Does ChatGPT Send Referral Traffic to Websites?

Yes, ChatGPT sends referral traffic to websites, but with two big caveats. It only happens when ChatGPT uses its live web search feature, which produces answers with inline citations and a Sources panel that link out to publisher pages. Clicking one of those citations is a genuine outbound referral. As of February 2026, however, ChatGPT triggered that search on only 34.5% of queries, so most answers send no referral at all. And of the referrals it does send, a large share arrives at your site with no referrer header, so your analytics quietly files it under Direct rather than as ChatGPT. The traffic is real, it is growing, and for most publishers it is both smaller than the hype suggests and larger than your dashboard admits.

Does ChatGPT actually send referral traffic, and how does the mechanism work?

ChatGPT search launched on 31 October 2024. When it runs, ChatGPT works as a retrieval-augmented system: it turns your prompt into one or more search queries, retrieves live documents, and generates an answer with inline citations plus a Sources panel. Clicking a citation or a source link opens the publisher's page, and that click is a real outbound referral, no different in principle from a click on a blue link in Google.

The important word is "when". ChatGPT frequently answers straight from its pre-trained knowledge with no links at all. The referral only exists at the end of a specific funnel:

1. The user asks a question.
2. ChatGPT decides to trigger a live web search (it often does not).
3. Your page is retrieved and cited.
4. The user clicks that citation.

Break any step in that chain and there is no referral. Semrush's clickstream analysis found ChatGPT triggered its search feature on just 34.5% of queries as of February 2026, down from 46% in late 2024 (Semrush, via Search Engine Land). In other words, roughly two out of three ChatGPT answers never reach step two, so they cannot send a referral to anyone.

How big is this traffic in practice?

Growing quickly, but still small and highly concentrated for most sites. The same Semrush study, drawing on more than a billion lines of US clickstream data across a 200 million-user panel from October 2024 to February 2026, found that outbound referral traffic from ChatGPT to the web grew 206% comparing January 2025 with January 2026 (Semrush, via Search Engine Land). The reach expanded too: ChatGPT referred traffic to roughly 71,000 unique domains per month in October 2024, peaking near 260,000 in October 2025 before easing to about 170,000 by February 2026 (Semrush blog).

One fact keeps this in proportion: the traffic is extremely top-heavy. Over 30% of all ChatGPT outbound clicks go to just 10 domains, and Google alone accounts for 21.6% of ChatGPT referral traffic, up from roughly 14% at the study's start (Semrush, via Search Engine Land). If a handful of giants absorb a third of every outbound click, the long tail of independent publishers is dividing a much smaller pool. In practice, most publishers who look will find AI-referred traffic sitting at roughly 1% or less of total sessions today. Fast growth on a small base is still a small base.

What source does ChatGPT show up as in GA4?

When ChatGPT referral traffic is attributed correctly, it arrives in GA4 as the session source / medium chatgpt.com / referral. You may also see older or variant referrers from the SearchGPT and Bing-backed era, and occasional oai.azure-style referrers. There is no automatic UTM tag: ChatGPT does not add campaign parameters to the links it shows, so recovery depends entirely on the referrer header surviving the click.

That last point is the crux. A referral only registers as chatgpt.com / referral if the browser passes a referrer along with the click. Very often it does not, and GA4 then treats the visit as Direct.

Why is my ChatGPT referral number lower than expected?

Because a large share of genuine ChatGPT clicks reaches your site as a naked request with no referrer, and GA4 classifies any referrer-less visit as Direct. This is not a GA4 bug. It is the sum of four separate stripping mechanisms, described in detail by Clickport:

• Referrer policy. chatgpt.com sends a strict-origin-when-cross-origin referrer policy, which can withhold the referrer on cross-site navigations.
• rel=noreferrer. Some inline links carry rel=noreferrer, which explicitly tells the browser to drop the referrer.
• In-app browsers. The iOS ChatGPT app opens links in a WKWebView and Android uses Custom Tabs. Both commonly drop the Referer header on external taps, and a great deal of ChatGPT usage is on mobile.
• Copy and paste. If a user copies a URL out of ChatGPT and pastes it into a browser, there is no referrer at all.

