How Much Traffic Does ChatGPT Actually Send? Data From Real Sites

Everyone says AI traffic is coming. Almost nobody publishes their numbers. Here are ours.

Methodology, stated plainly: this first edition is a deep look at one B2B SaaS site on the ezStats platform (~9,000 human visits) across January 1 to June 10, 2026, with AI referrals classified by our Search Everywhere report. One site is a case study, not a market study; we're publishing it because a real number with caveats beats a vibe, and we'll expand this into a quarterly multi-site report as our platform dataset grows. Treat every figure below as one site's reality, directionally useful, not a benchmark.

The headline numbers

Of all classified search-style traffic on this site in the period:

• Traditional search: 90.7% (1,195 sessions: Google 723, Bing 298, DuckDuckGo 110, Yahoo 63, Baidu 1)
• AI assistants: 8.0% (106 sessions: ChatGPT 90, Copilot 7, Perplexity 7, Claude 2)
• Social: 1.3% (17 sessions)

The full source table, since this is the part people quote:

So: AI is the second-largest search category on this site, ahead of social. ChatGPT alone (90 sessions) outsent Yahoo (63), and ran at about one-eighth of Google's volume. For a channel that effectively didn't exist three years ago, that's the story.

How AI visitors behave (the uncomfortable part)

Honesty requires showing the ugly columns too. On this site, AI-referred sessions were shallow: short durations and bounce rates in the 95-100% band, similar to or worse than other search channels on the same site.

Three plausible, non-exclusive readings:

1. Answer-first behavior is real. The AI already answered the question; the click is verification. A 10-second visit that confirms "yes, this product exists and looks legitimate" can still be a step in a real buying journey
2. Some of it isn't human. AI tools and agents fetch pages programmatically; despite bot filtering (this site's raw traffic was 40.5% classified bots), some automated fetches pass as sessions. Every published AI-traffic number, ours included, contains some of this
3. Landing-page mismatch. AI answers deep-link to whatever passage got cited, which is often not a page designed to convert a first visit

The takeaway isn't "AI traffic is junk." It's that session-quality metrics undersell a channel whose main work happens before the click.

Why every measured number is a floor

Two structural undercounts apply to all analytics tools, including ours:

• Missing referrers. Some AI tools and in-app browsers open links without a referrer header; those visits land in Direct, unattributable by anyone
• The zero-click effect. The biggest AI influence never produces a visit at all: the assistant answers, names brands, and the user either stops or later searches the brand name (showing up as branded organic or Direct). Sales teams hear "I asked ChatGPT" far more often than analytics shows chatgpt.com referrals

Which is why measuring AI traffic without measuring AI visibility (whether the assistants mention you at all) reads the scoreboard while ignoring the game. The visibility side is the leading indicator; here's the full explainer.

What to do with this

1. Get the channel visible in your own analytics. ezStats classifies it by default; GA4 needs a custom channel group; Plausible shows AI sources too
2. Judge the channel on trend, not volume. Single-digit share today; watch the slope quarterly
3. Check your visibility baseline. Run the free AI Visibility Score, one minute, no login, then work the citation playbook
4. If you have your own numbers, publish them. This space desperately needs more real data; we'll happily link to honest reports

FAQ

How much traffic does ChatGPT send to websites? It varies enormously by niche. On the B2B SaaS site in this study (Jan-Jun 2026), ChatGPT sent 90 sessions, about one visit for every eight from Google, making AI assistants 8% of classified search traffic. Most sites today see low single-digit shares, growing quarter over quarter.

Which AI tool sends the most referral traffic? ChatGPT, by a wide margin in our data: 90 sessions vs 7 each from Microsoft Copilot and Perplexity and 2 from Claude on the studied site. ChatGPT's user base makes it the dominant AI referrer for most sites.

Why is the bounce rate from ChatGPT traffic so high? Three factors: the AI already answered the question so visits are quick verification, some "sessions" are automated fetches by AI tools rather than humans, and AI answers deep-link to cited passages rather than landing pages. High bounce on this channel does not necessarily mean low value.

Does Google Analytics show ChatGPT traffic? GA4 records it under the Referral channel as source chatgpt.com; there's no AI channel by default. You can build a custom channel group to surface it, or use a tool that classifies AI traffic out of the box.

Is AI referral traffic undercounted? Yes, structurally. Some AI tools open links without referrer headers (those visits appear as Direct), and the largest AI influence, answers that resolve without any click or that drive later branded searches, never appears as AI traffic at all. Measured AI referrals are a floor.

Will AI traffic replace Google traffic? Not on current evidence; on this site traditional search was still 91% of search-style traffic. The realistic 2026 posture: treat AI as a fast-growing minority channel, measure it properly, and invest in AI visibility before your category's answers calcify around competitors.