You are paying for search work every month, your blog sessions have been sliding since spring, and you cannot tell whether you are funding something that stopped working. So you searched the obvious question, and you got the same answer four times in a row: not dead, just evolving. That answer is technically true and completely useless, because it never says who it is dead for, and you have a real suspicion you might be one of them. The most upvoted reply on the biggest recent discussion of this says it better than we could: most of the answers feel theoretical. Here is the non-theoretical version. For some businesses the honest advice this year is to stop paying, and we would rather name them than pretend the answer is the same for everyone.
Key Takeaways
If your traffic was mostly people asking simple questions, and you made money from that traffic itself rather than from what those readers did next, the thing you were buying has been taken away and it is not coming back.
AI summaries appeared on 18% of searches, but on 53% of queries of ten words or more and 60% of queries starting with a question word. Long, question-shaped keywords are the exposed ones.
With an AI summary present, a traditional result was clicked on 8% of visits against roughly 15% without one, and links inside the summary itself on 1%.
Google states plainly that its AI features need no additional requirements, no special schema, and no new machine-readable files. Anyone selling you one is selling you something Google says you do not need.
Rankings and sessions can hold while the work stops producing calls. If your report does not end in enquiries, it is not reporting on your business.
The honest answer starts with who it is dead for
Every page you have read on this hedges, so this one will not. There are businesses for which the search work being sold today is genuinely not worth buying, and if you are one of them you should stop.
The clearest case is a business whose traffic was mostly people asking simple questions, and whose money came from that traffic itself rather than from what those visitors did next. Ad-supported content, affiliate round-ups, definition pages, anything whose entire product was being the page that answered a quick question. That model depended on the click, the click is what has been taken away, and no amount of better writing brings it back. Being told to adapt is not useful advice when the thing you were selling has been absorbed into the results page.
The second case is narrower and gets less airtime. If you sell to an audience that has already moved to asking assistants directly, your searches are exactly the ones being answered before anyone reaches a website. A business-to-business marketing agency founder put it bluntly in public this month: if you are selling to marketers or sales people, divert the budget elsewhere. That is an uncomfortable sentence for a company that does search work to repeat, and it holds for that specific audience.
The third case is not about AI at all, and it catches more people than the other two combined. If you are paying a monthly fee, receiving a report full of rankings and sessions, and cannot point at a single enquiry that came from it, the problem predates AI Overviews by about a decade. Cancelling that counts as ordinary good sense rather than a strategic response to the future of search, and the current panic is simply giving you cover to finally act on it.
If none of those three describes you, keep reading, because the rest of this is about how to tell what is actually happening to your own site rather than to the market.
Your exposure is decided by query type, not by your industry
This is the part that lets you stop guessing, and it is the most useful thing on this page.
AI summaries do not appear evenly. The Pew Research Center recorded the real browsing of 900 US adults across nearly 69,000 Google searches and found summaries on 18% of searches overall, but on 53% of queries of ten words or more, and 60% of queries that began with a question word.
| Kind of search | How often an AI summary appeared |
|---|---|
| All searches in the study | 18% |
| Queries of ten or more words | 53% |
| Queries starting with a question word | 60% |
Read that as an instruction rather than a statistic. Open your own keyword list and sort it into two piles. Long, question-shaped searches go in the exposed pile. Short, commercial, local or brand-shaped searches go in the other one.
A plumber ranking for "emergency plumber" and a supplier ranking for "stainless fabrication near me" are in a very different position from a site ranking for "how long does epoxy take to cure". The first two are short, transactional, and resolve into somebody picking a business. The third is a question with a single answer, which is exactly what an assistant is built to hand over without sending anyone anywhere.
Most businesses turn out to be a mix, and the balance usually surprises them. Firms that thought of themselves as heavily exposed often find their commercial terms are untouched and only their blog is bleeding. Firms that felt safe sometimes find that most of what they rank for is questions. Either way you now know which half of your keyword list to argue about, which is more than the market-wide percentages will ever tell you.
The click loss is real, and pretending otherwise would be dishonest
Anybody telling you nothing has changed is arguing with measurements.
The same Pew work found that when an AI summary was present, a traditional search result was clicked on 8% of visits, against roughly 15% when no summary appeared. Links inside the summary itself were clicked on 1% of visits, which is the finding that should temper anyone's enthusiasm for citation as a traffic plan. Being named in the answer is worth having, and it is closer to being mentioned on the radio than to being sent a customer.
