The most repeated number in marketing this year is that search is dying. The most useful number is the one underneath it: across more than a million high-volume keywords, the demand that left the declining terms turned up almost exactly where you would expect it to go. Search is not shrinking so much as rearranging itself, and that distinction changes what you should publish next quarter.
Fractl's analysis with Search Engine Land of 1,010,848 keywords, each drawing at least 10,000 monthly searches across 379 brands and eight verticals, found 29% of those keywords in measurable decline. The ones that grew over the same period carry roughly the same total monthly volume, about 10.31 billion, as the ones that fell, about 10.29 billion. Broad definitional queries are losing volume to AI answers; specific, comparative, local and transactional queries are holding or growing. The response is to rebalance a content plan, not to publish less of it.
What the data actually says
Two independent studies published this year point the same way, which is worth more than either one alone.
Fractl, working with Search Engine Land, pulled Semrush data on 1,010,848 high-volume keywords, each with at least 10,000 monthly searches, spanning 379 brands across eight verticals. Their report is titled "Search Isn't Dying, It's Redistributing", and the finding behind that title is the important part: 29% of the keywords analysed are in measurable decline, and the ones that grew account for approximately the same total monthly search volume as the ones that fell. Aggregate demand is close to flat. What changed is which queries carry it.
Brainlabs analysed 1.35 million keywords across nine UK categories and found head terms in structural decline in seven of the nine, with long-tail queries growing over the same period. Their explanation matches Fractl's: AI tools have trained people to ask longer, more specific questions, so short definitional terms shed volume while specific queries gain it.
For context on how much of this was anticipated, Gartner predicted in 2024 that traditional search engine volume would fall 25% by 2026 as users moved to AI assistants. The observed decline among high-volume terms came in a few points beyond that, but the headline prediction missed the redistribution entirely, which is why "search volume is down 29%" has been quoted far more often than what it means.
The decline is not evenly distributed
The pattern tracks how completely a chatbot can finish the job. Where a full answer is a paragraph of general information, the query is exposed. Where the searcher needs to compare prices, check availability, read reviews of a specific supplier or reach a particular site, it is not.
| Query type | Example | Exposure to AI answers | What to do with it |
|---|---|---|---|
| Definitional and explanatory | "what is conversion rate optimisation" | High. The answer is complete in a paragraph | Keep for citation value, stop judging by clicks |
| Simple how-to | "how to add a favicon" | High. Steps are reproducible without a visit | Consolidate into larger resources |
| Comparison | "webflow vs wordpress for a business site" | Medium. People verify recommendations at the source | Invest. Add primary detail an assistant cannot synthesise |
| Cost and pricing | "website redesign cost 2026" | Medium. Ranges vary enough that people check several sources | Invest. Publish real ranges with the drivers behind them |
| Local and supplier | "web design agency near me" | Low. Navigational and commercial by nature | Protect. This is your commercial base |
| Transactional | "book a website audit" | Low. The destination is the point | Protect |
| Branded | "hyberx pricing" | Low and often growing as AI answers prompt brand lookups | Protect, and measure it as a demand signal |
Fractl's vertical breakdown follows the same logic: FinTech recorded the largest decline at -37.7% and Lifestyle the smallest at -15.2%. Categories where a competent paragraph resolves the question lose the most; categories where people need to look, compare and choose lose the least.
The mistake to avoid: deleting your explainer content
The obvious reaction to a page whose clicks halved while its rankings held is to delete or repurpose it. That is usually wrong, and it is worth being precise about why.
AI Overviews and AI Mode are served from Google's standard search index. A page that answers a definitional query clearly and is indexed is a candidate to be cited in the answer that replaced its click. Delete it and you do not recover the click; you remove yourself from the answer, lose the brand impression, and lose the associated search that often follows. The correct response is to change what you expect from that page, not to remove it.
Judge explainer content on impressions, citation and the branded searches that follow it. Judge comparison, pricing, local and transactional content on clicks and conversions. Applying a single metric across both is how good pages get deleted for the wrong reason.
Before changing anything, sort your top 50 pages into those two groups and look at the last twelve months separately for each. Most sites we look at find their commercial pages are stable or growing and their explainer pages are down 30 to 60% on clicks while impressions hold. That is not a content quality problem and the fix is not a rewrite; it is a measurement and allocation problem.
A rebalancing model for the next two quarters
Take the pages you plan to publish over the next six months as a fixed budget, then reallocate rather than expand. A workable split for a service business:
| Content type | Typical allocation before | Suggested allocation now | Why |
|---|---|---|---|
| Broad explainers and definitions | ~40% | ~15% | Still needed for topical coverage and citation eligibility, but with declining click return |
| Comparisons and decision guides | ~15% | ~30% | People verify an assistant's recommendation before spending money |
| Cost, pricing and benchmark content | ~10% | ~20% | High commercial intent, and numbers need a source the reader can check |
| Original data, teardowns and primary research | ~5% | ~20% | The only category an AI system cannot generate without you |
| Local, service and transactional pages | ~30% | ~15% | Already stable. Maintain rather than expand |
The percentages are a starting point to argue with, not a prescription. The direction is the part that holds: move effort from summarising what is already known toward publishing things that can only come from you, because that is simultaneously the content that keeps earning clicks and the content an AI system has to cite rather than paraphrase. That is the same conclusion we reached from the citation data in what AI search actually cites, arriving from the opposite direction.
