How to get your business recommended by ChatGPT

To get your business recommended by ChatGPT, you need three things: consistent information about your business everywhere it appears online, pages that AI models can crawl and quote directly, and mentions on the third-party sources those models already trust. This article is the practical playbook: how ChatGPT actually chooses businesses, a self-audit you can run in ten minutes, and the specific fixes - entity consistency, citable pages, structured data, review surfaces, llms.txt, and crawler access - in the order they matter.

How ChatGPT decides which businesses to recommend

ChatGPT draws on two different systems, and each responds to different tactics. The first is training data: what the model absorbed about your business before its knowledge cutoff. You influence this slowly, through mentions on established sites, press, directories, and forums that get scraped over months and years. The second is live retrieval: when a question looks current or local, ChatGPT runs a web search and cites pages it fetches in real time. You influence this quickly, through crawlable, quotable pages and search visibility.

The bar is high. One large 2025 analysis of local businesses found ChatGPT recommended only around 1 percent of them, while roughly a third of the same businesses appeared in Google local results. Being findable on Google does not mean an AI assistant will ever say your name. The same research found business websites made up the majority of the sources ChatGPT drew on, with third-party brand mentions and directories covering most of the rest. That split is effectively your to-do list: fix your own site first, then fix what everyone else says about you.

Run this ten-minute AI visibility audit first

Before changing anything, find out where you stand. Open ChatGPT (with web search on), Perplexity, and Claude, and paste in the questions a real customer would ask:

• "Best [your service] in [your city or niche]" - are you named at all?
• "[Your business name] reviews" - is what comes back accurate, and which sources are cited?
• "Who should I hire for [the problem you solve]?" - who wins, and which pages get cited for them?
• "Is [your business name] legitimate?" - does the assistant know you exist and describe you correctly?

Record the answers and the cited sources in a spreadsheet. The cited sources matter more than the answers: they tell you exactly which websites the models trust for your category. Those are the places you need to appear. Repeat the same prompt set monthly - that spreadsheet becomes your measurement baseline.

Entity consistency: make your business one unambiguous thing

Language models build an internal picture of your business from every mention of it across the web. If your name, address, phone number, description, and category differ between your website, Google Business Profile, LinkedIn, directories, and social profiles, the model's picture is blurry - and blurry entities do not get recommended, because the model cannot be confident about what it would be recommending.

• Pick one canonical name, one-line description, and category, and use them verbatim everywhere.
• Audit every profile you control: Google Business Profile, LinkedIn, industry directories, social accounts, footer text on your own site.
• Fix contradictions, dead listings, and old addresses. Conflicting information is worse than missing information, because retrieval systems resolve conflicts by dropping the uncertain entity.
• Have a clear About page that states who you are, what you do, for whom, and since when, in plain sentences a model can lift wholesale.

Citable pages: write content an AI can quote

When ChatGPT retrieves live pages, it quotes passages that answer the question directly. Most business websites are written to persuade, not to answer, so there is nothing quotable in them. The fix is structural, not stylistic: put the answer in the first two sentences under a heading phrased the way people ask the question, then support it with specifics.

• Question-form headings: "How much does X cost?", "How long does Y take?" - the queries you actually want to win.
• Answer-first paragraphs: the direct answer immediately under the heading, details after. Models rarely quote paragraph four.
• Real numbers, ranges, and honest tradeoffs. Vague pages do not get cited because there is nothing in them worth citing.
• One topic per page. A single page trying to answer twelve questions gets cited for none of them.

This discipline is the core of generative engine optimization, and it overlaps heavily with good SEO rather than replacing it. If the concept is new, the primer on what generative engine optimization is covers the fundamentals.

