How to use the AI Referrer Checker
The checker takes a referrer or a landing-page URL and tells you which AI assistant, if any, sent the visit. It runs in your browser and does not contact the site you paste. Here is the flow.
- Find the value to check. Copy a referrer URL from your analytics, a server log or a CRM record. You can also paste a full landing-page URL that includes UTM parameters.
- Paste it into the box. Use the full value, such as
https://chatgpt.com/orhttps://example.com/pricing/?utm_source=chatgpt.com. Partial values like a bare hostname work too. - Click check. The tool compares the hostname and the
utm_sourcevalue against a list of known AI assistants and AI search products. - Read the result. You get the assistant name, the signal that matched (referrer hostname or UTM parameter) and a note on how reliable that signal is.
- Use it to build a rule. If you are writing a filter for your analytics tool, copy the matching hostname or pattern and test it against a few more examples.
Tip: Check a handful of values from your real traffic, including ones you believe are direct or organic. You may be surprised which ones turn out to be AI visits.
What is an AI referrer?
When someone clicks a link on one web page and lands on another, their browser usually tells the destination site where they came from. That information is the HTTP Referer header (spelled with the historical typo), and it is also readable in JavaScript as document.referrer. Analytics tools use it to sort traffic into sources such as search, social and referral.
An AI referrer is simply a referrer that identifies an AI assistant or AI search product as the origin. When a person asks ChatGPT a question, reads the answer, and clicks a cited link, the browser can report that the visit came from ChatGPT’s website. The same goes for Perplexity, Gemini, Claude and others. Those visits are valuable, because they come from people who asked a specific question and chose your page from the answer.
Separately, some assistants add a tracking parameter to the links they produce, such as utm_source=chatgpt.com. That parameter travels in the URL of your landing page, so it survives in cases where the referrer header does not.
Why identifying AI traffic matters
AI search is a new channel, and most analytics setups bucket its visits under “referral”, “direct” or even “organic”. Without a deliberate step to separate them, you cannot answer basic questions:
- Which AI assistants send you visitors, and how many?
- Which of your pages get clicked from AI answers?
- Do those visitors convert differently from search or social traffic?
- Did a content change or a new citation lead to more visits?
Measuring this does not require a big investment. Even a simple filter on the referrer hostnames listed below gives you a starting picture. The more important habit is to measure it consistently, so you can compare weeks and months. Visits from AI assistants are usually a small share of total traffic for most sites today, and they may behave differently from other sources, so it is worth tracking them separately rather than guessing.
How AI assistants pass referrers
There is no single standard. Each product behaves differently, and behavior changes as products update. What follows reflects what is commonly observed, but you should always verify against your own data.
ChatGPT
ChatGPT links generally arrive with a referrer from chatgpt.com. Older traffic may show chat.openai.com, which was the original address before the product moved to the chatgpt.com domain. In addition, ChatGPT has added utm_source=chatgpt.com to links it shows in answers. That means a landing-page URL such as /pricing/?utm_source=chatgpt.com is a strong signal even when the referrer header is missing. Count both patterns, and keep the older hostname in your rules so historical data still classifies correctly.
Perplexity
Perplexity visits typically show a referrer from perplexity.ai, often as www.perplexity.ai. Because Perplexity is built around cited sources, it tends to send a relatively clean and recognizable signal. Match on the registrable domain, so both with and without www are caught.
Google Gemini
Gemini’s web app lives at gemini.google.com, and clicks from it can carry that hostname as the referrer. An earlier product name, Bard, used bard.google.com, which you may still see in older data. Gemini inside Android and other apps may behave differently, and Google’s AI features inside Search are a separate case, covered below.
Microsoft Copilot
Copilot traffic can show a referrer from copilot.microsoft.com. Before that, the Bing chat experience used Bing addresses, so older data may appear as bing.com with a chat path or none at all. Microsoft’s products and branding change often, so treat this as one to monitor rather than a stable rule.
Claude
Claude’s web app is at claude.ai, and clicks from it can carry that referrer. Claude’s volume of outbound clicks is often lower than other assistants, in part because of how and when it cites sources, but the pattern is the same: match the hostname.