How to tell which of your own keywords are declining
Industry averages do not tell you about your account. Run this on your own data:
- Pull 16 months of Search Console query data and compare the last three months with the same three months a year earlier. Use impressions, not clicks, to isolate demand change from click-through change.
- Segment by query shape. Tag each query as definitional, comparison, cost, local, branded or transactional. This is the step most audits skip and it is where the finding lives.
- Look for the split. If definitional impressions are down while commercial impressions are flat, you are seeing redistribution, which is an allocation decision. If everything is down together, you have a visibility problem, which is a different piece of work entirely.
- Check the generative AI report. Search Console's Search generative AI performance reports rolled out to all sites worldwide by the end of August 2026, covering impressions by page, country and date. They show impressions and not clicks, so treat them as a presence signal: are you in the answer at all?
- Watch branded and direct. Rising branded search alongside falling informational clicks is the signature of AI-assisted discovery working in your favour. Falling branded search is the number to worry about.
Repeat it quarterly. The pattern in this data is a trend rather than an event, and a single month's comparison will mostly show you seasonality. If you want a refresher on the underlying method, our keyword research guide covers the mechanics; this is that process with one extra dimension added.
What stays true
Three things in this data cut against the loudest current advice.
- Total demand is roughly flat. People have not stopped searching. The keyword list you built in 2023 has aged, which is a normal thing for a keyword list to do.
- Commercial queries are the resilient ones. The searches closest to spending money are the least exposed, which means the content closest to revenue is the least at risk. That is an unusually comfortable position for a marketing budget to be in.
- Publishing less is the wrong lesson. The growing half of the distribution is longer, more specific and more numerous. Covering it takes more pages, not fewer, but pages with a narrower job each.
The planning question for the rest of 2026 is not whether to keep investing in content. It is whether your next twenty pages are aimed at the half of the distribution that is growing or the half that is shrinking, and answering that takes an afternoon with your own Search Console export.
Frequently asked questions
Is search volume actually declining in 2026?
Partly. Fractl's analysis with Search Engine Land of 1,010,848 keywords with at least 10,000 monthly searches each, across 379 brands and eight verticals, found 29% of those keywords in measurable decline. However, the keywords that grew over the same period carry roughly the same total monthly volume, about 10.31 billion against about 10.29 billion for the declining set, so aggregate demand is close to flat. Demand is redistributing between query types rather than disappearing.
Which types of search queries are losing the most volume?
Broad definitional and simple how-to queries, because an AI answer completes them without a visit. Brainlabs found head terms in structural decline in seven of nine UK categories while long-tail queries grew. Comparison, pricing, local, branded and transactional queries are the most resilient, because the searcher still needs to compare options, check a specific supplier or reach a particular destination.
Which industries are most affected by AI-driven search decline?
Information-heavy categories where a chatbot can answer completely. In Fractl's vertical breakdown, FinTech recorded the largest decline at -37.7% and Lifestyle the smallest at -15.2%. The pattern tracks how much of the decision a paragraph of general information can resolve, rather than anything specific to those industries' marketing.
Should I delete blog posts that lost traffic to AI Overviews?
Usually no. AI Overviews and AI Mode are served from Google's standard search index, so an indexed page that answers a definitional query clearly remains a candidate to be cited in the answer that replaced its click. Deleting it does not recover the click and removes you from the answer. Change the metric you judge that page by, from clicks to impressions, citation and the branded searches that follow, rather than removing it.
How do I find out which of my own keywords are declining?
Export 16 months of Search Console query data and compare the last three months with the same three months a year earlier, using impressions rather than clicks so you measure demand rather than click-through. Tag each query as definitional, comparison, cost, local, branded or transactional. If definitional impressions fall while commercial impressions hold, you are seeing redistribution. If everything falls together, you have a visibility problem instead.
Can Search Console show my AI Overviews performance?
Partly. Google's Search generative AI performance reports rolled out to all sites worldwide by the end of August 2026 and show impressions within generative AI features on Search and Discover, broken down by page, country and date, with a device breakdown on the Search report. They do not include click data, so treat them as a presence signal rather than a traffic measurement.
The takeaway
The 29% figure is real and the conclusion drawn from it usually is not. Demand left broad definitional queries and arrived at specific, comparative and commercial ones, leaving aggregate volume close to flat. So the work is reallocation: fewer explainers, more comparisons, pricing, and primary material only you can publish; keep the explainers you have, because they are still what gets cited; and judge the two groups by different metrics. Run the query-shape split on your own Search Console export before you act on anyone's industry average, including this one.
Sources & further reading
- Search Engine Land, "What 1 million keywords reveal about AI's impact on search"
- PPC Land, "AI cuts overall search demand 29%, FinTech loses 38%, Fractl finds"
- Brainlabs, "AI Overviews and Declining Search Volume: What 1.35M Keywords Reveal"
- Google Search Central, "Introducing Search generative AI performance reports in Search Console"
- Search Console Help, "Generative AI performance report (Search)"