How Much Traffic Is AI Taking From Publishers?

The short answer: AI is taking meaningful traffic from publishers, but almost none of it is arriving as an AI chatbot referral. AI chatbots such as ChatGPT still account for well under 1% of publisher page-view referrals. The traffic that is genuinely disappearing is being lost upstream, inside Google's own results, where AI Overviews and zero-click answers suppress the clicks your existing rankings would otherwise earn. Pew Research found that when a Google AI summary is present, users click a traditional search result only 8% of the time versus 15% without one. Those are two different mechanisms, and conflating them is the single most common mistake publishers make when they try to size the damage.

How much traffic is AI actually taking from publishers?

To answer this properly you have to separate two things that look similar in your analytics but are nothing alike.

The first is demand-side click suppression. When Google places an AI Overview above your result, it answers the query on the results page, and the user often never clicks through. This is where publishers are losing real volume today. The second is supply-side referral substitution: AI chatbots sending their own traffic to your pages in place of a search click. This is growing quickly, but off such a tiny base that it is nowhere near offsetting the first.

Here is the scale of each, drawn only from measured data.

The disproportion is the whole story. In January 2025, The Atlantic received around 8.4 million organic search visits versus 186,000 from ChatGPT in the same month. The chatbot referral is real and worth capturing, but it is a rounding error next to the search traffic that AI Overviews are quietly eroding.

There is also a hidden imbalance on the crawling side. Cloudflare's crawl-to-refer ratio measures how many pages an AI platform requests for every referral visit it sends back. In July 2025 Cloudflare reported Anthropic's Claude at roughly 38,000 crawls per referral, OpenAI at about 1,091 to 1, and Perplexity at about 195 to 1. A high ratio means heavy content extraction with very little traffic returned. So even the referral traffic AI does send is dwarfed by how much of your content it consumes to produce answers, and because native app referrals carry no referrer, the true ratios are likely worse than published.

Is my traffic loss from AI Overviews or AI chatbots?

For almost every publisher, the honest diagnosis is AI Overviews, not chatbots. The maths makes this obvious. If chatbot referrals are under 1% of your visits, they cannot explain a 10%, 20% or 30% decline. A drop of that size has to come from the clicks you were already earning in Google.

The Overview effect is both large and intensifying. Ahrefs, analysing 300,000 keywords of aggregated Google Search Console data, found the position-1 click-through rate on AI Overview keywords fell from 0.073 in December 2023 to 0.016 in December 2025. The correlated CTR reduction grew from 34.5% in April 2025 to 58% by December 2025. In plain terms: ranking first is worth roughly half of what it used to be on any query that triggers an Overview.

Pew's behavioural data explains why. When an AI summary appeared, users not only clicked less, they left more often. 26% ended their browsing session after seeing a summary, versus 16% for searches without one, and only 1% clicked a link inside the summary itself. The AI Overview is not an alternative doorway to your site. For most users, it is the destination.

A quick way to attribute your own loss:

• Check Search Console first. If impressions are flat or up but clicks are falling, your rankings are intact and something is intercepting the click. That pattern points to AI Overviews, not to chatbots or a ranking drop.
• Check which queries lost clicks. Overviews trigger disproportionately on informational and question-shaped queries (more below). If your losses cluster there, that is the signature.
• Size the chatbot channel before blaming it. If your combined ChatGPT, Gemini, Perplexity and Copilot referrals are a fraction of a percent, they are not your problem, and they cannot be the fix for a large decline either.

Why does GA4 make this so hard to see?

Because GA4 does not report AI traffic cleanly, and it never has by default. Two things get in the way.

First, coverage. Google added a native AI Assistant default channel on 13 May 2026, which auto-tags qualifying visits with the medium ai-assistant and no configuration required. But it only recognises a fixed list of AI referrers. Google's initial release named ChatGPT, Gemini and Claude, and the live default-channel documentation has since expanded to include Copilot, Grok and Deepseek. Notably, Perplexity is still absent, so Perplexity visits stay buried in generic Referral unless you build a custom channel group with a source regex that matches its domain, placed above the default Referral channel.

Second, and more serious, is a structural blind spot. A session is only tagged as an AI referral if the inbound request carries a referrer header from a recognised AI domain. But a large share of AI-influenced visits carry no referrer at all. When a user copies a URL out of a chatbot answer and pastes it into the address bar, or arrives from a native mobile app, no referrer is sent. That visit lands in Direct, indistinguishable from a bookmark or a typed URL.

The consequence is unavoidable: GA4's referral-based AI numbers are a floor, not the true figure. Any honest AI-traffic dashboard has to flag that part of your Direct traffic is AI-influenced, rather than reporting referral counts as if they were complete. This is exactly why a clean read of your existing GA4, with the AI channels built out properly and the Direct caveat stated plainly, is more useful than another tracking script that still cannot see copy-paste visits.

How do I tell an AI decline from a seasonal dip?

