Google AI Mode: What It Means for Publisher Traffic

Google AI Mode is a distinct, more conversational version of search that expands on AI Overviews. It launched in Google Labs around March 2025, was announced at Google I/O in May 2025, and rolled out to all US users from around 13 June 2025. For publishers the practical answer is blunt: AI Mode does still send clicks to external sites, but Google blends those clicks into your ordinary organic numbers so you cannot see them separately, and the weight of independent evidence shows AI answers suppress click-through and raise zero-click behaviour. This article explains exactly what you can measure, what you cannot, and how to keep earning citations and clicks inside AI Mode.

Is Google AI Mode the same as AI Overviews?

No. They are related features from the same AI stack, but they behave differently, and it matters for how you read your data.

AI Overviews are the AI-generated summary boxes that appear at the top of an otherwise normal results page. AI Mode is a separate, full-screen conversational experience that expands on that idea. Its defining mechanism is query fan-out: instead of running your one query, AI Mode issues multiple related searches across subtopics at once and assembles a wider set of supporting links than classic search would show. You then ask follow-up questions and it keeps the thread going.

That difference in surface produces a difference in measurement. In an AI Overview, the feature does not occupy a single discrete ranking position. In AI Mode, each component inside the answer (a link card, an image block, a carousel) gets its own position under Google's standard search element rules, as Google documents in its AI features guidance. So a single AI Mode answer can generate several impressions for your site at several different positions.

The one thing they share is the part publishers care about most: both are reported inside the same Search Console Performance report, under the same 'Web' search type, with no way to tell them apart. So while AI Mode and AI Overviews are genuinely different experiences for the user, they are effectively indistinguishable in your reporting.

Does AI Mode send any traffic to publishers?

Yes, some. This is the point most doom-laden takes get wrong. AI Mode is not a pure dead end.

Inside AI Mode, the counting rules are specific:

• A click on any external link within the answer counts as a click in Search Console.
• Your page appearing anywhere in the AI response counts as an impression.
• Each element in the answer has its own position, so you can rank in several places within one response.

AI Mode data has been included in Search Console Performance totals since 16 June 2025. So the clicks are real and they do land in your reports. The problem is the rate. The behavioural evidence is consistent that AI answers suppress the click, even when good links are on offer.

The strongest independent study is from the Pew Research Center, which tracked real browsing behaviour from 900 US adults across 68,879 searches. When a Google AI summary appeared, users clicked a traditional search result only 8% of the time, versus 15% of the time when no summary was shown. They clicked a link inside the summary itself only 1% of the time. In other words, the presence of an AI answer roughly halves the odds of a click to a publisher.

Zero-click behaviour rises at the same time. In the same Pew data, users ended their browsing session entirely 26% of the time when an AI summary appeared, versus 16% when it did not. A separate randomised field experiment by researchers at the Indian School of Business and Carnegie Mellon University, reported by PPC Land, found that when an AI Overview was shown, outbound organic clicks fell by 39.8% and zero-click searches rose by 34.5%, with no measurable effect on sponsored clicks. Because it randomised whether users saw the AI Overview, it isolates cause rather than correlation. The traffic that disappears is organic, publisher-bound traffic.

The trend is also getting worse. Ahrefs, studying 300,000 keywords, measured a 34.5% lower average CTR for the top-ranking page on AI Overview keywords in April 2025. By its December 2025 update that had widened to a 58% lower CTR. Position-1 average CTR fell from 0.073 in December 2023 to 0.016 in December 2025, when Ahrefs' model suggests it would have been around 0.037 without AI Overviews.

So the honest summary is: AI Mode sends traffic, but per impression it sends far less than a classic blue-link result did, and the gap is growing.

A worked example of the decoupling

Imagine an informational article that used to rank position 1 for a question-format query.

The figures are illustrative and drawn from the Ahrefs position-1 CTR benchmarks, not a single measured page. The pattern is the point: impressions hold or rise while clicks collapse. That divergence, impressions flat or up but clicks and CTR down, is the fingerprint of AI answers eating your click, and it is the thing you can actually watch in your own data.

Can I see AI Mode traffic separately in Google Search Console?

No, and this is the single most important thing for publishers to understand. There is no filter, segment, or search type in Search Console that isolates AI Mode or AI Overview clicks, impressions, or CTR.

Both features are counted inside the existing Performance report under the 'Web' search type, as Google confirms in its AI features documentation. An AI-driven click and a classic blue-link click sit in the same bucket. You cannot natively see what share of your Search Console clicks came from an AI experience.

Two further wrinkles make attribution even harder inside AI Mode itself. First, because of query fan-out, one user question can surface your page against several subtopic queries, so your impressions get spread across queries you never targeted. Second, and more disruptive, follow-up questions in AI Mode are treated as brand-new searches. Clicks, impressions and positions are attributed to the new query, not the original one, as Search Engine Land explains. So a single conversational session that starts on one topic and ends on a click to your site can be scattered across multiple, disconnected query rows. You cannot reconstruct the journey.

GA4 does not rescue you either. Google AI Overview and AI Mode clicks originate on google.com and are indistinguishable from ordinary organic clicks in the referrer, so you cannot isolate them in GA4 by referrer at all. The wider AI-referral picture is also undercounted: when a user clicks a link inside an AI app or embedded browser, the referrer header is often stripped, so GA4 buckets the session as Direct rather than Referral. As the AuthorityTech attribution guide sets out, Perplexity passes perplexity.ai reliably, ChatGPT began appending utm_source=chatgpt.com in June 2025 but still drops attribution on its mobile app, and Google's AI clicks pass no distinguishing signal at all.

