What AI-Referred Traffic Actually Earns Publishers

AI-referred traffic is worth more per visitor than organic search on almost every engagement and conversion metric, but there is far less of it, and whether that adds up to real money depends entirely on how you make money. If you monetise by subscriptions, leads or product sales, the conversion premium is large enough that each AI visitor can be worth several search visitors. If you monetise by ad impressions, AI traffic is thinner than it looks, because there is very little of it today and it tends to land on high-intent pages rather than deep, ad-heavy article journeys. The honest answer to "what does AI traffic earn" is this: measure it on the metric that matches your business, and expect the total to be small for now even where the per-visitor value is high.

Does AI-referred traffic earn more than search traffic for publishers?

Per visitor, yes, and the gap is not marginal. The most defensible dataset comes from Microsoft Clarity, which analysed 1,277 publisher and news domains and disclosed both its sample and its method. It found that LLM-referred visitors converted to sign-ups at 1.66%, against 0.15% from search, 0.13% direct and 0.46% social. Subscription conversion was 1.34% for LLM traffic versus 0.55% for search. Referrals from Copilot achieved the highest subscription conversion of any AI source, converting at 17 times the direct rate and 15 times the search rate, and 52% of the 1,277 domains studied converted some AI traffic into sign-ups or subscriptions in the month measured.

The engagement data points the same way. Ahrefs found AI-referred visitors viewing roughly 50% more pages per session than traditional search visitors, and Similarweb's analysis found AI-influenced visitors viewing close to twice as many pages and spending close to twice as long on site as non-AI-influenced visitors. That deeper engagement is the mechanism behind the conversion premium: people arriving from an AI answer have often already had their basic question resolved, so the click through to your site is a considered one, lower in the funnel and closer to an action.

There are two important caveats before you treat this as free upside. First, the base is tiny. Across ten major industries, AI platforms drive about 1% of overall web traffic, and Microsoft's own data shows LLM traffic remaining under 1% of the total even after growing 155.6% over eight months. A large multiple on a small base is still a small number. Second, the widely quoted 4x to 5x conversion figures from single-vendor reports carry selection bias, because AI often arrives on high-intent queries near the point of decision. Treat those vendor multiples as directional. The Clarity and Pew numbers are the ones to lean on, because they publish their samples.

Where "more valuable" quietly becomes "less valuable"

For an ad-funded publisher on Raptive, Mediavine or AdSense, the calculus flips. Your revenue is not conversions, it is impressions. A visitor is worth roughly the number of ad impressions they generate multiplied by your page RPM. AI visitors help on pages per session, but their volume is around 1% of traffic, so the total incremental ad revenue is small today no matter how engaged each one is. Worse, high-intent AI visitors often land on conversion-oriented pages, product pages, comparison pages and answer pages, rather than the deep, ad-dense article journeys that drive display revenue. So the same traffic that a subscription publisher celebrates can look thin to a pure-display publisher. Neither is wrong. They are measuring different things.

How do I measure earnings from ChatGPT and Perplexity referrals in GA4?

Start from an uncomfortable fact: you almost certainly cannot see most of your AI traffic in GA4 yet. The engines behave differently. Perplexity's Comet browser usually passes a referrer and shows up cleanly as perplexity.ai / referral. ChatGPT, and especially the Atlas in-app browser and shared-link webviews, frequently strips the referrer header, so those sessions are logged as Direct or (not set). Given that ChatGPT accounts for about 87.4% of AI referral traffic, this means the single largest source of AI referrals is the one GA4 is worst at attributing. Left uncorrected, your AI value is scattered into Direct and undercounted.

The fix is to build your own attribution rather than trust the defaults. Two steps:

• Classify known AI hostnames. Build a custom channel group or referral rules in GA4 that recognise chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, among others, and group them into an "AI" channel. Decide deliberately whether to treat each as referral or as a form of organic-AI.
• Inspect the raw source/medium. Go to Reports, then Acquisition, then Traffic acquisition, and set the primary dimension to Session source / medium. That is where perplexity.ai / referral will appear, and where you can watch how much of your Direct traffic spikes when a piece gets cited by ChatGPT.

Because so much ChatGPT traffic hides in Direct, treat a rise in Direct that coincides with an AI citation as a signal, not noise. This is exactly the reporting gap Ramprt is built to close. It reads your existing GA4 read-only, applies AI channel classification for you, and shows AI referrals by engine plus what that AI-referred traffic earns against your site average, without adding a tracking script. You can see it working on the free live demo.

