Why search traffic is being quietly rerouted through LLMs

I look at a fair number of Google Analytics dashboards in a normal month. Across most of the small and mid-sized UK sites I work with, something has been quietly happening to organic traffic over the last eighteen months. Not a single dramatic drop, but a slow downward drift in informational queries that did not have an obvious cause when you looked at rankings or content.

The cause was not on the site. It was on the results page, in the AI Mode panel, in ChatGPT and in Perplexity. A growing share of the questions that used to end on a website now end on an AI answer. Most of the affected sites are not losing rank. They are losing the click. This post is the developer and SEO read on what is happening, what is causing it and what it means for how you measure organic traffic from here.

The mechanic of a zero-click answer

When a user runs a query in Google AI Mode, ChatGPT search, Perplexity or Claude with web access, the assistant fans the query out across a set of sources, pulls relevant passages and assembles an answer in the chat surface. Some answers cite the sources prominently. Some bury the citations in a small drawer. In every case the user has read the answer before they reach the citation, which means most of them never click.

Google AI Overviews are the most visible version because they sit on top of an existing Google results page. Studies by independent SEO researchers published in early 2026 found that AI Overviews are associated with a substantial drop in click-through on the underlying organic links, with one Ahrefs analysis reporting a 58 percent reduction in click-through for top-ranking pages compared to a non-AI Overview query baseline. The effect varies by query type and is most pronounced on informational queries where the AI summary genuinely answers the question.

AI Mode goes further. It is a separate conversational interface that does not always show ten blue links at all. The user asks a question, gets a structured answer and follow-up prompts. Sources appear, but the format is built for reading not for clicking through. Google has been adding features through 2025 and into 2026, including a Further Exploration block that surfaces additional links beneath the AI answer, in response to publisher pressure over the click decline.

What the publisher data shows

Chartbeat tracked Google referral traffic to publishers across its network through 2025 and reported a roughly one-third decline in global Google traffic to publishers in the year to November. Press Gazette covered the data in early 2026. The decline was uneven. Lifestyle, weather and utility publishers were hit hardest because the queries that bring users to those sites are the same kind of queries AI products answer well. News and analysis sites were less affected because users still want the live update or the named writer.

The Reuters Institute Journalism, Media and Technology trends report for 2026 surveyed publishers worldwide and found that, on average, they expect search traffic to almost halve over the next three years, with the headline figure reported at around a 43 percent expected decline. The mood among the publishers I read is closer to managed decline than panic, with most planning to put less effort into traditional Google SEO and more into direct audience relationships, email, podcasts and brand search.

For small UK businesses, the headline numbers from publisher data are not directly transferable. Most of what I see on client sites is a softer version of the same trend. Informational queries shed traffic. Branded and high-intent commercial queries hold up. The drop is not uniform across pages, which is the clue that the cause is query-type and not site-wide.

Why this is hard to see in analytics

The traffic is being rerouted, not lost from the index. Your rankings can be unchanged or improving while your traffic is falling. That is what makes it hard to spot. The page is still being matched to the query. The click is just not happening because the answer was delivered inside the AI product before the user got to your link.

Search Console will sometimes show this directly. Impressions hold up while clicks fall, which pulls average position down only because click-through has collapsed. If you look at top queries and see that your informational pages have lost click-through on queries the AI tools answer well, you are looking at the effect.

Attribution makes it worse. A user who asks Claude or ChatGPT and reads your answer in the chat surface, then comes back days later to look you up by brand, lands as a direct or branded organic visit. The credit for the AI citation does not show up anywhere in your analytics. The visit looks like it came from nothing, when in fact the AI surface was the discovery channel.

Why publishers feel it sharpest

Publisher business models depend on volume. Programmatic advertising, affiliate revenue, sponsored content and newsletter signups all key off page views. A 30 percent decline in clicks is a 30 percent decline in advertising revenue, give or take. That is why the Press Gazette coverage has tracked the pain in detail and why the publisher response has been the loudest.

The legal response has started too. Antitrust filings in the US and the UK have begun to point at AI Overviews and AI Mode as evidence that Google is using publisher content to keep users on its own surfaces. Several publishers have started blocking AI crawlers, which carries the trade-off that they then do not appear in citations either. The economics of that trade-off depend on whether your category produces clicks from AI products at all.

For commercial sites that are not publishers, the picture is different. A consultancy or a B2B services business does not need a million page views. It needs the right enquiries. The AI surface can still feed those enquiries if you are getting cited well, because a buyer asking ChatGPT for the best UK Craft CMS freelancer is at a more useful point in the decision than a user typing the same query into Google.

What this means for organic traffic baselines

The 2023 organic traffic baseline is no longer a fair comparison. If your dashboard still uses it, you are reading every month as a decline. The honest baseline is the post-AI-Overviews period from early 2025 onwards, recalibrated by query type. Informational pages should be expected to bleed traffic. Decision-stage and brand pages should hold or grow.

The metric I now track alongside organic traffic on client sites is brand search volume and direct visits to specific pages from the content. If a particular blog post is referenced often in AI answers, you may see a small lift in direct visits to that URL or in brand searches that mention concepts the post owns. These are weak signals, but they are the only signals available until analytics catches up.

I also pay attention to whether my pages get cited when I run my own questions through ChatGPT, Claude, Perplexity and Google AI Mode. It is a small, manual check, but it tells you whether your content is in the consideration set the AI is drawing from.

What this is not

It is not the end of SEO. Most of what was good SEO is still good GEO. Strong on-page clarity, structured data, fast pages, internal linking, link equity and freshness. I have written about the overlap and the gaps in generative engine optimisation explained.

It is also not a reason to panic about your existing content. Pages that were earning informational traffic are still doing useful work for users who arrive through any surface and for AI products that are citing them. The work to do is not to delete pages or chase a new tactic. The work is to update your baseline expectations, look at where traffic is shifting and decide which categories of page are worth more or less attention from here.

What I recommend in practice

Rebase the dashboard. Stop comparing to 2023. Compare to a rolling twelve-month window starting after AI Overviews became widespread in your category. Segment by query intent. Watch impressions alongside clicks.

Write differently for informational queries. The AI surfaces favour content that gives a direct answer in the first paragraph, with the supporting reasoning afterwards. Buried lead paragraphs and SEO intros that delay the answer no longer help and may actively hurt your citation odds.

Lean into entities. Make sure your business, your name, your products and your unique terminology are described consistently across your site and across the web. Entity consistency is one of the few signals AI systems can use to decide that you are a credible source.

Keep doing the technical SEO work. Schema, sitemaps, internal links, page speed. None of this has stopped mattering. Most AI search products are still running on top of a search index and the index still ranks by familiar signals.

If you want help thinking about how AI surfaces are affecting your particular sites, see AI integration services or get in touch and send me a URL.