How to tell if AI is sending you traffic

The hostnames to watch

Answer engines pass a normal HTTP referrer when someone clicks a citation, so this is ordinary referral tracking with a specific list. As of this page’s last update, the ones worth having in your filter:

• chatgpt.com, and chat.openai.com for older sessions
• perplexity.ai, plus www.perplexity.ai
• claude.ai
• gemini.google.com, and bard.google.com if you have history going back that far
• copilot.microsoft.com, and bing.com/chat
• you.com, phind.com, poe.com if your audience is technical

One regex covers the lot:

.*(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|
gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|
bing\.com\/chat|you\.com|phind\.com|poe\.com).*

Keep it on one line when you paste it in. Escape the dots, or chatgpt.com also matches chatgptXcom, which is the sort of thing that never bites you until a competitor registers the domain.

Setting it up in GA4

GA4’s default channel groups do not have a bucket for this. Most AI referrals currently land in Referral, and a meaningful share land in Direct, because some clients strip the referrer. So you have to build the view yourself.

1. Admin, then Data display, then Channel groups. Create a custom group, copying the default as a starting point.
2. Add a new channel at the top of the list called AI Search. Order matters: channel groups evaluate top down and the first match wins, so if it sits below Referral it will never fire.
3. Condition: Source matches regex, and paste the regex above.
4. In Reports, then Acquisition, switch the dimension to your new channel group. Custom channel groups are not retroactive in every report, so also build a Comparison with the same condition if you want history.

Worth doing at the same time: add Explore → free-form with landing page against that segment. Which page an answer engine sends people to is more actionable than the total, because it tells you which page got cited.

Setting it up in Plausible

Considerably less work. Plausible already splits referrers sensibly, so you can either read them straight off the Top Sources panel, or make it permanent:

1. Filter on Source, choose contains, and enter the hostnames separated by the pipe character.
2. Save that filtered view as a Segment so it persists.
3. For a permanent tile, add a goal or use the shared-link dashboard with the filter baked into the URL.

Fathom and Umami work the same way. Any tool that lets you filter on referrer hostname and save the filter is fine, and if yours does not, the server logs will.

Why the number will look disappointing

Set your expectations before you look, because the first reading tends to be demoralising and it should not be.

Most AI answers are never clicked. That is the product working as designed. Someone asks a question, the model reads your page, the answer appears in the chat, and the person gets what they needed without visiting you. You were cited. You got no session. Zero-click has been the direction of travel for years and answer engines take it to its conclusion.

One to three percent of total sessions is a normal reading for a small site doing this well. If you are seeing that, nothing is broken.

The traffic that does arrive tends to be unusually good, which is the compensation. Someone who read a synthesised answer, saw you named as a source, and chose to click through has done more qualification than a search click. Compare conversion rate on the segment rather than volume, and it usually looks a lot better than the session count suggests.

The category this cannot measure

Google AI Overviews are invisible to all of this. An AI Overview renders on the Google results page, so a click from one is a click from google.com, indistinguishable from any other organic click in GA4 and in Search Console. There is no filter that separates it.

What you can watch instead:

• In Search Console, find queries where impressions hold steady, average position holds steady, and click-through rate falls. That divergence is the signature of an answer being read instead of a link being clicked. Compare a rolling 28 days against the same window a year earlier.
• Search your own key queries in an incognito window once a month and note whether an AI Overview appears at all. It varies by query type far more than by site.

The measurement that matters more

Referral traffic tells you how often you were clicked. For answer engines the more useful question is how often you were named, and no analytics tool reports that. You have to go and look.

Here is the whole method, and it takes about fifteen minutes:

1. Write down five questions a buyer would actually ask before finding you. Not your brand name. Things like “best invoicing tool for freelance designers” or “how do I get honest feedback on my SaaS”.
2. Ask each of them, once a month, in ChatGPT, Claude, Perplexity and Google AI Mode. Use a fresh session with no memory or personalisation, or you are grading your own homework.
3. Record two things per question per engine: were you named at all, and were you cited with a link. A spreadsheet with twenty cells is plenty.

Twenty data points a month, tracked over six months, will tell you more about your AI visibility than any dashboard. Almost nobody is doing this, which is exactly why it is worth doing.

Who should not bother

If you get under a thousand sessions a month, this segment will be single-digit visits and you will read noise as signal. Spend the fifteen minutes on the citation check above instead, which works at any volume because it does not depend on your traffic at all.

If you sell locally, or on a marketplace, or your acquisition is entirely outbound, this measures something that is not your growth channel. And if the honest answer is that you would not change anything based on the number, do not build the report. A dashboard nobody acts on is just a thing to feel bad about on Mondays.

Common questions

Once you can see the traffic, the next question is what to publish so there is more of it. The AEO checklist covers that, and comparison pages are where most indie SaaS founders find the fastest return.