The result in every case is a request with no referrer, which GA4 files as Direct. Because ChatGPT adds no UTM, there is nothing to recover the attribution from. So the chatgpt.com / referral line in your reports is a floor, not the true total. When people say published referral figures understate ChatGPT traffic, this is exactly why. If you have watched your Direct channel swell while your AI referral line stayed tiny, a chunk of that Direct is almost certainly misfiled ChatGPT.

How do I see ChatGPT referrals in Google Analytics 4?

There are two reliable approaches, plus a server-log method to catch what they miss. All three capture only part of the picture, for the reasons above, but together they are the right starting point.

Option 1: segment the traffic acquisition report

In GA4, open Reports, then Traffic acquisition, and change the primary dimension to Session source / medium. Filter or search for chatgpt.com / referral. This gives you sessions, engagement and any configured conversions for the attributable ChatGPT slice, with no setup required.

Option 2: build a custom AI channel group

For an ongoing view, create a custom Channel Group. Add a channel (for example "AI") whose rule matches Source against a regex of AI domains, such as chatgpt.com, perplexity.ai and gemini.google.com. The critical step is ordering: place this AI channel above the built-in Referral channel, so AI hits are intercepted before they fall into generic Referral. Channel rules are evaluated in order, so a misordered rule leaves ChatGPT lumped in with every other referring site.

Option 3: close the gap with server logs

Because both GA4 methods miss the Direct-misfiled portion, the most complete picture comes from cross-referencing your server logs. Look for verified AI-crawler hits from OAI-SearchBot and ChatGPT-User (verification method below) to see which pages ChatGPT is fetching and citing, then watch for correlated lifts in Direct traffic to those same URLs. If a page starts getting fetched by ChatGPT-User and its Direct traffic rises at the same time, that is your missing referrals surfacing as Direct.

Which OpenAI bot controls whether ChatGPT can cite and refer you?

OpenAI runs several distinct bots, and confusing them is the most common way publishers accidentally cut off their own referral traffic. Per OpenAI's Overview of OpenAI Crawlers:

The practical rule for robots.txt: if you want ChatGPT referrals, allow OAI-SearchBot and ChatGPT-User. You can still block GPTBot on training grounds without losing a single citation, because ChatGPT search uses a different crawler. Blocking OAI-SearchBot and ChatGPT-User, on the other hand, is what actually removes you from the surface that sends the clicks.

Do not trust the user-agent string alone, since it is trivially spoofed. OpenAI publishes IP-range JSON files for verification at openai.com/searchbot.json, openai.com/chatgpt-user.json and openai.com/gptbot.json. Verify a bot by matching the source IP against the relevant list.

Does ChatGPT traffic earn well for ad-funded sites?

The honest answer has two halves. On volume, no, not yet for most sites: at around 1% or less of total sessions, ChatGPT is not moving your pageview-driven ad revenue in any material way. On quality per visit, the signals are encouraging but soft.

ChatGPT referral clicks are widely reported to convert well because the user is pre-qualified. The model has effectively pre-vetted the site before the click, so the person who arrives has already been told your page is the relevant answer. Several 2025 and 2026 vendor analyses report AI-referred visitors converting well above typical organic search visitors. Treat that as directional, not gospel: these are third-party marketing studies, not audited data, and conversion behaviour varies enormously by niche.

For a display-ad publisher there is a further wrinkle: even well-converting visitors only lift revenue if they view enough monetised pages, and if you cannot see the traffic, you cannot value it. The misattribution problem is therefore a revenue problem, not just a reporting one. If a growing, high-intent slice of your audience is landing in Direct, you will underinvest in the content ChatGPT actually cites. The fix is to measure net effect at page level, combining the attributable referral line, the correlated Direct lift, and your own conversion or engagement data, rather than judging ChatGPT by the chatgpt.com / referral figure alone.

How is this different from Google AI Overviews?

They are opposite mechanisms, and it matters. ChatGPT search is a chat surface whose citations are the primary way out to the web, so its default behaviour is to link. Google AI Overviews sit on the results page and mostly suppress clicks.

