Does ChatGPT Use My Website?
Short answer: sometimes, in two completely different ways, and the difference between them decides what you should do about it.
This question comes up constantly and it is usually asked with a mix of hope and suspicion. Hope, because being used by ChatGPT sounds like free visibility. Suspicion, because being used without permission sounds like something has been taken. Both reactions are reasonable and both miss what is actually happening.
This article explains the two routes into ChatGPT, how to check whether your site is reachable by either, what you can control, and what to do with the answer.
The Two Ways ChatGPT Can Reach Your Website
Almost all the confusion about this subject comes from treating these as one thing. They behave differently, they are controlled differently and they matter differently.
Training: what the model learned before it met you
Large language models are trained on very large quantities of text gathered from the web, among other sources. If your website existed and was publicly accessible when that gathering happened, some of it may have contributed to what the model knows.
Two things follow. It is historical, so it reflects your website as it was rather than as it is. And it is diffuse: the model has not stored your pages, it has absorbed patterns from an enormous corpus in which your site is a vanishingly small part. You cannot point at a sentence and say that came from you.
For most small and medium businesses, this route contributes very little. You are not famous enough for a model to have formed a reliable impression of you from training alone.
Retrieval: what it looks up while answering
This is the one that matters. When ChatGPT searches the web during a conversation, it retrieves current results, reads them and uses them to build the answer, citing sources as it goes.
It retrieves primarily through Bing. That single fact is the most actionable thing in this article, and it surprises most people, because fifteen years of habit points everyone at Google.
Retrieval is live, current and influenceable. If your site is indexed, reachable and readable, it can be used today, regardless of what any model learned two years ago.
Which One Should I Care About?
Retrieval, almost entirely.
Training is largely out of your hands, historical, and slow to change. Retrieval is happening right now, reflects your site as it stands, and responds to work you do this month. Every practical recommendation worth following concerns retrieval.
This also reframes the anxiety people bring to the question. The worry is usually “is my content being taken”. The more useful worry is “can it be reached at all”, because a site that cannot be retrieved is simply absent from the answers your buyers are reading.
How Do I Check Whether ChatGPT Can Reach My Site?
Four checks, none of them difficult, in the order worth doing them.
1. Check your robots.txt
Visit your own domain followed by /robots.txt. You are looking for any rule blocking these: GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended, CCBot.
Some site builders and security plugins add these blocks by default, and plenty of businesses have them without knowing. If you want to be found in AI answers, they should not be there.
2. Check you are actually in Bing
Search Bing for a distinctive phrase from your homepage in quotation marks. If nothing comes back, Bing does not have you properly indexed, and ChatGPT will struggle to retrieve you no matter how well you rank in Google.
Register with Bing Webmaster Tools and submit your sitemap. Twenty minutes, and it is the highest-return technical task on this page.
3. Check the crawlers are not being refused at the server
This is the one almost nobody checks and it catches people out. Your robots.txt can welcome every crawler while your firewall, security plugin or host quietly rejects requests that do not look like a human browser. AI crawlers send minimal headers, which is precisely what some security rules are configured to block.
Ask whoever manages your hosting to look in the server logs for GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot, and confirm they are receiving normal responses rather than being turned away. It takes them five minutes. We have found sites inviting crawlers in at one layer and blocking them at another, with nobody aware because nothing looks wrong from outside.
4. Ask it, with browsing on
In a logged-out session with web search enabled, ask a question your website should be able to answer, without naming your company. Then ask one that names you. If it can find you when prompted but never otherwise, you are reachable but not recommended, which is a different problem and a more common one.
Can I Stop ChatGPT Using My Website?
Largely, yes. Whether you should is a different question.
You can disallow the crawlers in robots.txt. Publishers with paid content and licensing deals have done exactly this, and for them it makes sense, because their content is the product.
For most businesses it is self-harm. You are not protecting an asset, you are removing yourself from the answers your buyers now read, and the question still gets answered using a competitor who stayed visible. You lose the citation and gain nothing.
The exception worth considering is content genuinely behind a paywall or a login, which should not be publicly crawlable anyway for reasons that have nothing to do with AI.
