Why Isn't My Website Showing Up in ChatGPT? 7 Reasons and the Fix for Each

You've asked ChatGPT about your own business and it either drew a blank, described a competitor, or said something about you that stopped being true in 2023. More than 900 million people a week now use it, so a website not showing up in ChatGPT is a real commercial problem rather than a curiosity.

The good news is that it's a diagnosable one. ChatGPT has two ways of knowing a business exists: a live search crawler that fetches pages when someone asks, and whatever the underlying model absorbed about you during training. Most cases, whether they arrive as "why isn't my website showing up in ChatGPT", "why doesn't ChatGPT know my company" or "AI can't see my website", come down to one of those two channels being cut, and each channel fails for a short list of reasons. Below are the seven we check, in the order that finds the fault fastest.

The Short Answer: Seven Reasons a Website Isn't Showing Up in ChatGPT

1. OAI-SearchBot is blocked in your robots.txt. That's the crawler OpenAI uses for ChatGPT search. Block it and you've opted out of the live channel.
2. Your host or CDN refuses the crawler before robots.txt is read. Cloudflare has blocked AI crawlers by default on new domains since 1 July 2025.
3. Your site isn't in Bing's index. OpenAI names Microsoft as a search provider for ChatGPT search.
4. Nobody else on the web mentions you. Models learn brands from mentions, not from your homepage.
5. Your site never says plainly what you are: no Organization schema, no consistent name and address, no discovery files.
6. Your content is stale, or ChatGPT's memory of you is. AI assistants prefer fresh pages, and a model trained before your rebrand still remembers the old one.
7. You judged it on one prompt. Research shows the same question rarely returns the same list twice.

Check What ChatGPT Actually Says About Your Business First

Before touching anything, establish which of three states you're in, because they have different fixes. Ask ChatGPT, with search enabled, "what does [your business name] in [your town] do?" and then "recommend a [your trade] in [your town]".

• Not found at all. "I couldn't find a business by that name" or "no official website found". Start at reason 1 and work down.
• Found, but wrong. Old address, discontinued service, a price from three years ago. Skip to reason 6, then reason 5.
• Found, but never recommended. It knows you exist and picks someone else when asked for a recommendation. That's reasons 4 and 5, and it takes longest to fix.

If you'd rather not run the prompts yourself, the Live ChatGPT Snapshot on our sister site does it for you and shows the raw response, including how ChatGPT describes your competitors when asked the same thing.

Reason 1: OAI-SearchBot Is Blocked in Your robots.txt

OpenAI runs three separate user agents, and people block the wrong one all the time. Here is what each does, straight from OpenAI's crawler documentation:

OpenAI's three user agents and what blocking each one costs you

User agent	What it does	Blocking it means
OAI-SearchBot	Crawls to surface sites in ChatGPT search results	You disappear from ChatGPT search answers
ChatGPT-User	Fetches a page when a user asks ChatGPT about it	ChatGPT can't read a page a user points it at
GPTBot	Crawls content that may be used to train models	Nothing, as far as ChatGPT search is concerned

The one that governs whether you appear in ChatGPT search is OAI-SearchBot, and OpenAI is unambiguous about it:

OAI-SearchBot is for search. OAI-SearchBot is used to surface websites in search results in ChatGPT's search features. [...] it can take ~24 hours from a site's robots.txt update for our systems to adjust.

OpenAI, Overview of OpenAI Crawlers

Two things in that passage matter. First, the search crawler and the training crawler are different agents with different names, so you can refuse GPTBot (if you object to training use) and still allow OAI-SearchBot without contradiction. A lot of the "block AI" robots.txt templates that circulated in 2024 and 2025 disallow every OpenAI agent in one block, and the site owner never notices they've switched off ChatGPT search along with training. Second, the 24-hour figure is OpenAI's own, so if you fix robots.txt today and nothing changes by tomorrow afternoon, the fault is elsewhere on this list.