The contrast is stark in the data. Pew Research Center tracked 68,879 searches from 900 US adults in March 2025 and found that when a Google AI Overview appeared, users clicked a traditional search result in just 8% of visits, versus 15% without the summary, and clicked a link inside the AI Overview in only 1% of visits (Pew Research Center). Users also ended their browsing session on 26% of pages with an AI summary, against 16% without. Ahrefs put a trend on the same effect: the presence of an AI Overview correlated with a 34.5% lower average click-through rate for the top-ranking page in its April 2025 study of 300,000 keywords (Ahrefs), a figure it revised to 58% when it re-ran the study on December 2025 data (average position-one CTR on AI Overview keywords fell to 0.016, from 0.073 in December 2023) (Ahrefs update).

So AI Overviews tend to take clicks away, while ChatGPT search, when it searches at all, hands a limited number of clicks out. Neither replaces Google search volume, but they pull in different directions, and you should not lump them together in a single "AI traffic" bucket.

Why won't Google Search Console show my ChatGPT referrals?

Because it never could. Google Search Console only reports Google Search, so ChatGPT referrals have never appeared there and never will. GSC is simply the wrong instrument for this job; GA4 and your server logs are where ChatGPT traffic lives.

It is worth noting the parallel gap on Google's own AI surfaces. AI Overview and AI Mode appearances have been folded into standard Search performance metrics, with a single URL appearing in both an AI Overview and a blue link counting as one impression. Google launched a separate Generative AI performance report in beta on 3 June 2026, initially to a subset of UK sites, but in v1 it reports impressions only, broken down by page, country, device and date, with no clicks, CTR, position or query data (Google Search Central). So even for Google's AI, click-level visibility is thin, and for ChatGPT, GSC gives you nothing at all.

The bottom line for publishers

ChatGPT does send referral traffic. It is real, growing at triple digits year on year, and it appears to convert well because the model pre-qualifies the visitor. But it only fires on the roughly one-in-three queries where ChatGPT searches the live web, it is heavily concentrated among a few large domains, and a meaningful share of the clicks you do earn are hiding in your Direct channel. The right posture is neither panic nor dismissal. Allow OAI-SearchBot and ChatGPT-User so you stay eligible to be cited, set up the GA4 segment and channel group so you see the attributable portion, and cross-reference server logs and Direct lifts so you catch the rest.

If you would rather not build all of that by hand, try Ramprt free. It reads your existing GA4 with no new tracking script and shows AI referrals by engine, which AI crawlers are fetching your content, and what AI-referred traffic earns versus your site average, so you can see the ChatGPT picture that your default dashboard hides. For the wider context, see our related guides on why ChatGPT traffic shows as Direct in GA4 and how to track AI referral traffic in GA4.

Frequently asked questions

Does blocking GPTBot stop ChatGPT sending me traffic?

No. GPTBot is for model training only. ChatGPT search surfaces and cites pages through OAI-SearchBot and fetches live pages through ChatGPT-User, both different crawlers. You can block GPTBot and remain fully eligible for ChatGPT search referrals.

Why does my Direct traffic keep rising as ChatGPT grows?

Because chatgpt.com sends a strict referrer policy, some links carry rel=noreferrer, and the iOS and Android in-app browsers drop the Referer header. A referrer-less visit is classified as Direct by GA4, so genuine ChatGPT clicks land in Direct rather than as chatgpt.com / referral.

What percentage of my traffic should I expect from ChatGPT?

For most sites, AI-referred traffic is around 1% or less of total sessions, and over 30% of all ChatGPT outbound clicks go to just 10 domains. It is growing fast, up 206% year on year per Semrush, but remains small on an absolute basis for the long tail of publishers.

Can I see ChatGPT traffic in Google Search Console?

No. Search Console only reports Google Search and has never shown ChatGPT referrals. Its new Generative AI performance report covers only Google's own AI features, and in v1 shows impressions, not clicks. Use GA4 and your server logs to see ChatGPT.

How do I confirm a bot really is OpenAI and not a spoof?

Match the request's source IP against OpenAI's published IP-range files at openai.com/searchbot.json, openai.com/chatgpt-user.json and openai.com/gptbot.json. Do not rely on the user-agent string, which can be faked.