The half you cannot get from GA4 at all: Google AI Overviews

There is a second, larger AI effect that never appears as a GA4 referral channel: Google's AI Overviews and AI Mode. When a link inside an AI Overview is clicked, Google counts it as an ordinary Search Console click, not a distinct AI channel, and the AI Overview occupies a single position with all its links sharing that position. So Overviews mostly show up as fewer clicks on your existing organic listings, not as a new referral line. To understand your full AI picture you have to read GA4 referral data (ChatGPT, Perplexity, Gemini) and Search Console click behaviour (AI Overviews) side by side. They describe different halves of the same shift.

Why does my AI referral RPM change so much month to month?

Because most of the month-to-month movement is measurement noise, not a real change in what the traffic is worth. Four mechanisms drive the volatility.

• Referrer stripping is inconsistent. Whether a given ChatGPT session carries a referrer depends on the client, the app version and whether it is an in-app webview. A month where more sessions land in Direct will show a lower measured AI RPM even though the underlying visitors are identical.
• The sample is tiny. At around 1% of traffic, a handful of high-value sessions, or their absence, swings the average sharply. Small numbers are volatile numbers.
• Engine mix shifts fast. Between September and November 2025, ChatGPT referrals grew 52% year on year while Gemini grew 388%. As the mix between engines that pass referrers and engines that strip them changes, your measured AI RPM moves before anything about visitor quality has.
• Search Console aggregation distorts CTR. AI Overview and AI Mode impressions are folded into the standard Web performance report, not broken out. Because an Overview can generate an impression when its link is expanded into view without a click, pages that appear often in AI features show rising impressions and falling clicks, which mechanically depresses your reported CTR even when your ranking has not moved. Ahrefs measured this as a 34.5% lower CTR for top-ranking pages, comparing March 2024 with March 2025.

Google's dedicated Generative AI features report launched in beta on 3 June 2026, but in version one it is impressions only, with no click, CTR, position or query data. So you still cannot isolate AI-feature click value inside Search Console. The practical response to volatile AI RPM is to stop reading month-to-month and start reading rolling three-month trends, and to fix your GA4 channel rules first so you are at least comparing like with like.

Is it worth optimising content for AI traffic when the volume is small?

It depends on your monetisation model, and this is the decision that matters most. Run the two calculations honestly.

A worked example makes the split concrete. Take 100,000 monthly sessions. AI referrals at 1% are 1,000 sessions. For a display publisher at, say, three page views per session and a page RPM in the low tens of pounds, that is a small addition to a large impression base, real but not transformative. For a subscription publisher, apply the Clarity gap: 1,000 AI sessions at a 1.34% subscription rate is around 13 subscribers, whereas 1,000 search sessions at 0.55% is around 5 or 6. If a subscriber is worth 50 pounds a year, that difference alone is meaningful, and it compounds as AI referral volume grows.

The critical framing, though, is that AI referral gains are not offsetting the search losses running underneath them. Publisher referral traffic is falling. A Digital Content Next survey of 19 publishers reported a median year-on-year referral decline of 10%, with news brands down 7% and non-news brands down 14%, and individual case studies cited alongside it showed click-through rates falling from around 5% to under 1% and traffic drops of up to 25%. AI referrals at 1% of traffic come nowhere near closing that gap in ad-impression terms. So for a display publisher, "optimise for AI traffic" is not a rescue plan for lost search volume. It is worth doing to defend citation share and to capture the high-intent visitors AI does send, but it will not refill the impression bucket that AI Overviews are draining.

What optimising for AI actually means

Because ChatGPT drives around 87.4% of AI referral traffic, referral-side optimisation is overwhelmingly a ChatGPT exercise today: clear, extractable answers, well-structured pages, and content that is genuinely worth clicking through to after the summary. Google AI Overviews are a different job. There the aim is not a new referral channel but preserving the clicks you still get on your organic listings, since an Overview can end the session before any click. Pew found that users ended their browsing on 26% of pages that carried an AI summary versus 16% of pages with only traditional results, and clicked a traditional result in just 8% of visits with a summary present against 15% without. That is the click erosion you are defending against.

Putting it together: what to actually do

The sequence is straightforward. First, fix measurement, because you cannot value what you cannot see. Build AI channel rules in GA4 and read Search Console alongside them. Second, choose the right yardstick: impression maths if you sell ads, conversion maths if you sell subscriptions or products. Third, judge results on rolling quarters, not volatile single months. Fourth, keep the numbers in proportion. AI-referred traffic is a small, high-quality stream that is growing quickly and matters most to subscription and lead businesses. For pure display publishers it is a useful floor and a signal of where attention is heading, not a replacement for the search referrals that AI Overviews are quietly taking away.

If you want the AI view of your own GA4 without setting all this up by hand, Ramprt reads your existing analytics and shows AI referrals by engine, what AI-referred traffic earns against your site average, and where Google AI Overviews are decoupling your impressions from your clicks. Try the free live demo.