Other AI products
Other products you may see include you.com, poe.com, phind.com, duckduckgo.com (which has an AI chat feature, though its normal search traffic is ordinary search), grok.com and various browser-based assistants. Some of these also add their own UTM parameters. The checker’s list covers the common ones, but new products appear regularly, so review your referral reports occasionally for unfamiliar AI-looking hostnames.
Google AI Overviews and AI Mode
Visits from Google’s AI features inside Search generally arrive with a Google referrer, the same as ordinary organic search clicks. In most analytics tools they are therefore indistinguishable from regular organic traffic unless you use Search Console reporting or other methods Google provides. Links in these features sometimes include a text fragment (a part of the URL starting with #:~:text=) that highlights the quoted passage on your page, but that is not a reliable or complete signal and is not visible to most analytics tools by default. Do not expect a referrer-based check to find these visits.
The reality: many AI visits arrive without a referrer
It is important to be honest about what referrer-based detection can and cannot do. Many visits from AI assistants carry no referrer at all. When that happens, analytics tools classify the visit as direct traffic, and the checker has nothing to match. Several things cause this:
- Native apps. When a link is opened from a mobile or desktop app, the app often does not pass a referrer to the browser or in-app web view.
- Referrer policies. Websites and apps control how much referrer information is sent. A strict policy, or links marked with
rel="noreferrer", strips it entirely. The common default sends only the origin (hostname), not the full path. - Copy and paste. Assistants often show URLs as plain text. If the person copies a link and pastes it into the address bar, the visit has no referrer.
- Security and privacy settings. Some browsers, extensions and privacy modes trim or remove referrers.
- Mentions without links. An assistant may recommend your brand by name, and the person then searches for you or types your address. That visit shows up as search or direct, with no trace of the AI origin.
The consequence is simple: the AI traffic you can detect is a floor, not a ceiling. The real number is higher by an amount nobody can know precisely. If you only count referrer-matched visits, you will under-report AI’s influence. If you treat a rising share of direct traffic as proof of AI influence, you will over-claim. A calibrated view uses several signals together and says clearly what is measured and what is inferred.
Signals beyond the referrer
Since the referrer is unreliable on its own, combine it with other signals.
UTM parameters
The utm_source=chatgpt.com pattern is the clearest explicit marker for ChatGPT. Check your landing-page reports for utm_source values that name an assistant. Remember that UTM parameters are added by the product, not by you, and the set of assistants that add them can change.
Landing-page patterns
Pages that receive a burst of direct traffic with unusual depth, such as a specific deep-link in a comparison article, may be getting AI clicks. This is circumstantial evidence only, but it can guide you to investigate.
Server logs and AI crawler visits
AI crawlers that fetch your pages are a different thing from visitors clicking through, but they tell you which engines can read your content. Seeing a crawler such as one operated by OpenAI, Anthropic or Perplexity on a page is a useful precondition signal. You can identify those with our AI Bot User-Agent Checker. Crawler visits are not human visits and should never be mixed into your visitor counts.
Post-purchase and sign-up questions
Ask new customers how they heard about you, and include “AI assistant such as ChatGPT or Perplexity” as an option. Self-reported attribution is imperfect but catches the visits that no referrer would reveal.
Branded search trends
A rise in searches for your brand name after a period of being cited by assistants is consistent with AI influence. It is not proof, but it is a useful corroborating signal when the other measures move in the same direction.
Setting up AI traffic reporting
Once you know what to look for, you can set up a simple, consistent report.
In Google Analytics
Create a custom channel group or a segment that matches session source against the AI hostnames, using a case-insensitive regular expression. A starting pattern is shown in the examples below. Also include matches on the utm_source value. Apply it to a report of landing pages so you can see which pages AI visitors reach. Check your own data before relying on the result: sources can appear in different formats such as chatgpt.com, chatgpt.com / referral or with a medium value.
In server logs
Search the referrer field in your access logs for the same hostnames and the query string for utm_source=. This catches visits that JavaScript analytics might miss because of blockers.
With a tracking snippet or plugin
A dedicated tracker can do the matching for you and keep a maintained list of assistants. The free aeoora WordPress plugin records AI referral visits for roughly 40 assistants without cookies and without storing IP addresses, and shows them in a live dashboard. For other sites, the aeoora Cloud tracking snippet does a similar job, as described below.
