot do render. Everything else reads your raw HTML and nothing more.
Test it yourself in one line, and read what actually comes back rather than what you expect:
curl -A "Mozilla/5.0 (compatible; OAI-SearchBot/1.0; +https://openai.com/searchbot)" \ https://yourdomain.com/your-service-page | sed 's/<[^>]*>//g' | tr -s ' \n' ' ' | wc -c
If that returns a few hundred characters, your service page is invisible to most AI crawlers. One caution learned the hard way: check the HTTP status too. A bot-protection challenge page also returns almost no text, and it looks identical to an empty page if you only count characters. A 403 means you were blocked, which is a completely different problem from a page that renders empty.
Of everything studied, this has the strongest evidence behind it. A factorial experiment across 252,000 paired trials and six models, with brand names anonymised and source order counterbalanced, found that including explicit price information had a consistent, significant effect on whether a business was recommended. The same study found formatting-only changes had little impact.
An honest note on that source: it is an unreviewed preprint (arXiv 2605.25517), not a peer-reviewed paper, and its authors disclose an internal pilot. We cite it because the design is unusually rigorous, not because it has been refereed.
The practical objection is real: many service businesses cannot quote without seeing the property. The answer is a published range plus the reason, not silence. “Typically $1,200 to $3,500 an acre depending on how heavy the growth is, exact price after a site visit” is a price for the purposes of an AI answer, and it is completely honest. A page that says only “contact us for a quote” gives the engine nothing to repeat.
One trap: a JavaScript price calculator publishes no price into the HTML. It computes in the browser, so a crawler sees an empty widget. The number has to be in the text.
Across 1.4 million prompts, the strongest single predictor of whether a page was cited was how closely its title matched the sub-query the engine generated — cosine similarity of 0.656 for cited pages versus 0.484 for uncited ones.
In practice this means Forestry Mulching in Louisa County, VA | Business Name rather
than Services or Home. It is the cheapest change on this page and one of
the most reliable.
Not a list of 150. The three to ten that appear in answers for your category — which, as the table above shows, are different in every industry. Claim the record, complete every field, and make the name, address and phone match your website exactly.
Across 4,216 AI-recommended businesses with verified ratings, the correlation between citation frequency and star rating was r = 0.013 — effectively zero. Having reviews makes you eligible to be recommended. Having more of them, past that point, does not appear to buy additional recommendations.
Ask for them lawfully. The FTC's rule on consumer reviews (16 CFR 465, effective October 2024) prohibits incentivising reviews conditioned on sentiment, suppressing negative reviews, and insider reviews without disclosure. And note that Yelp prohibits soliciting reviews at all — asking can get them filtered.
Things widely recommended that the evidence does not support:
Ask the same question twice and you will get two different answers. Across 2,961 runs, the odds that two runs of an identical prompt returned the same list of brands were worse than 1 in 100. A single check proves nothing in either direction.
So: ask each question at least eight times, per engine, and report each engine separately. Record whether you were cited (your domain appears as a source) separately from whether you were mentioned (your name appears in the text) — they are different outcomes and most tools blend them. And when a run fails, discard it rather than counting it as “not recommended,” which silently understates every figure.
You do not need all three levers finished before results appear. Start by asking ChatGPT ten questions a real customer would ask — "best [your service] in [your town]," "who should I hire to fix X" — and note which businesses get named and which sources get cited. Those cited sources are your to-do list.
Who we are: ASN Intelligence runs AI-visibility campaigns for small businesses, and asncheck.com is our education hub. If you want the baseline without the manual checking, the free scan at asncheck.com shows whether ChatGPT, Perplexity, and Google AI currently mention your business.
How long does it take to show up in ChatGPT recommendations?
Expect weeks to months, not days. Simple fixes — a corrected directory profile or a server-rendered page — can be picked up on an engine's next crawl, while building the review volume and third-party mentions that change recommendations is cumulative work. Ask ChatGPT the same customer questions monthly and track which businesses get named.
Can I pay ChatGPT or OpenAI to recommend my business?
No. ChatGPT recommendations are not advertising placements, and nobody can guarantee your business a mention. Recommendations come from patterns in training data and from live web sources the model retrieves while answering. The reliable path is earning presence in reviews, directories, and communities, plus a machine-readable website — not buying placement.
Does the same work help with Perplexity, Copilot, and Google AI Overviews?
Yes. Every major AI engine relies on the same fundamentals — trusted third-party sources, readable server-rendered pages, and corroborated, consistent facts — though each engine favors slightly different sources. Work done to earn ChatGPT recommendations generally improves visibility in Perplexity, Microsoft Copilot, and Google AI Overviews at the same time.