A robots.txt that allows the search and retrieval agents while leaving training as your own choice looks like this:

# Allow ChatGPT search and user-requested page fetches
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

# Training crawler: your call. Blocking it does NOT affect ChatGPT search.
User-agent: GPTBot
Disallow: /

Paste your domain into our free Robots.txt Checker if you want the file parsed for you, or use the AI Crawler Checker, which reads the robots.txt rule for each of 14 AI crawlers and then tests live access as well, which brings us to reason 2.

Reason 2: Your Host or CDN Refuses the Crawler Before robots.txt Is Read

robots.txt is a request. A web application firewall is a refusal. If your host, CDN or security plugin returns a 403 or 429 to the crawler's user agent, the crawler never reads your robots.txt at all, and every robots.txt-only checker on the internet will tell you the site is wide open.

This has become the common case rather than the edge case. Cloudflare announced on 1 July 2025 that every new domain on its network starts with AI crawlers blocked, so that, in its words, "every new domain starts with the default of control, and eliminates the need for webpage owners to manually configure their settings to opt out" (Cloudflare press release). Cloudflare lets AI companies declare whether a crawler is for training, inference or search, and lets you allow them by category, but the default is off and a great many site owners never open that panel.

We measured a starker version of this on 18 August 2026. WordPress.com returned 403 to GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User and Applebot, and 429 to Meta-ExternalAgent, Bytespider and CCBot, across eight separate *.wordpress.com hostnames, while a normal browser request got 200 either side of every probe. Its robots.txt names none of those agents. A neutral, unnamed test bot walked straight in, which is what proves the rule is aimed at AI crawlers specifically rather than at bots in general. We wrote up the wider pattern in Is Your WordPress Host Quietly Blocking AI Bots?

The uncomfortable finding from the same session is that a host's own marketing site tells you nothing about its customer hosting. Of three real customer sites on the same platform, two allowed everything and a third refused two crawlers. Blocking is configured per package, not per host, and we have had to ask our own platform provider to unblock a crawler on a specific customer site more than once. The only test that means anything is a test of your URL.

You can run that test in two lines. The first asks your site the way ChatGPT search does; the second asks it the way a browser does. The status codes should match.

curl -sI -A "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; OAI-SearchBot/1.0; +https://openai.com/searchbot" https://www.example.co.uk/ | head -1
curl -sI -A "Mozilla/5.0 (Windows NT 10.0; Win64; x64) Chrome/128.0 Safari/537.36" https://www.example.co.uk/ | head -1

Two 200s and you're fine at this layer. A 403 or 429 on the first line only, and something between the crawler and your PHP is making the decision. One caveat from our own testing: the answer can vary by source IP, because some firewalls score datacentre ranges more harshly than a home connection, and real crawlers come from datacentres. A clean result from your laptop is the gentle version of the test.

Reason 3: Your Site Isn't in Bing's Index, and ChatGPT Search Leans on Microsoft

OpenAI's own crawler is only part of how ChatGPT search finds pages. The launch announcement said ChatGPT search draws on "third-party search providers, as well as content provided directly by our partners", and OpenAI's help article on ChatGPT search says it "sometimes partners with other search providers". The only provider whose privacy policy that article links to is Microsoft's.

Read that plainly: a site Bing cannot find is a site ChatGPT search is working to locate with one hand tied. Most UK businesses have never opened Bing Webmaster Tools because Bing's share of UK searches is small, and they've never noticed that the index behind Copilot and behind part of ChatGPT search is the same one. Two fixes, both free:

• Verify the site in Bing Webmaster Tools, submit your sitemap, and check that your priority pages actually show as indexed. If you already use Google Search Console, Bing will import the verification.
• Turn on IndexNow, which pings Bing the moment a page changes rather than waiting for a crawl. Most SEO plugins have a one-click switch for it.

Do a quick site:yourdomain.co.uk search on Bing before and after. A site showing 3 pages when it has 40 has found its problem.