Best practices
Keep your list of AI hostnames in one place
Maintain a single documented list of the hostnames and UTM values you treat as AI sources, and reuse it in every report and filter. Review it every few months, because products rebrand and new ones appear.
Match on the registrable domain, not the exact string
Referrers can have subdomains, www, trailing paths or different protocols. Match on the domain portion, in lowercase, and handle both perplexity.ai and www.perplexity.ai. A pattern anchored too tightly will miss real visits.
Keep historical hostnames
When a product moves domain, such as ChatGPT’s change from chat.openai.com to chatgpt.com, old data stays under the old name. Keep both in your rules so trend lines are continuous.
Do not treat all matches as the same quality
Segment by assistant and by landing page. Visitors from one assistant may reach different pages, stay longer or convert differently from another’s. Lumping them together hides what is working.
Track conversions, not just visits
A small number of AI visits that convert at a good rate matters more than a large number that bounce. Tie AI sources to your key events, such as sign-ups, demo requests or purchases, and compare them with other channels using the same definitions.
Be transparent about uncertainty
When you report AI traffic to colleagues or clients, state that the figure covers only visits with a detectable signal. Present it as a minimum, and show the direct-traffic caveat. This builds credibility and avoids overreach.
Respect privacy
Referrer analysis does not require personal data. Prefer approaches that do not set cookies or store IP addresses where possible, and follow the consent rules that apply to your audience and region.
Examples
These examples show what typical values look like and how the checker would classify them. They are illustrative patterns, not guaranteed formats for any product.
Input: https://chatgpt.com/
Result: ChatGPT (referrer hostname: chatgpt.com)
Input: https://example.com/pricing/?utm_source=chatgpt.com
Result: ChatGPT (UTM source: chatgpt.com)
Input: https://www.perplexity.ai/
Result: Perplexity (referrer hostname: perplexity.ai)
Input: https://gemini.google.com/
Result: Google Gemini (referrer hostname: gemini.google.com)
Input: https://claude.ai/
Result: Claude (referrer hostname: claude.ai)
Input: https://copilot.microsoft.com/
Result: Microsoft Copilot (referrer hostname: copilot.microsoft.com)
Input: https://www.google.com/
Result: Not identified as AI. Google AI Overviews and AI Mode clicks
look like ordinary Google referrals.
Input: (empty)
Result: No referrer. Could be direct, an app, or a stripped AI visit.
A case-insensitive regular expression you could adapt for an analytics segment, matching the source or referrer hostname, might look like this. Test it against your own data, and extend it as you discover more sources:
(chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|claude\.ai|you\.com|poe\.com|phind\.com|grok\.com)
And a quick way to search a standard web server log for the same patterns from a command line, which is useful when JavaScript analytics is blocked:
grep -Ei "chatgpt\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|copilot\.microsoft\.com" access.log | wc -l
A small piece of client-side code can also read the signal in the page itself, which is how many trackers work. This is a simplified illustration, not a complete implementation:
const ref = document.referrer ? new URL(document.referrer).hostname : "";
const utm = new URLSearchParams(location.search).get("utm_source") || "";
const ai = /(chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|copilot\.microsoft\.com)/i;
const isAiVisit = ai.test(ref) || ai.test(utm);
Common mistakes
- Assuming no referrer means no AI. Many AI visits have no referrer and appear as direct. Undercounting is the default, not the exception.
- Assuming all direct traffic is AI. The opposite error. Direct traffic includes bookmarks, email apps, messaging and dark social. Rising direct traffic is a hint, not a measurement.
- Forgetting the old ChatGPT hostname. Excluding
chat.openai.combreaks historical comparisons. - Matching too narrowly. A rule that requires
https://www.misses subdomain and protocol variations. - Expecting Google AI features to show up. AI Overviews and AI Mode clicks look like organic Google visits in most tools.
- Counting crawler hits as visitors. AI crawlers fetching pages are not people clicking through. Keep them in a separate report.
- Never revisiting the list. Products rebrand, add domains and change how they pass referrers. An unreviewed list slowly becomes wrong.
- Reporting precise-sounding numbers. Give ranges or say “at least” when the data only supports a floor.
