How to track visitors coming from ChatGPT, Perplexity and Gemini to your WordPress site

Visitors arriving from ChatGPT, Perplexity and Gemini identify themselves through exactly two signals: the HTTP referrer (chatgpt.com, perplexity.ai, gemini.google.com) and, for some assistants, a utm_source parameter appended to the link — ChatGPT has tagged outbound links with utm_source=chatgpt.com since 2025. Filter your analytics for those strings and you can see this traffic today, with no new tools; the honest catch is that whatever number you get is a floor, not a total, because a large share of AI-referred sessions arrive stripped of both signals.

Most guides on this topic jump straight to GA4 channel-group configuration. That's the last step, not the first. If you understand the two raw signals — where they come from, when each assistant sends them, and when they silently disappear — every analytics setup becomes obvious, and you'll stop trusting the numbers more than they deserve.

What signals do ChatGPT, Perplexity and Gemini actually send?

Each assistant leaves one or both of these fingerprints on a click-through: a referrer header naming its domain, and sometimes a UTM parameter it appends to your URL. As of July 2026, this is what we see in practice:

Two details matter more than the table suggests. First, the UTM parameter survives situations the referrer doesn't — it's part of the URL itself, so it persists through redirects and referrer-stripping browsers. That makes utm_source=chatgpt.com the single most reliable AI-traffic signal that exists right now. Second, the referrer tells you the assistant but never the conversation — you will know a visitor came from Perplexity, but not what they asked. There is no AI equivalent of a search-terms report, and no setup changes that.

How do you see AI visitors in GA4?

Open Reports → Acquisition → Traffic acquisition, switch the primary dimension to "Session source / medium", and search for chatgpt, perplexity, and gemini — sessions that kept their referrer or UTM are already there. For a permanent view, create a custom channel group with a regex condition on source, something like chatgpt|openai|perplexity|gemini|copilot, so AI referrals stop being scattered across the generic Referral channel.

That's the whole GA4 setup, and it works identically on WordPress or anywhere else. The part the setup guides rarely say plainly: GA4 can only classify sessions that carry a signal. It cannot recover the stripped ones, and Google Analytics undercounts your AI traffic for reasons that are structural, not configuration mistakes — consent-mode modeling, in-app browsers, and the Direct bucket swallowing everything unlabeled.

How do you track AI visitors on WordPress without GA4?

You need any layer that records referrers and query strings — three realistic options, in increasing order of usefulness:

1. Server access logs. Every request's referrer is already in your logs; grep for the domains above. Free and complete, but it mixes humans with bots, counts requests instead of sessions, and nobody reads raw logs weekly.
2. A general stats plugin. Jetpack Stats, Koko Analytics and similar record top referrers, so chatgpt.com will start appearing in the referrer list once the traffic exists. Fine for awareness; there's no per-assistant view and UTM parameters are usually ignored or need manual filtering.
3. Purpose-built AI-traffic measurement inside WordPress. This is the gap we kept hitting while building Contexta: site owners wanted one report answering "which assistants send me people, and to which pages" — so its AI traffic report matches both the referrer and UTM signals server-side, before consent banners and ad blockers thin out what JavaScript analytics can see, and breaks it down per page.

The per-page breakdown is the part that changes decisions. Site-wide "you got 40 visits from ChatGPT" is trivia; "these three articles get cited and this money page never does" tells you what to write next.

Why will your numbers always undercount?

Because both signals are fragile, and the biggest slice of AI influence produces no visit at all. The failure modes stack: native mobile apps and in-app browsers frequently strip referrers, so those sessions land as Direct traffic. Some browsers apply strict referrer policies that truncate or drop cross-origin referrers. Redirects and security proxies can shed UTM parameters. And when an assistant answers the user's question inline — which is the normal case — the user gets value from your content with zero clicks, leaving nothing to measure on your side.

When we tested this against our own properties, the pattern that emerged wasn't a precise percentage — anyone selling you one is guessing — but a consistent direction: measured AI referrals lag visibly behind how often the same pages turn up cited when you probe the assistants by hand. Treat your tracked count as the tip of the iceberg, useful for trends rather than totals: if chatgpt.com sessions doubled month over month, that direction is real even if the absolute number is understated.

What should you actually do with the data?

Use it to find which of your pages AI assistants already trust, then feed that pattern. The workflow that works: filter AI-referred sessions by landing page, list the top ten, and look at what those pages have in common — they're almost always the ones with direct, self-contained answers near the top. That's your template for the next round of content, and a stronger signal than any generic best-practices list, because it comes from your own site.

The second use is defensive: a baseline. AI referral traffic is growing from a small base, and platform behavior changes without notice — referrer formats have already changed once (chat.openai.com → chatgpt.com). If you don't have a stable measurement in place, you can't tell a real drop in AI visibility from a tracking artifact. Set the filter up once, check it monthly, and date any anomaly so you can correlate it with platform changes later.