Reason 4: Nobody Else on the Web Mentions Your Business

This is the reason behind "found, but never recommended", the one people describe as "ChatGPT ignores my site", and it's the one no technical fix touches. Rand Fishkin of SparkToro put it in one sentence back in 2024 and nothing since has contradicted him:

The currency of large language models is not links. The currency of large language models is mentions.

Rand Fishkin, SparkToro, How Can My Brand Appear in Answers from ChatGPT, Perplexity, Gemini, and Other AI/LLM Tools?

What that means in practice is that the model learned which businesses go with "emergency locksmith Wolverhampton" from the places those words sit next to each other on the open web: directories, reviews, local news, trade bodies, forum threads. Your own site says it once. The web needs to say it many times, in crawlable public text, for the association to form. It also explains a question we get from agencies who rank well: "we're number one on Google for our main keywords, so why doesn't ChatGPT use us as a source?" Because ranking and mentions are different signals. Ahrefs analysed 15,000 prompts in August 2025 and found that only 12% of the links cited by ChatGPT, Gemini and Copilot appear in Google's top ten for the same prompt; for ChatGPT's in-text citations the overlap was 8%. Your Google position is not a ticket in.

The counter-example is worth more than the theory. In April 2026 we launched a brand-new domain for a Brewood locksmith, Lockerfella, on our standard Linux hosting with zero backlinks and zero reviews on the site at go-live. By day ten to twelve it was number one in ChatGPT and in Gemini for "Brewood locksmith", and ChatGPT was quoting real prices and named recent jobs from the site verbatim. What carried it was not off-site authority, because there wasn't any yet. It was 18 area pages each written from the owner's real work diary, with named jobs and prices from £30 to £249, full schema, and AI discovery files at launch. The Lockerfella case study has the numbers. Specific, verifiable, first-hand content is the one form of "mention" you fully control, and a model reads it the same way it reads a directory listing: as words that now sit next to your name.

Reason 5: Your Website Never Says Plainly What Your Business Is

A page that reads beautifully to a person can be an unlabelled pile of nouns to a crawler. If nothing on the site states, in machine-readable form, that "365i is a web hosting company in Kettering, founded 2002, trading as BSolve IT Limited", then an AI system assembling an answer has to infer it. Inference is where "found, but wrong" comes from, and often where "ChatGPT doesn't know my business" comes from too: it has met your words without ever meeting your entity.

Three things close that gap, in ascending order of effort. An Organization or LocalBusiness schema block on every page with the same name, address and phone number the rest of the web uses for you. An about page that says who you are in the first paragraph rather than the fourth. And a set of AI discovery files: llms.txt, brand.txt, ai.json and the rest, which describe the business to AI systems in the format they read most easily.

The obvious objection is "do AI crawlers even fetch those files?", and we hold the logs to answer it. On a site hosted on our own platform, with the files generated by our own free WordPress plugin, the first eleven days after install showed 15 file reads and none of them from OpenAI. By the nineteenth day the picture had changed: 53 reads from 8 distinct AI bots across 5 operators, including ChatGPT-User 11 times, OAI-SearchBot 6 and GPTBot 6, every one of them a 200. In the 30-day window to 15 July 2026, OAI-SearchBot's single most-requested file on that site was /brand.txt, and brand.txt and faq-ai.txt were read more often than llms.txt itself. The honest limitation: llm.txt, the singular variant, was never fetched by anyone, and crawler behaviour on day one tells you nothing about day forty.

Reason 6: Your Content Is Stale, or ChatGPT's Memory of You Is

Two different faults hide under "ChatGPT gives wrong information about my business", and they need separating.

The live channel prefers fresh pages. Ahrefs' study of just under 17 million cited URLs (July 2025) found that AI assistants cite pages 25.7% fresher on average than the pages in Google's organic results, and that ChatGPT has the strongest preference of all: the URLs it cites are 458 days newer than the organic results for the same query. A services page last edited in 2022, with no visible date, is competing against pages updated last month, on a surface that is measured to care.

