Is SEO Dead Now That People Ask AI?

The short answer, and the two measurements under it

Start with the strongest public measurement of the click question, a Pew Research Center study published in July 2025. It is worth knowing how it was built, because the method is unusually direct. Pew used browsing data from 900 US adults who had agreed to install a tracking app, covering every search they ran in March 2025, then re-ran each of those searches through a scraping service in April to see what the results page had looked like. That gave 68,879 unique Google searches, 12,593 of which produced an AI summary.

Three findings matter. Users who met an AI summary clicked a traditional result on 8% of visits, against 15% for visits with no summary. Clicks on a link inside the summary itself happened on 1% of visits, so the citation links are not quietly replacing the blue ones. And the browsing session ended outright on 26% of pages carrying a summary, against 16% without.

Set against that, Google's own post from August 2025, written by Liz Reid, who runs Search. Her claim: "Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year," plus "average click quality has increased and we're actually sending slightly more quality clicks to websites than a year ago." Quality is defined in the post as a click where the user does not quickly return to the results page. The post disputes third-party traffic reports as resting on "flawed methodologies, isolated examples, or traffic changes that occurred prior to the roll out of AI features."

Note what Google did not do. No percentage, no click count, no date range, no method. "Our data shows" is the whole citation. I take the direction of the claim seriously because Google is the only party that can see the aggregate, and I treat the absence of a single number as what it is.

Why a halved click rate and steady total clicks are both true

These two accounts are not in conflict, and understanding why is most of the answer to the headline question. Pew measured a rate: clicks per results page. Google reported a total: clicks summed across all of Search. A rate can fall while a total holds, provided the number of searches rises enough to cover the difference. Google says exactly that happened, in the same post: "with AI Overviews, people are searching more and asking new questions that are often longer and more complex."

So the plausible shape of the last two years is more searches, each one less likely to produce a click, with the total landing in roughly the same place. That is a redistribution, not a collapse. Google concedes the redistribution plainly: traffic "decreased to some sites and increased to others."

Which puts the real question to anyone running a small site. Averages describe the web, and nobody operates the web. If total clicks held while the mix shifted, somebody absorbed a loss so that somebody else could absorb a gain. The pages that lost are the ones whose entire value was a fact a model can state in a sentence. The pages that gained are the ones a reader needed to open because the answer depended on their situation, or because they had already decided to trust a particular source.

This is also why asking "is SEO dead" produces such contradictory answers from people who are all being honest. A site that published 400 short definition pages has watched something die. A plumber with one good service page in one city has noticed nothing at all. Both are describing their own traffic accurately.

The queries that stopped paying, and the ones that did not

Pew's data has a second layer that gets quoted far less than the click figures, and it is more useful for planning. The study recorded which kinds of searches triggered a summary at all. One or two word searches produced one on 8% of occasions. Searches of ten words or more produced one on 53%. Searches phrased as a question produced one 60% of the time. The median summary ran 67 words, and 88% of summaries cited three or more sources. All of that is in the same report.

Read that as a map of where the risk sits. A fully formed question, typed out in a sentence, is the query most likely to be answered on the results page. A short, blunt, two-word query is the one least likely to be. Which is roughly the inverse of what the content advice of the last decade told everybody to target. The long, natural, question-shaped keyword was the safe recommendation precisely because it was easy to rank for, and it is now the one most exposed.

Three kinds of page come out of this in good shape. First, anything where the answer depends on the reader's particulars: a price that varies, a decision with tradeoffs, a procedure with choices in it. A summary can describe the shape of the answer and still leave the reader needing yours. Second, anything local, where the user needs a name, a phone number and an address, not an explanation. Third, anything that is the thing itself rather than a description of the thing: a tool, a calculator, a template, a product page, a demo.

And one kind comes out badly: the page whose job was to define a term, state a date, convert a unit or list five obvious tips. Not because Google punished it. Because the reader got what they came for without arriving. If that describes most of a site, the honest reading is that the model did not kill the business so much as reveal what the traffic had been worth all along.

The part of the plan that did not change at all

Here is the part that gets lost in the argument. Almost everything that made a small site findable in 2019 still makes it findable now, and none of it is AI-specific. A page has to exist at a stable URL. A crawler has to be allowed to fetch it. The title has to say what the page is. The site has to be navigable by links rather than only by a search box. Pages have to load on a phone on a bad connection. Somebody somewhere has to link to you.

Every surface that answers questions, Google's included, is reading pages it found by crawling the web. There is no second index for AI, no parallel submission route, no separate ranking system that bypasses the ordinary one. A page nothing can crawl is invisible to all of it at once. That makes basic technical hygiene more load-bearing than it was, not less, because the same work now serves several consumers.

If your site is new, the AI question is almost certainly not your problem. Your problem is that you have no links, no history and nine pages, and you would be invisible in a search engine from 2015 for exactly the same reasons. I wrote the unglamorous version of that work up in SEO for a new website, and the traffic side of it in how to get traffic to a new website. Neither has needed revising because of AI search, which is itself a data point about how much changed.