This is the question that separates panic from diagnosis. A seasonal dip and an AI-driven decline can look identical in a single line chart, so you have to interrogate the shape of the loss, not just its size.

Use these tests, in order.

1. Compare year on year, not week on week. Seasonality repeats. If the same dip appeared in the same weeks last year and the year before, it is a season, not AI. AI Overview loss shows up as a structural step-down that does not recover when the season turns.
2. Split clicks from impressions. A seasonal dip usually moves impressions and clicks together, because demand itself fell. An AI Overview dip holds impressions roughly steady while clicks fall, because the demand is still there but the click is being intercepted on the results page.
3. Segment by query type. Pull your losing queries. If the decline is concentrated in question-phrased and long-tail informational queries, that matches the exact profile of AI Overview exposure. A broad seasonal fall hits navigational and branded queries too.
4. Check the CTR trend at your top positions. If you are still ranking first but earning fewer clicks per impression over time, that is the fingerprint of Overview suppression, consistent with the Ahrefs finding that position-1 CTR keeps falling on Overview keywords.

A worked example. Say a recipes and how-to publisher sees traffic down 22% year on year in the same quarter it always softens. Impressions are down only 4%. Clicks are down 22%. The loss is concentrated in "how to" and "what is" queries, and the site still ranks in the top three on most of them. That is not seasonality. Seasonality would have pulled impressions down with clicks. This is AI Overview interception, and the fix is not to wait for the season to turn, because it will not come back on its own.

Which content loses the most traffic to AI?

Not all pages are equally exposed. AI Overviews trigger far more often on some query types than others, and Pew's data lets you map your risk precisely.

Across the 68,879 Google searches Pew analysed in March 2025, about 18% produced an AI summary. But that average hides enormous variation by query shape:

The pattern is clear. Informational, explanatory and long-tail content is the most exposed. Definitions, explainers, "how to" guides, "what is" pages and question-led articles are exactly the material AI Overviews are best at summarising and most likely to trigger on. If your library skews toward that kind of content, your AI Overview exposure is well above the 18% headline average.

The least exposed content is short, navigational and branded. One-to-two-word queries produce a summary only around 8% of the time, so pages people reach by searching your brand or a specific product name are relatively insulated. Transactional and highly specific queries also tend to keep the click, because the user still needs to arrive somewhere to act.

The practical implication is to prioritise by exposure, not by traffic alone. A high-traffic explainer sitting on question-shaped queries is far more at risk than an equally trafficked branded page, and it should be first in the queue for whatever defensive work you do, whether that is deepening the content beyond what a summary can replicate, targeting queries that resist summarisation, or building direct audience relationships that do not depend on the search click at all.

Can I control which AI crawlers use my content?

Yes, and this is more precise than most publishers realise. AI crawlers are distinguishable and independently controllable, so you do not have to choose between all-in or all-out.

OpenAI, for example, runs separate bots with different jobs: GPTBot for model training, OAI-SearchBot which powers ChatGPT's search citations, and ChatGPT-User for user-initiated fetches. GPTBot and OAI-SearchBot are separately controllable in robots.txt, and OpenAI publishes verifiable IP ranges as JSON for each. That means you can disallow GPTBot to keep your content out of training while still allowing OAI-SearchBot, so you remain eligible to be cited in ChatGPT search. Extraction and visibility are separate decisions. Note that ChatGPT-User follows a user's explicit request to fetch a page rather than crawling automatically, so it is a different kind of access again.

Google works differently. Google-Extended is a robots.txt control token, not a separate crawler. Disallowing it blocks the use of your content for Gemini model training and grounding, but Google states plainly that "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search." Because AI Overviews are generated from the live Googlebot search index, there is no robots.txt lever that opts you out of AI Overviews while keeping you in Search. The two draw on the same index.

One practical caution: verify crawlers by reverse DNS or by matching the source IP against the vendor's published ranges, not by trusting the user-agent string, which is trivially spoofable. The major providers publish these ranges precisely so operators can confirm a bot is genuine before allowing or blocking it.

So what should a publisher actually do?

Start by measuring the right thing. The number that matters is not your chatbot referral count, it is your net click loss on the search queries you already rank for. Chatbot referrals are worth capturing and will grow, but today they are too small to explain, or to offset, a real decline. The volume you are losing is in Google, on informational queries, where the Overview answers before the click happens.

The concrete steps: split clicks from impressions to confirm the mechanism, segment your losses by query type to find the exposed content, and build out your GA4 channels so ChatGPT, Gemini, Claude, Perplexity and Copilot are all counted, while accepting that Direct hides more AI traffic than any referral figure will ever show.

If you want that read without building the reports yourself, try Ramprt free on your existing GA4. It gives a clean, plain read of what AI is doing to your traffic, referrals by engine, which crawlers are taking your content, and how AI-referred visits compare to your site average, with no new tracking script. For the wider picture, see our companion pieces on the impact of Google AI Overviews on publishers and how to track AI traffic in GA4.