The model's memory is a snapshot. Everything the underlying model "knows" about you without searching was fixed at its training cutoff. If you rebranded, moved, dropped a product line or changed prices after that date, and the search channel doesn't fire for the question being asked, ChatGPT answers from the snapshot. That is where "hallucinated discontinued features" comes from: the feature existed, on your site, at the time the model read it. You can't edit a trained model. What you can do is make the live channel win: keep the current page crawlable (reasons 1 and 2), keep it indexed (reason 3), keep it dated and updated, and state the change plainly somewhere a crawler will read it, including an update log on the page itself.

We've been on the receiving end of the milder version. In May 2026, ChatGPT told a prospect vetting us that 365i doesn't appear on Nominet's public registrar list. True as far as it goes: our .uk registrations route through our platform partner 20i under the STACK tag, which is a detail that had never been written down anywhere crawlable. The answer wasn't wrong so much as incomplete, and the fix was a page that says exactly how 365i registers your domain in plain words. The agency owner concerned moved 30-plus client domains to us once he'd heard it from a human, but he shouldn't have had to ask.

Reason 7: You Judged It on One Prompt

The last reason is the reader, and I include it because we've made the mistake ourselves. In January 2026 SparkToro and Gumshoe published the largest public test of AI recommendation consistency to date: 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times over November and December 2025. Their finding: "there's a <1 in 100 chance that ChatGPT or Google's AI, if asked 100X, will give you the same list of brands in any two responses", and for the same list in the same order it's nearer 1 in 1,000 (SparkToro, 28 January 2026).

So one prompt that misses you is not a verdict, and one prompt that names you is not a win. Ask the same question five or six ways, from a logged-out session, over a few days, and count how often you appear. If you're in the set a third of the time, you're visible and the work is reason 4. If you're never in it, work the list from the top.

What Fixing robots.txt Won't Do: The Limits

Everything above about crawlers is necessary and none of it is sufficient, and there's data that looks, on a fast read, like it contradicts reason 1. BuzzStream examined 4 million AI citations across 3,600 prompts and found that among the top 50 news sites, 82.4% of those blocking OAI-SearchBot and 88.2% of those blocking GPTBot were still being cited (BuzzStream, updated April 2026). The explanations offered are that the content was crawled before the block went in, that it sits in Common Crawl's archive from years of CCBot visits, and that not every crawler obeys robots.txt.

None of that helps a plumbing firm in Kettering. The Guardian is cited despite blocking because a million other pages describe what the Guardian published; it has reason 4 solved a thousand times over. A small business has no such cushion, so the crawler channel is most of what it has, and a block costs it far more than the same block costs a national newspaper. The other honest limits: there is no "submit your website to ChatGPT" form, no guarantee of a citation once every check passes, and no ranking to track in the way you'd track a Google position, for the reason given in reason 7. What you get for doing the work is eligibility: your site showing up in ChatGPT for a steadily rising share of the prompts that ought to name you.

How to Fix a Website Not Showing Up in ChatGPT, in Order

The seven checks in diagnostic order, with how to run each and how long before you'd expect to see a change

Check	How	Change within
1. OAI-SearchBot allowed in robots.txt	Robots.txt Checker, or read the file	About 24 hours (OpenAI's figure)
2. Host or CDN not refusing the crawler	The two-line curl test, or the AI Crawler Checker	Immediately once unblocked, then 24 hours
3. Indexed in Bing	Bing Webmaster Tools, sitemap, IndexNow	Days to a few weeks
4. Third-party mentions exist	Directories, reviews, local press, first-hand case pages	Months, and ongoing
5. Entity stated plainly	Organization schema, about page, AI discovery files	Weeks, as pages are re-crawled
6. Content current and dated	Update key pages, show dates, keep an update log	Weeks
7. Measured properly	Several prompts, several days, logged out	Now

If you host with us and the two-line test returns a 403 or 429 on your own package, raise a ticket with the output and we'll trace it. On the platform this site and our customers' sites run on, our own logs show OAI-SearchBot, ChatGPT-User and GPTBot all fetching with 200s, so a refusal is something specific to the package and can be found.

Frequently Asked Questions