The demand an answer engine cannot sit in front of

The Pew figure I keep coming back to is the session-ending one: 26% of pages with a summary ended the browsing session, against 16% without, per the study. That is a picture of someone who had a question, got it answered, and left. There was never a version of that visit where they became a customer. The person was never shopping.

Which points at the demand no answer engine sits in front of: people who already know your name. Somebody typing your business into a search box has done the deciding somewhere else, and they are not asking a question an AI summary can satisfy. They want your page. That search resolves to you or it fails, and no amount of synthesis changes it.

I want to be careful here, because this point gets oversold. I have found no credible public measurement of branded search volume rising because of AI search, and I am not going to invent one. The argument for brand is structural rather than empirical: a query that names you can only be answered by you. What it implies for a small operation is a reordering of effort. The work that makes people remember a name, which is to say doing something distinctive, being visible in places with their own audiences, and having a product worth mentioning, now sits above the work of competing for generic informational phrases.

One uncomfortable corollary: this is slower, harder to measure and far less teachable than keyword work, which is why most advice still points at the keyword work.

What I would change, and what I would leave alone

What I would actually change, given the two measurements above:

Stop writing pages whose entire payload is a fact. If a competent model can answer the question in two sentences and be right, the page is not an asset. Write the version that requires judgment instead, and put something in it that only you have: your numbers, your procedure, the thing that went wrong when you tried it.

Keep one page per thing you sell, and make it good. The product page, the service page, the contact page. These are the pages that convert and the ones a named search lands on, and they are consistently the least maintained pages on a small site.

Measure impressions as well as clicks in Search Console. If impressions hold while clicks slide on informational pages, you are watching the effect described above rather than a ranking problem, and the fix is different. Ranking problems want better pages. This wants different pages.

Spend an hour on crawler permissions and page structure, then stop. That is the whole technical surface of AI visibility for a site this size. I went through what the acronym is actually selling in what AEO is and whether a small site needs it, and the citation side in how to get mentioned by ChatGPT and Claude. Both are short on purpose.

What I would not do: rip up a site that is working, buy a monitoring subscription before having anything to monitor, or hire anyone selling a new acronym. The retainer has been renamed and repriced. The work underneath it has not changed as much as the invoice suggests.

A verdict by site type

The verdict depends on what the site is for, so here it is by type.

Local service business. Least affected of anybody. Your searches are short, local and transactional, which is the category least likely to produce a summary at all. Maps, reviews and a page that loads still decide this. Carry on.

Selling a product. Mildly affected, mostly upstream. People may research the category without visiting you and arrive later knowing more. Your product page matters more and your explainer posts matter less. Keep the ones that help somebody choose.

Content site funded by ads. Badly affected, and it is not recoverable by technique. The model was paid impressions on traffic that came for facts. Both AI summaries and the honest reading of your analytics say the same thing about that model.

Brand-new site. Unaffected, because you have nothing to lose yet. Build the thing, get the pages crawlable, get a few real links. The AI question is a distraction at your stage and will still be there in six months.

Somebody already asks for you by name. This is the position worth engineering toward, and the one the last two years have made more valuable rather than less.

The phrase "SEO is dead" has been in circulation since roughly 2008 and has never once been right, which is not an argument that it cannot become right. It is a reason to want evidence. The evidence available says a rate fell, a total held, and the mix underneath moved. That is a real change with real losers, and it is not the end of the practice.

The version I would defend: being findable still pays, and the kind of page that pays has narrowed. Facts are free now. Judgment, specificity, local presence and a name people remember are not. A small site that leans on those four was never competing for the traffic that went away.

Common questions

Is SEO still worth learning in 2026?

Yes, with a narrower focus than a few years ago. The technical core has not changed: a crawlable site, one clear page per topic, sensible titles, internal links and real links from elsewhere. What is worth less is keyword research aimed at short informational questions, since those are the queries most likely to be answered on the results page without a click. Learn the fundamentals, skip the tactics aimed at thin content.

Did AI Overviews reduce traffic to small websites?

It depends entirely on what the site's pages do. Pew Research measured click rates roughly halving on searches where an AI summary appeared, so pages that existed to state a fact lost traffic. Google says total organic clicks across the web held steady while distribution shifted between sites. Local, transactional and product pages have been largely untouched; short definition and listicle pages took the loss.

Should I stop writing blog posts if AI can answer the question?

Stop writing the posts whose answer a model can give correctly in two sentences, because the reader will never need to arrive. Keep writing where the answer depends on circumstances, involves a tradeoff, or contains something only you have: your own figures, your procedure, the mistake you made. Those pages still earn a click because a summary of them is not a substitute for them.

Do I need an AI visibility tool or a new agency to handle this?

Almost certainly not at a small-site scale. The technical surface is a crawler-permissions check and page structure, which is an afternoon of work, not a retainer. Monitoring tools report which prompts mention you, which is interesting and rarely actionable until you have enough traffic for the answer to change a decision. Spend the money on the product or on the one page that converts.