Structured data and llms.txt: speak the machine formats

Structured data (JSON-LD schema) turns your pages into facts a machine can parse without guessing. At minimum, add Organization or LocalBusiness schema on your homepage with your canonical name, URL, logo, and contact details, Service or Product schema on offer pages, and FAQPage schema on pages with question-and-answer content. It takes a developer an afternoon and removes ambiguity from every retrieval.

llms.txt is an emerging convention: a plain-text file at your domain root that gives AI systems a curated map of your site - who you are, what you do, and links to your most important pages with one-line descriptions. Adoption by the model providers is still uneven, so treat it as cheap insurance rather than a ranking lever: an hour of work that makes your best pages easier to find, with no downside.

Then check crawler access, because everything above is worthless if AI crawlers are blocked. Review your robots.txt for rules affecting GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot. Many sites block these bots by default via a CDN, firewall setting, or an aggressive robots file someone added in 2023 - which quietly removes you from live AI retrieval entirely. Blocking training crawlers is a legitimate choice; doing it accidentally is not.

Get mentioned where AI models already look

Your own site is the biggest single source, but third-party mentions and reviews carry weight your site cannot: they are independent corroboration. Models weight sources that already rank well and get cited often - and your ten-minute audit told you exactly which ones those are for your category.

• Reviews on the surfaces models cite for your niche: Google reviews for local service, G2 or Capterra for software, Clutch for agencies, Trustpilot for e-commerce.
• Respond to negative reviews factually. Assistants summarize review sentiment; an unanswered pattern of complaints becomes the answer to "is this business any good?"
• Earn coverage on industry publications and communities where your buyers ask questions - one substantive mention on a trusted site outweighs fifty directory listings.
• Publish genuinely useful expertise under a named author. Quotable, specific writing is what gets picked up and repeated, which is how training-data visibility compounds.

Measure it, and set honest expectations

Re-run your prompt set monthly across ChatGPT, Perplexity, and Claude, and track three things: whether you are named, what is said about you, and which sources are cited. Also watch your server logs or analytics for AI crawler visits and for referral traffic from chatgpt.com and perplexity.ai - both are early signals that retrieval is picking you up.

Timelines: crawler-access and schema fixes can affect live retrieval within weeks. New citable content typically takes one to three months to start earning citations. Training-data presence moves on model-release cycles - many months. If someone promises you top ChatGPT recommendations in thirty days, they are selling the same snake oil the SEO industry has always sold. And if your category is one where nobody asks AI assistants for recommendations, this whole effort may not be worth much yet; check your own customer behavior before investing heavily.

Most of this playbook is execution: schema, crawler configuration, content restructuring, and monthly measurement. It is the kind of systems work Kaev does under AI and automation, but everything above is doable in-house by a motivated founder with a developer's help.

Common questions

How does ChatGPT choose which businesses to recommend?

Two ways: from its training data, built from mentions of your business across the web over time, and from live web retrieval, where it searches and cites pages in real time. Training presence responds to third-party mentions and reviews; live retrieval responds to crawlable, answer-first pages that rank in search.

How long does it take to show up in ChatGPT recommendations?

Crawler and schema fixes can influence live-search answers within weeks. New content usually takes one to three months to earn citations. Deep training-data presence follows model release cycles and takes many months. Anyone promising fast guaranteed results is overselling.

Does blocking GPTBot hurt my AI visibility?

Blocking GPTBot keeps your content out of future training data, and blocking OAI-SearchBot removes you from ChatGPT's live search citations. Many sites block these accidentally through CDN or firewall defaults. Check your robots.txt and bot-protection settings before assuming a content problem.

Is getting recommended by AI different from normal SEO?

It overlaps heavily. Good SEO still feeds the search results AI assistants retrieve from. The additions are entity consistency across every profile, answer-first page structure that models can quote, structured data, AI crawler access, and mentions on the third-party sources models cite.

Do I need an llms.txt file for my business website?

It is optional but cheap. llms.txt is a plain-text site map for AI systems at your domain root. Provider adoption is still uneven, so treat it as an hour of low-risk insurance, not a ranking lever. Fix crawler access and schema first.