Reading your referral report: a worked example
Imagine a small software company reviews 30 days of traffic. Its analytics shows 12,000 sessions. A segment built from the hostname list finds 90 sessions from chatgpt.com, 35 from perplexity.ai, 10 from gemini.google.com and 6 from claude.ai. These numbers are invented for illustration, but the reasoning applies to real data.
The team can say, with confidence, that at least 141 sessions came from AI assistants. They cannot say that is all of them. Looking at the landing pages, they see that most AI sessions land on two pages: a comparison article and a pricing explanation. That is a useful finding, because it shows which content is getting cited and clicked.
Next, they look at the direct channel. It contains 2,400 sessions, and a few deep pages receive more direct visits than their internal linking would explain. That might be AI influence or it might be email and chat links. They do not add those sessions to the AI total. Instead they note the pages as candidates and ask new sign-ups where they heard about the product.
Finally, they compare conversion. If AI-referred sessions sign up at a higher rate than average, that is worth reporting, with the sample-size caveat. With only 141 sessions, a handful of conversions can swing the rate widely, so the right conclusion is “promising, keep measuring” rather than a firm figure.
A simple monthly routine
- Export sessions for your AI segment by source and landing page.
- Scan the referral report for unfamiliar hostnames that look AI-related and add them to your list.
- Compare AI sessions and conversions with the previous month.
- Check the pages that receive the most AI visits and make sure they are accurate, current and have a clear next step.
- Write down the caveats in the report so readers do not mistake a floor for a total.
Frequently asked questions
Does ChatGPT always send a referrer?
No. ChatGPT links often arrive with a referrer from chatgpt.com and, in many cases, with utm_source=chatgpt.com added to the URL, but not every click carries either. Visits from apps, copied links and privacy-restricted browsers can arrive with nothing. Count what you can detect and treat it as a minimum.
Why does my analytics tool show these visits as direct?
When there is no referrer and no UTM parameter, analytics tools have no source to report, so they label the session as direct. This happens often for visits that start in a native app or from a pasted link. It is a limitation of the data, not of your setup.
Can I tell Google AI Overviews visits from normal organic search?
Not reliably with referrers. Clicks from Google’s AI features generally carry a Google referrer, like regular search clicks, so they blend into organic traffic in most analytics tools. Search Console and Google’s own reports are the places to look for any breakdown Google offers, and its reporting may change over time.
Is utm_source=chatgpt.com something I need to set up?
No. ChatGPT adds it to the links it produces. You do not configure it, but you can read it in your analytics and use it in filters. Do not add it to your own links, because that would pollute your data with traffic that did not come from ChatGPT.
Will the checker work for referrers from new AI products?
It recognizes a maintained list of known assistants and AI search products. New or niche products may not be on the list, and the tool will say the value is not identified. If you see an unfamiliar AI-looking hostname in your referral report, add it to your own rules and check how the product describes its links.
Does the checker store the URLs I paste?
The check happens in your browser and does not contact the site you paste. If you are concerned about sensitive URLs, remove query strings that contain personal data before pasting. Check the page’s privacy notice for details on what the site itself collects.
How much traffic should I expect from AI assistants?
It varies enormously by site, topic and how much your content is cited. For many sites today it is a small share of total visits, and it can grow quickly if you start being cited for important questions. Measure your own baseline rather than relying on general claims, and track it over time.
Is the referrer the same as the user agent?
No. The referrer says where a human visitor came from. The user agent identifies the software making a request, and it is how crawlers announce themselves. A person clicking from ChatGPT uses a normal browser user agent. A crawler fetching your page for an AI product uses a bot user agent. Both are useful, and they answer different questions.
Go further with aeoora
Referrer rules take you part of the way. aeoora Cloud tracks real AI-assistant visits with a lightweight snippet that does not use cookies by default, and shows which assistants and pages drive traffic and conversions alongside your visibility in AI answers. Read the AI referral traffic docs and the tracking snippet guide to see how it works, then compare plans on the pricing page or start a 2-day trial. A card is required to start.
To see which AI crawlers are reading your site, the AI crawler analytics docs explain how verified bots are detected from your logs. WordPress users can get referral tracking and a crawl log for free with the aeoora WordPress plugin. And you will find more utilities in the free tools collection, including the AI Prompt Ideas Generator.