If My Content Is Used, Do I Get Anything Back?
Sometimes, and this is where expectations need managing.
When ChatGPT retrieves live results it cites sources with links, and people do click them. That traffic shows up in analytics and it is usually small in volume.
It is also, in our experience, remarkably good traffic. One of our clients is a UK law firm in a specialist field. Around 30 percent of the enquiries arriving from AI-referred visits convert, against 3.9 percent across their website traffic as a whole. Roughly eight times. Seventeen enquiries from about fifty-six visits in a month.
Fifty-six visits is invisible on a traffic chart. Seventeen enquiries is a good month. If you judge this channel on volume you will dismiss it. If you judge it on what arrives, you will not.
The honest caveat, since this article is about being precise: analytics currently groups ChatGPT, Perplexity, Gemini and Copilot into a single channel, so that figure covers AI referrals as a whole rather than ChatGPT alone.
What Makes ChatGPT More Likely to Use My Pages?
Reachability gets you eligible. Being used is a separate question, and five things move it.
- Bing visibility. Retrieval runs through it. Everything else is secondary if this is missing.
- Extractable answers. The question as a heading, the answer in the first two sentences underneath. Assistants quote passages, not pages, so give them a passage worth quoting.
- Specifics rather than claims. Numbers, timeframes, defined methods, named people. “Leading provider of bespoke solutions” cannot be quoted. “Most cases are resolved within eight weeks” can.
- Corroboration elsewhere. Information that appears consistently across your site, your listings and independent coverage is safer for an assistant to state than a claim existing in one place.
- Freshness. Recency influences citation. A site whose newest content is two years old reads as a business that may no longer be trading.
We go through the full picture in how to get recommended by ChatGPT, which covers the seven signals in more detail than there is room for here.
Does This Work the Same Way for Other Assistants?
The principles transfer. The plumbing does not.
| Assistant | Retrieves through | What that means for you |
| ChatGPT | Primarily Bing | Bing Webmaster Tools is not optional |
| Perplexity | Its own live index, always cites | Watch its citations closely, it shows its working |
| Google Gemini | Google index | Your existing Google work largely carries over |
| Google AI Overviews | Google index | Reaches the most buyers, so it matters most |
| Microsoft Copilot | Bing | The same work as ChatGPT covers this |
The practical consequence is that Bing covers two of the five and Google covers two more, so the technical groundwork is smaller than the list of assistants suggests.
What Should I Do This Week?
Four things, about an hour and a half in total, and they cover most of what matters.
- Read your robots.txt and remove any AI crawler blocks you did not intend.
- Register with Bing Webmaster Tools and submit your sitemap.
- Ask your host to confirm the AI crawlers are getting normal responses in the server logs.
- Pick your three most important pages and move the answer to the top of each, under a heading phrased as the question.
Then leave it a month before judging anything. If you want a fuller diagnosis first, why is my business not showing up in AI search works through the eight usual causes in order.
What About My PDFs, Images and Downloads?
Worth knowing, because a great deal of the most useful material on business websites sits in formats that get overlooked.
PDFs
These can be crawled and used, and they are frequently where the genuinely specific content lives: price lists, specification sheets, guides, technical documents. If a PDF is publicly linked and readable, it is retrievable. If it is a scanned image with no text layer, it is not, and a surprising number of business documents fall into that second group.
Two practical points. Give PDFs descriptive filenames rather than a reference number, and link to them from a page with context around them, because a document sitting on its own with no explanatory page is much harder to interpret.
Images and video
The file itself carries almost no useful meaning to a text-based assistant. What it can read is the alt text, the caption, the surrounding copy and any transcript. If your best explanation of something is inside a video with no transcript, it is invisible for these purposes, and adding one is usually a small job with a disproportionate effect.
Anything behind a form
Gated content is not retrievable. That is the trade you make when you put something behind a form, and it is a real trade now rather than a neutral one. It is worth asking whether your best guide is earning more as a lead magnet than it would as a widely cited public asset. For some businesses the answer has changed.
Is Any of This Legal, and Should I Worry About Copyright?