Two vendor studies point the same way and should be read with their limits attached. Ahrefs compared 150,000 informational keywords that triggered AI Overviews against 150,000 that did not, and estimated position-one click-through 34.5% below its counterfactual, from an expected 4.0% down to an observed 2.6%. Seer Interactive tracked 3,119 informational queries across 42 client organisations over fifteen months and found the same direction.
Both are worth knowing and neither is a census. Ahrefs and Seer are companies with products in this market, Seer's is drawn from its own client portfolio rather than from the population, and both deliberately looked at informational keywords, which are the most exposed kind. None of that makes them wrong. It means the honest reading is that there is a serious and well-evidenced informational-click problem, not that there is a single traffic loss percentage that applies to your business.
We have written separately about what zero-click search means for a small business and about whether AI is killing website traffic, so this piece stays on the money question rather than re-explaining the phenomenon.
There is no separate AI checklist you can buy
A whole product category has appeared to sell you readiness for this, and Google has been unusually direct about it.
Its own guidance on AI features states that there are no additional requirements or special optimisations for AI Overviews and AI Mode, and specifically that no new machine-readable file, AI text file, or special schema type is needed. A page has to be indexed and eligible to show a snippet in ordinary Search, and that is the entry condition.
That single fact should change how you read a proposal. If the pitch involves a new file to install, a proprietary AI schema, or a subscription that promises inclusion in AI answers, it is selling you something the search engine says is not a requirement. Nobody can promise a citation, in the same way nobody can promise a ranking.
What genuinely has gained value is unglamorous and mostly already on your invoice. Pages that are crawlable and indexed with their important text in the HTML rather than assembled by scripts. Structured data that matches what is visible on the page. Accurate Business Profile information where you serve a local market. First-hand information and original evidence rather than a restatement of what everyone else published. That last one has teeth: Google's guidance on helpful content names content produced mainly to attract search visits, and summaries that add nothing, as warning signs.
If you want the longer treatment of how citation actually works, we cover it in generative engine optimization and in a separate piece on the tools people use to track it.
Three questions decide whether your current retainer is worth renewing
You do not need an audit to answer these, and they take an afternoon.
The three questions, in the order worth asking them
- 1
One, what is the search work actually for
Write down what you want it to produce. Enquiries, booked jobs, applications, quote requests, orders. If the honest answer is "traffic" or "rankings", you have found the problem and it is not AI. Traffic was always a proxy, and it has just become a much worse one.
- 2
Two, has anything reached you from it
Go through the last three months of enquiries and count how many can be traced to a search someone ran. Not sessions. People. If the number is zero and has been zero for a year, the spending is not working, whatever the report says about position four.
- 3
Three, is the gap in the exposed half or the commercial half
Open Search Console, set sixteen months, put impressions and clicks on one chart, then split by query. Losing clicks on question-shaped searches while commercial searches hold is the AI story. Losing both together is a ranking, indexing or seasonality story, and treating it as an AI problem will send you fixing the wrong thing for a quarter.
The third question is the one people skip, and it is the one that decides everything else. Google publishes its own guide to debugging a traffic drop precisely because the shape of the drop tells you the cause, and most drops are not what the owner first assumed.
What to stop paying for, and what to start asking your report to prove
Some of what a retainer covered five years ago has genuinely stopped earning its place.
Publishing interchangeable definition pages to capture an informational click is the clearest one. That was always a thin proposition and it is now a losing one, because the click it was built to catch is the click most likely to be absorbed. Volume for its own sake is the same story with a bigger invoice attached. A law firm publishing hundreds of generated pages a month is not adapting to AI search, it is producing exactly what Google's guidance names as a warning sign, at scale.
What deserves the money instead is the work that survives whether or not anybody clicks a blue link. Pages that carry your prices, your process, your actual work and a way to act. Technical health, because being indexed and eligible for a snippet is the entry condition for everything else. Accurate local information if you serve a local market, which is the part an answer engine has least ability to replace, and which we cover under local SEO. Original evidence, because a summary can restate a definition and cannot restate what you found doing the job.
And the reporting has to change with it. Ask for enquiries, calls and booked work, segmented into the exposed and commercial halves of your keyword list. Add referrals arriving from assistants, which will be small and is worth watching for direction. If you want a second free source of directional data, Microsoft has begun publishing an AI Performance report in Bing Webmaster Tools; Bing is not where most of your customers search, but it is currently more forthcoming about AI surfaces than the alternative.
The strongest signal will reach you in conversation before it reaches a dashboard
This one never shows up as a metric, and it is the most useful thing on the list.
