Sign in
Back to insights
Seen from above, a small robot lowers a handwritten card onto a table of documents, with threads running from the card out to each one and a single gold photograph among them.

Is SEO Dead? The Click Is Dying, But SEO Is Not

8 min read

Google users click a search result 8% of the time when an AI summary is on the page. Without one, 15%. They click a link inside the summary 1% of the time.

That is Pew Research, tracking 68,879 real searches made by 900 people. It is the best evidence anyone has that the thing we all built websites for is going away. So is SEO dead? No. But the click is, and those are not the same thing.

What the numbers actually say

Pew followed the real browsing of 900 US adults for a month rather than asking them what they do, which is what makes it worth more than most of what gets published on this. Three findings matter.

  • A search with an AI summary produced a click on a traditional result 8% of the time. Without a summary, 15%. That is a little under half.
  • Users clicked a source cited inside the summary in 1% of visits. The citations are decorative, as far as traffic goes.
  • People ended their browsing session entirely on 26% of pages with a summary, against 16% without. The summary does not just replace the click. It ends the trip.

Zoom out and the picture is the same. Zero-click searches, where nobody goes anywhere at all, now account for 68% of US queries. In Google’s AI Mode, which replaces the results page with a conversation rather than sitting above it, there is no organic list to click in the first place.

What Google says, and what it has shown

Google disputes the story. Its head of Search says total click volume is “relatively stable” year on year, and that what AI removes is mostly bounce clicks, where someone opens a page, takes one fact, and leaves. The clicks that survive, the argument goes, are better ones.

That is a reasonable hypothesis. It has been repeated in a blog post, on a podcast and in interviews, and in none of them has Google published the figures. Pew showed its working on 68,879 searches. One of these two positions can be checked.

Both can also be true at once, which is the part worth sitting with. Total clicks across the whole web can hold steady while your site loses badly, if the clicks concentrate into a smaller number of very large destinations. Stable in aggregate is not the same as stable for you.

Why SEO is not dead

Here is the thing that gets lost in the funeral notices. An AI summary is not written from nowhere. It is assembled from pages the search engine already ranked, and being cited in one is downstream of ranking for the query. You cannot be quoted by a system that never found you.

So the discipline survives almost entirely intact. Crawlable pages, sensible structure, content that answers the question, speed, images that are not enormous. All of it still applies, for the same reasons. What changed is the reward. It used to be a visit. Now it is often a mention, and a mention is worth less, which means the same work buys less than it did.

That is a real loss and worth being honest about. It is not the same as the work being pointless.

It already had a better name, and Google retired it

When Google launched this in May 2023, it called it the Search Generative Experience. A year later it renamed the feature AI Overviews and the phrase quietly disappeared.

Which is a shame, because it was the more accurate name. What sits on a results page now is an experience being generated for one person, not a list of ten blue links to be optimised against. The industry has landed on generative engine optimisation and answer engine optimisation instead, which are the same idea wearing an acronym.

Names matter here because they change what you measure. If you are still counting positions, you are counting the thing that used to pay.

What actually gets you into the answer

There is early research rather than folklore on this. A team at Princeton built a benchmark of around 10,000 queries across nine datasets and tested what makes a generative engine quote a source. The three strongest techniques improved visibility by 30 to 40% relative to the unedited page, and low-ranking sources gained the most.

What worked was unglamorous:

  • Add statistics. A model reaching for something to say reaches for a number.
  • Quote credible sources, by name, in the text.
  • Cite where your claims came from. The pattern a model rewards is the pattern a careful writer already uses.

Keyword density did not make the list. Neither did anything that looks like the SEO of ten years ago.

One structural change is worth making everywhere: answer the question in the first paragraph rather than building to it. A retrieval system reads the top of the page to decide whether the page is about the thing. An article that spends four paragraphs clearing its throat is an article a model gives up on.

The part nobody puts in these articles: images

Google Lens now handles close to 20 billion visual searches a month, by Google’s own count. People point a camera at a thing and ask about it. That is a search with no text query in it at all, and it is one of the fastest growing kinds.

Meanwhile the models answering text questions are multimodal. They do not only read your words about the picture. They read the picture. And then they read the filename, the alt text, the caption, the paragraph around it, and decide together what the image shows.

Which quietly promotes two things that were treated as chores. Alt text stops being only an accessibility obligation and becomes the sentence that tells a machine what your picture is. A filename stops being housekeeping and becomes a label the retrieval step can read before it decodes a single pixel. Neither was ever wasted effort. Both now do a second job.

The practical version has not changed much from what we wrote about image SEO, which is reassuring rather than disappointing: describe the image honestly, name the file for what it shows, keep it small enough not to slow the page. The difference is who is reading. It used to be a crawler building an index. Now it is also a model building an answer, and a person pointing a phone at a shelf.

What to do about it

  • Stop measuring rank alone. Track whether you are being cited in AI answers for your own queries. Ask them and look.
  • Answer in the opening. Put the conclusion first and the reasoning after it. This helps models and readers equally.
  • Be quotable. Write sentences that survive being lifted out of the page, because that is what happens to them now.
  • Bring your own numbers. Original measurement is the thing an aggregating machine cannot manufacture, and the thing it most wants to quote.
  • Describe every image properly. Alt text and filenames are now inputs to a model, not just an index. See 25 alt text rewrites for what good looks like.
  • Own a destination worth arriving at. If a summary can replace your page, it will. A tool, a dataset, a calculator or a service survives summarising in a way that a listicle does not.

Questions people ask

Should I stop investing in SEO?

No, but you should expect less traffic per unit of effort and plan for it. The work that gets you cited is mostly the work that got you ranked. What has changed is what you get at the end of it.

Do AI summaries appear on everything?

Not yet. In Pew’s data they appeared on about 18% of searches, and they skew heavily to question-shaped queries: 60% of searches phrased as a question got one, against 8% of one or two word searches. Informational content is taking the hit first.

Is there any upside?

Two. Being quoted by an answer engine is a strong endorsement in front of someone with clear intent, even without a click. And the techniques that earn it, original data and named sources, are the ones that were always worth doing and were often skipped because keyword volume paid better.

Does this change what I do with images?

It raises the stakes on the same practices rather than replacing them. Descriptive filenames, real alt text and files small enough not to slow the page were already right. They are now read by more systems, in more places, including one holding a camera.

Sources: Pew Research Center, Google users are less likely to click on links when an AI summary appears, from 68,879 searches by 900 US adults collected March 2025; Search Engine Land, on zero-click search reaching 68% of US queries; Google, AI in Search is driving more queries and higher quality clicks, for Google’s position, which is stated without figures; Aggarwal et al., GEO: Generative Engine Optimization, KDD, for the benchmark and the 30 to 40% visibility figures; Google, on Lens handling nearly 20 billion visual searches a month. Where Google and independent researchers disagree above, both are given and the difference is that only one of them has published data.