This question comes up almost every time and it deserves a straight answer rather than a shrug.
The law here is genuinely unsettled. There is ongoing litigation internationally about training data, and the UK position on text and data mining has been consulted on more than once without a settled outcome. Anybody telling you the position is clear is overstating it in one direction or the other.
What is reasonably clear in practice: live retrieval with attribution behaves much like a search engine reading your page and citing it, which is the arrangement most businesses have been comfortable with for twenty-five years. Training is the contested area, and it is also the part you have least control over.
Our practical advice for a typical business, and it is advice rather than a legal opinion: if your content is marketing, being read and cited is the point, and blocking it costs you more than it protects. If your content is the product, take proper advice and consider licensing rather than simply blocking. If you are unsure which you are, ask whether you would be pleased or annoyed to find a paragraph of it quoted with your name attached.
How Long Until Changes Take Effect?
Faster than SEO in some respects and slower in others, which catches people out.
- Crawler access fixes can take effect within days. Once a crawler can reach a page, it can use it.
- Bing indexing typically takes one to four weeks after submitting a sitemap, sometimes faster for an established site.
- Structure and specificity edits tend to show over four to eight weeks, as pages are re-fetched and re-evaluated.
- Corroboration and authority takes six months and beyond, because it depends on things being published elsewhere that you do not control the timing of.
- Training data changes only when models are retrained, which is not a timeline you can plan around.
The uneven pace is why we sequence work the way we do. Do the fast mechanical things first, so there is visible movement while the slower work compounds quietly underneath.
How We Handle This at TU
We check reachability before anything else, because there is no point improving content that cannot be fetched. Robots rules, Bing indexing, server-level crawler access, and whether the important pages are actually readable rather than assembled by script in a way that leaves little for a crawler to read.
That last one catches a lot of modern websites. A page that looks rich to a person can be close to empty to a crawler, and the business has no idea because it looks perfect on screen.
Then we work on extractability and specifics, which is mostly editing rather than writing, and it is the fastest content work there is. New material and third-party coverage come afterwards, because they compound more slowly and they perform far better once the foundations are sound.
We run our own website the same way, with an owner, a monthly report and a number to hit, and we publish what we learn including the parts we are still testing. It is worth asking anyone selling you AI visibility to show you their own results first.
If you would like to know whether ChatGPT can currently reach and use your site, we will check it as part of an AI visibility snapshot. You keep the findings whether or not you work with us.
Frequently Asked Questions
Does ChatGPT use Google or Bing?
Bing. When ChatGPT searches the live web it retrieves primarily through Bing’s index, which is why registering with Bing Webmaster Tools is one of the fastest wins in AI visibility. Google still matters, but for Gemini and AI Overviews rather than for ChatGPT.
Was my website used to train ChatGPT?
Possibly, if it was publicly accessible when training data was gathered. There is no way to confirm it for an individual site, and for most businesses it matters far less than live retrieval, which is current and influenceable.
How do I block AI crawlers if I want to?
Add disallow rules for the named crawlers in robots.txt. For most businesses we would advise against it, because you remove yourself from answers your buyers read while the question still gets answered using a competitor.
If I block them, will my content stop appearing?
New retrieval stops. Anything already absorbed in training is not retrievable or removable on request. Blocking is forward-looking only.
Does ChatGPT send me any traffic?
Yes, when it cites sources during live retrieval. Volumes are usually modest. Conversion rates are frequently far above site average, which is why the channel is worth more than its visit count suggests.
Why can ChatGPT describe my competitor but not me?
Usually because more independent information exists about them. If everything about your business originates from your own website, there is nothing to corroborate against, and these systems are cautious about naming a business on its own say-so.
Do I need to do anything different for ChatGPT versus Google?
Some of it. The Bing work is specific to ChatGPT and Copilot. The rest, extractable structure, specific claims, consistent entity signals and independent coverage, benefits everything, including conventional Google rankings.
My site is built in a modern framework. Does that matter?
It can. If the content is assembled in the browser rather than served in the page, some crawlers will see very little. Ask your developer whether the important content is present in the initial response, and test one page with scripting disabled to see what remains.