If enquiries start arriving from people who already know your process, your rough pricing and what you do not do, they read about you somewhere and possibly inside an answer they never clicked. That is the shape of the new funnel, and it will not appear as a session, a referral or an attributable source. It will appear as a shorter sales conversation.
So ask. Ask every new customer how they came to you, write the answers down, and read them monthly. The answers have started changing, and a business that notices the change six months before its analytics does will make better decisions about where the money goes. Nothing in any dashboard is going to tell you that a customer arrived pre-convinced.
Should you keep paying for search work?
1. Which of your keywords is most exposed to being answered without a click?
2. Your impressions are flat but clicks are falling on question-shaped searches only. What does that suggest?
3. A proposal offers a special AI schema file that will get you into AI answers. What is the right response?
Pick an answer to begin.
Frequently Asked Questions About Whether SEO Is Dead
Is SEO dead in 2026? Not as a whole, and genuinely yes for some businesses. If your revenue came from informational traffic itself, the thing you were buying has largely been absorbed into the results page. If your revenue comes from people who need to compare, price, book or hire, the searches that produce those customers are far less affected.
Should I cancel my SEO retainer? Cancel it if you cannot trace a single enquiry to it over a year, or if the reporting only ever shows rankings and sessions. Change it if the work is aimed at informational traffic that is no longer converting. Keep it if it is producing enquiries, because nothing in the current shift makes that stop working.
Will AI replace SEO? It has replaced one part of it, which is capturing clicks by answering simple questions. The underlying work of being crawlable, indexed, accurate and worth quoting is what feeds the AI answers themselves, so it has become more load-bearing rather than less.
Do I need special schema or an AI file to appear in AI answers? No. Google states there are no additional requirements or special optimisations for its AI features, and specifically no new machine-readable files, AI text files or special schema. A page needs to be indexed and eligible to show a snippet, which is the ordinary requirement.
How do I tell an AI problem from a normal traffic drop? Look at impressions and clicks together over sixteen months, then split by query type. Impressions steady while clicks fall on question-shaped searches is the AI pattern. Both falling together is usually a ranking, indexing or seasonality problem, and Google publishes its own guide to diagnosing those.
Does being cited in an AI answer bring traffic? Barely, on current evidence. Links inside AI summaries were clicked on about 1% of visits in the Pew study. Being cited is worth having as a visibility and credibility signal, and it is not a traffic strategy by itself.
If informational content is losing clicks, should I stop writing? Stop writing the kind that only restates a definition, because that was always the weakest thing on any site and it is now the most exposed. Keep writing the kind that carries something only you have: your pricing, your process, your finished work, your view of the trade-offs.
What This Means for You
Some businesses should stop paying this year, and we would rather say which ones than issue another reassurance. If your traffic was people asking questions and your money came from that traffic, the model has been taken apart. If your reporting has never produced an enquiry, the AI conversation is simply giving you permission to notice.
For everyone else the change is real, measured and narrower than the headlines. The click loss is concentrated in question-shaped searches, the searches that bring you customers are largely intact, and the work that survives is the work that was always worth doing: being findable, being accurate, and being worth quoting. What has changed is that traffic is no longer an acceptable proxy for any of it. Ask your reporting for enquiries instead, and a great deal of this argument resolves itself.
If you want somebody to go through your own keyword list and tell you honestly which half is exposed, that is a normal part of what we do at Web Leveling. We build fast custom sites and run the search engine optimization behind them, and every site we build is owned outright by the business that paid for it. Tell us what is bothering you about your search spending and we will come back within one business day with a straight read, including when the honest answer is that you should spend less.
Terms
The words this argument hides behind
Tap a term to see what it means.
AI Overview. Google's AI-generated summary above some results. It sits inside Google Search, so ordinary Search eligibility is what governs whether your page can be used in it.
Informational query. A search whose whole purpose is an answer, usually long and often starting with a question word. These are the most exposed searches and the ones that lead to a sale least often.
Commercial query. A search from somebody trying to hire, buy, book or compare. Shorter, far less likely to draw a summary, and where nearly all local revenue comes from.
Click-through rate. The share of people shown your page who click it. The number that falls first when an answer appears above you, which is why watching clicks alone is misleading without impressions beside them.
Impressions. How often your page was shown at all. Impressions holding while clicks fall is the signature of being skipped rather than being outranked.
Counterfactual. What a study estimates would have happened without the thing being measured. Ahrefs' 34.5% figure is a comparison against an estimate, not a before-and-after of the same pages.
Vendor study. Research published by a company selling into the market it measured. Useful, directional, and never the same class of evidence as an independent panel study.