Why doesn't ChatGPT mention my business when people ask for recommendations?
Most often ChatGPT doesn't mention your business for one of three reasons: its crawlers can't access your site, your brand isn't present in the third-party sources the model leans on (listings, reviews, comparison articles, forums), or your pages aren't written in a form that's easy to quote. Ranking well in Google helps but doesn't guarantee anything – AI answers are assembled from a different mix of sources and a much smaller set of citations. The fix is a sequence: make yourself crawlable, make yourself quotable, then make yourself mentioned elsewhere.
How does ChatGPT decide which businesses to mention?
There are two separate mechanisms, and confusing them causes most of the wasted effort.
The first is the model's internal knowledge, formed during training on a large text corpus with a cutoff date. If your company was widely written about before that cutoff, the model may recall it without browsing. You cannot edit this directly or quickly.
The second is retrieval: when the question looks current or local, ChatGPT searches the web, opens a handful of pages, and builds the answer from them with links. This is the part you can influence – it behaves like a search layer with a very short shortlist, usually a few sources per answer rather than ten blue links.
In practice, recommendation-style prompts ("best X in Y", "which tool should I use for Z") trigger retrieval plus synthesis: the model gathers candidate names from lists and reviews, then checks individual sites. If your brand never appears in the gathering stage, your website quality is irrelevant – you're not in the candidate pool.
Can AI crawlers actually reach my site?
Start here, because it's binary and fast to check. Several AI companies use named user agents, and many sites block them by default – through robots.txt, a CDN bot-fighting rule, or a security plugin someone enabled last year.
Key agents to know:
| User agent | Operator | What it does |
|---|---|---|
| GPTBot | OpenAI | Crawls content, primarily for model training |
| OAI-SearchBot | OpenAI | Crawls for ChatGPT's search index |
| ChatGPT-User | OpenAI | Fetches a page when a user or action requires it live |
| PerplexityBot | Perplexity | Builds Perplexity's search index |
| ClaudeBot | Anthropic | Crawls for Claude |
| Google-Extended | Controls use of content for Gemini, not Google Search indexing | |
| Bingbot | Microsoft | Powers Bing and, through it, Copilot |
A permissive robots.txt for a business that wants AI visibility:
User-agent: GPTBot
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: ClaudeBot
Allow: /
User-agent: Google-Extended
Allow: /
User-agent: *
Allow: /
Sitemap: https://example.com/sitemap.xmlTwo caveats. Blocking GPTBot limits training use but doesn't necessarily remove you from answers that cite pages found through other indexes. And robots.txt is a request – enforcement varies, and a CDN rule or WAF challenge page can block a crawler even when robots.txt allows it. Check your server logs for these user agents; zero hits over several weeks is a strong signal that something upstream is blocking them.
Why do I rank in Google but never get cited by AI?
Classic search rewards pages that match a query well enough to deserve a click. AI answers reward passages that can be lifted and restated as fact, from sources the system treats as reliable on that topic.
Three common mismatches:
- Your page answers the query only after the reader scrolls. A landing page with a hero slogan, a carousel and testimonials before any substance gives a model nothing concrete to extract in the first screens of text.
- Key facts live in images, PDFs or JavaScript widgets. Price tables rendered client-side, service areas shown only on a map, hours inside a graphic – all invisible or unreliable to a crawler that reads HTML.
- Your claims are unverifiable. "Leading provider in the region" is marketing language. "Operating in Kraków and Katowice since [check: year], 40+ staff, on-site response within 4 hours" is extractable.
Also note that AI answer engines often prefer a third-party list over a vendor's own site when the question is comparative. A query like "best accounting software for freelancers" pulls from review sites and roundups, because the model treats a vendor page about itself as biased.
Which sources do AI answers lean on for recommendations?
Where the answer comes from depends on question type:
- Local service questions – map listings, directory profiles, review platforms, local media.
- Software and B2B tools – review marketplaces, comparison articles, documentation, Reddit and niche forums, YouTube transcripts.
- Product purchases – retailer pages, spec databases, "best of" roundups.
- Professional services – association registries, industry directories, case-study coverage, LinkedIn-style profiles.
The practical implication: visibility in AI answers is partly an off-site job. If every roundup of your category lists eight competitors and not you, the model has no reason to produce your name. Getting added to those lists – through outreach, accurate directory profiles, contributed expert commentary, or simply asking a reviewer to re-test – often moves the needle faster than rewriting your homepage.
Audit it yourself: search your category the way a customer would, open the top ten results that are lists or comparisons, and record whether you appear. That sheet is your off-site backlog.
How do I test whether ChatGPT knows my business?
Run a structured test rather than one lucky prompt.
- Write 10–20 prompts a real buyer would type: category + city, problem statements ("my warehouse scanner keeps losing Wi-Fi, who can fix it"), comparisons ("X vs Y vs alternatives"), and one direct brand prompt ("what is [brand] and what do they do").
- Run each in ChatGPT with web search on, then with it off, and note the difference. Answers without browsing reveal what the model "remembers"; answers with browsing reveal who wins retrieval today.
- Repeat in Perplexity, Gemini, Copilot and Google AI Overviews. Coverage differs noticeably between them – being invisible in one and cited in another is normal.