So what can you actually do? Measure the decoupling pattern on your own data rather than chasing a filter that does not exist:

• In Search Console, watch for queries and pages where impressions are flat or rising while clicks and CTR fall. On informational, question-format content that is the observable signature of AI answers intercepting the click. Compare period over period, not against a headline industry percentage.
• In GA4, build a custom channel group with a referral regex such as chatgpt\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com. This recovers the AI referrals that do pass a referrer. It will still miss the referrer-less sessions that fall into Direct, so treat a rising, hard-to-attribute Direct share as a soft proxy for AI-referred traffic you cannot name.

Because Google blends the data, the impact is only visible as a trend on your own site. That is exactly why chasing a single "X% of searches now have AI" figure is a distraction. Reported AI Overview prevalence has swung from 6.49% of queries in January 2025 to a peak near 25% in July 2025, then down to 15.69% by November 2025, per Semrush data compiled by QuickSEO, with other 2026 trackers reporting anywhere from roughly 15% to 65% depending on keyword set and method. The number that matters is the one on your own dashboard.

How do I stay cited in AI Mode?

Start by separating two decisions that are constantly confused: whether AI is allowed to train on your content, and whether you appear in AI answers. They are not the same lever.

The opt-out trap

Google-Extended is a robots.txt token only. It is not a crawler you will ever see in your server logs. It controls whether your content is used to train and ground Gemini and Vertex AI. Crucially, disallowing Google-Extended does not remove you from AI Overviews or AI Mode, because those features are served from the standard Googlebot index. The only way to fully exit AI Mode is to exit Google Search indexing itself, for example with noindex or nosnippet, which also kills your normal organic traffic. There is no clean escape hatch, and that impossibility of selective opt-out is a central grievance in publisher complaints against Google. So for almost every ad-funded publisher, the realistic goal is not to hide from AI Mode but to be the source it cites and clicks through to.

What actually earns the citation and the click

AI Mode assembles answers from the standard index, so classic quality and structure signals still drive whether you are pulled in. The behavioural data points to what to prioritise:

• Own the query shapes that trigger AI answers. Longer queries (BrightEdge found queries of roughly 8 words or more are about 7x more likely to trigger an AI Overview) and question-format queries disproportionately trigger AI answers. That maps almost exactly onto informational publisher content, so structure pages around specific questions and their direct answers.
• Answer the sub-questions, not just the headline. Query fan-out means AI Mode is running several subtopic searches at once. Content that cleanly covers the adjacent sub-questions has more surfaces to be cited against and picks up more of those per-element positions.
• Front-load a clear, self-contained answer. Because each element in AI Mode carries its own position, passages that stand alone as a crisp answer are easier to cite and rank as a component.
• Give people a reason to click past the summary. With top-of-page click rates falling toward the low single digits, the click now goes to pages that offer something the summary cannot flatten: original data, tools, first-hand reporting, depth, and specificity the model has to send the user to get.

Know your AI crawlers, and verify them properly

If you do want to make training-versus-answering decisions deliberately, you need to recognise the crawler landscape. It splits into training bots, search-index bots, and user-triggered fetchers, and blocking one is not the same as blocking another.

The vendor documentation is the authority here: OpenAI publishes its user agents and IP-range JSON files (gptbot.json, searchbot.json, chatgpt-user.json), Anthropic documents ClaudeBot, Claude-User and Claude-SearchBot and notes the older Claude-Web and anthropic-ai tokens are deprecated, and Perplexity documents PerplexityBot and Perplexity-User.

Two practical cautions. First, PerplexityBot is a search-index bot that surfaces and links sites, and it is not used to train models, so blocking it can cost you Perplexity citations rather than protect you. Second, user-agent strings are trivially spoofable, so verify crawlers by IP range or reverse DNS, not by name alone. OpenAI, Google, Perplexity, Common Crawl and Anthropic all now publish machine-readable IP-range files (Anthropic at claude.com/crawling/bots.json), so IP verification is available for the major crawlers.

The takeaway for AI Mode specifically: none of these robots.txt decisions affect your presence in Google AI Mode, which runs off Googlebot. Your AI-crawler policy is a training and third-party-citation decision. Your Google AI Mode presence is decided by ordinary indexing and content quality.

What publishers should actually do next

The industry-percentage debates about how common AI answers are miss the point for any individual site. The variance in the numbers, from roughly 15% to 65% depending on who is counting, is itself the argument for ignoring the headline and measuring your own GA4 and Search Console.

A grounded action list:

1. Track the decoupling on your key informational pages: impressions versus clicks versus CTR, period over period. Rising impressions with falling clicks is your AI-answer signal.
2. Add the AI-referral channel-group regex in GA4 to recover the referrals that do pass, and watch Direct as a rough proxy for what does not.
3. Do not block Google-Extended expecting it to remove you from AI Mode. It will not. Make training opt-outs a separate, deliberate call.
4. Verify AI crawlers by IP, not user agent, before you block anything, so you do not accidentally cut off the search-index bots that cite you.
5. Invest editorially in the things a summary cannot replace: original data, tools, first-hand expertise, and depth.

If you want to see the decoupling and the AI-referral picture on your own site without wiring up regex and custom channel groups by hand, try Ramprt free on your existing GA4. It reads your data read-only, with no new tracking script, and its AI tab surfaces AI referrals by engine, which crawlers are taking your content, and how AI-referred visitors compare to your site average. For the wider referral gap, see our guide to why AI traffic does not show up in GA4, and for the crawler side, AI crawlers explained.