- For each answer, log which domains were cited. Those domains are your competition for citation space.
- Note factual errors about your brand. Wrong pricing, a closed office, an old product name – these come from stale sources you can usually correct.
- Re-run the same prompt set monthly. Answers vary between sessions, so track trends, not single outputs.
What on-site changes make a page quotable?
Write the answer first, then the context. Each important page should state what you do, for whom, where, at what price range and with what constraints – in plain sentences, high on the page.
Add a short factual block that a model can lift verbatim: services list, coverage area, response times, certifications, languages, payment methods. Keep it in HTML text, not an image.
Mark up the facts so machines don't have to guess:
{
"@context": "https://schema.org",
"@type": "LocalBusiness",
"name": "Example Repair Co.",
"url": "https://example.com",
"telephone": "+48 000 000 000",
"address": {
"@type": "PostalAddress",
"streetAddress": "Main St 1",
"addressLocality": "Kraków",
"postalCode": "30-001",
"addressCountry": "PL"
},
"areaServed": ["Kraków", "Katowice"],
"priceRange": "$",
"openingHours": "Mo-Fr 09:00-18:00",
"sameAs": [
"https://www.linkedin.com/company/example",
"https://g.page/example"
]
}Add an FAQ block with the questions prospects actually ask, marked up as FAQPage:
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "Do you work with small businesses?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Yes. Typical clients have 5-50 employees, and the minimum engagement is one month."
}
}]
}Structured data doesn't force a citation, and support for specific types varies by engine, but it removes ambiguity about entities, locations and offers.
Finally, publish an llms.txt at your root as a plain-language map of what matters. Adoption is still uneven – treat it as low-cost insurance rather than a guaranteed channel:
# Example Repair Co.
> On-site IT hardware repair for small businesses in Kraków and Katowice.
## Key pages
- [Services and prices](https://example.com/services): full price list, response times
- [Coverage area](https://example.com/areas): districts served, travel fees
- [FAQ](https://example.com/faq): warranty, turnaround, payment
## Facts
- Founded: [check: year]
- Team: [check: headcount]
- Languages: Polish, EnglishAlso make sure nothing on the page tells engines to stay away:
<meta name="robots" content="index, follow, max-snippet:-1, max-image-preview:large">A max-snippet limit or a stray noarchive can shrink what an engine is willing to reproduce.
What should I fix first?
| Priority | Problem | Fix | Effort |
|---|---|---|---|
| 1 | AI crawlers blocked by robots.txt, CDN or WAF | Allow named agents, verify in server logs | Low |
| 2 | Key facts hidden in JS, images or PDFs | Move to server-rendered HTML text | Medium |
| 3 | Brand absent from category roundups and directories | Outreach, claim and complete profiles | High |
| 4 | Pages lead with slogans, not answers | Rewrite intros as direct answers | Medium |
| 5 | No structured data | Add Organization/LocalBusiness, Product, FAQPage | Low |
| 6 | Conflicting NAP and outdated facts across the web | Correct listings, update stale pages | Medium |
| 7 | No llms.txt | Publish a short source map | Low |
How long before ChatGPT starts mentioning me?
Changes reach retrieval-based answers as fast as the underlying indexes refresh – that can be days to weeks for a well-crawled site, longer for a new domain. Changes to the model's internal knowledge only arrive with future training rounds, which you can't schedule.
Off-site work has a lag of its own: a new roundup mention has to be published, crawled and indexed before it can influence an answer. Plan in quarters and measure with a fixed prompt set, because a single re-prompt on the same day can give a different result for reasons unrelated to your site.
FAQ
Does blocking GPTBot remove me from ChatGPT answers?
Not necessarily. GPTBot is associated mainly with training; ChatGPT's browsing uses other agents and external search indexes. Blocking GPTBot reduces the chance of being recalled from training data, while a blocked search-oriented agent or a CDN rule is more likely to be what keeps you out of cited answers.
Will more backlinks make AI mention me?
Links help indirectly by improving crawling and classic rankings, but AI answers respond more to being named in relevant, retrievable text. A sentence listing your brand inside a well-ranked comparison article is usually worth more than a link from an unrelated site.
Why does ChatGPT give wrong facts about my company?
It's usually repeating an outdated or incorrect source – an old directory entry, a stale press article, a competitor's comparison page. Find the sources through the citations in the answer, request corrections where possible, and publish a clear, current facts page so newer crawls have something better to use.
Should I create pages specifically for AI?
No. Create pages that answer buyer questions precisely, then make them machine-readable. Thin pages written only to be quoted tend to perform poorly in both classic search and AI retrieval.
Why do Perplexity and Gemini cite me but ChatGPT doesn't?
Each service uses its own index, crawler permissions and selection logic, so coverage genuinely differs. Treat each engine as a separate channel, check them separately, and fix the blockers specific to the one where you're absent.
You can see where you stand before changing anything: run a free check of your site in Згадка at /en/audit to get GEO and SEO scores, a list of fixes with ready-made robots.txt, llms.txt, schema.org and FAQ files, and tracking of whether AI answers cite you.
