Is SEO dying or changing? What our logs and Google's own documents show

Is SEO dying or changing? What our logs and Google's own documents show

Between 21:19 UTC on 24 September and 06:39 UTC on 25 September 2026, OpenAI's GPTBot fetched 15,285 pages from trustlist.uk. Googlebot fetched 221. That is about 69 pages taken by one AI crawler for every page taken by the crawler that feeds the world's largest search engine. In the same nine hours, 27 page views on our site arrived from a Google search results page.

Over a longer period the picture is the same. In the 92 days to 18 September 2026, Google showed our pages 652,463 times and sent us 657 clicks. Four in five of those clicks, 528, went to pages about one named business or product, not to the broad ranking pages we had built to catch search traffic. If you asked us on the evidence of our own site whether SEO, in the sense we had practised it, is dying, we would have to say that ours was.

The question this article tries to answer is whether that is true of SEO in general, or whether something narrower is dying. We start with what we counted, then look at what Google itself says and documents, then at independent measurement, and end with a method you can apply to your own site. Our own traffic story is told in more detail in what 657 clicks taught us, and the view from the engines' side is in the changing role of search engines.

What we counted on our own site

Two data sets underpin our view, and both have limits worth stating first.

The log sample. It is one overnight window of 9.3 hours in UK time, from our edge proxy, counting HTML page requests only. It is a sample, not a month. In it the site served 77,221 page requests. Browser-like user agents made 35,401 of them (45.8%), but our analytics show much of that is an automated browser farm, so it is not all human. All AI companies' crawlers and fetchers together made 27,591 requests (35.7%): GPTBot 15,285, OAI-SearchBot 5,359, Meta's crawlers 4,519, Applebot 1,747, Anthropic's ClaudeBot 614, and ChatGPT-User, which fetches pages live for a person's question, 66. Bingbot made 2,180 requests and Googlebot 221. We let AI crawlers and SEO tools in deliberately.

Of the 35,401 browser page views, 76.1% carried no referrer, 23.8% came from our own pages and 0.1%, which is 27 views, came from Google search. None came from a social network.

Search Console. Over the 92 days, listing pages (one company or product) collected about 33,000 impressions and 528 clicks, a 1.6% click-through rate at average positions of 6 to 9. Ranking pages, which list many firms for a category or place, collected about 255,000 impressions and 67 clicks, a rate of 0.03% at positions 60 to 80. About a third of the queries we examined in the "no clicks" list, some 310 of 943, were searches by contact scrapers: a person's name, a company name and "phone".

What this shows is specific. Search still sends people to pages that are the best available answer to a precise question, such as "what does this company do". It does not send people to pages that are one of thousands of near-identical lists. Meanwhile, most of the reading of our site is done by machines, and the machines that read most are not the ones that send clicks.

That is our site. The rest of the article asks how far it generalises.

What Google says about AI Overviews and AI Mode

Google's position is easy to find, because it has written it down.

Its Search Central page on AI features says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." To be eligible, a page must be indexed and eligible to be shown in Google Search with a snippet. The usual controls apply: nosnippet, data-nosnippet, max-snippet and noindex limit how content is shown. Traffic from AI features is counted in Search Console's Performance report under the Web search type, not separated out. The page describes AI Mode and AI Overviews using a "query fan-out" technique, issuing several related searches behind the scenes, which Google says can show a wider and more diverse set of links than a traditional search.

That last point matters for anyone who believes SEO is now "optimising for AI". On Google's own account, there is no separate discipline for its AI features. A page that is indexed, eligible for a snippet and useful is eligible; a page that is not, is not.

Google's leadership also makes claims about traffic. In August 2025 Liz Reid, head of Google Search, wrote that total organic click volume from Google Search to websites had been "relatively stable year-over-year", that Google was sending slightly more "quality clicks" than a year before, and that user trends were "shifting traffic to different sites", with some sites losing and others gaining. The post named the kinds of content that were gaining: forums, videos, podcasts and posts with first-hand perspectives, and in-depth reviews, original posts and thoughtful first-person analysis. It published no figures to support the aggregate claim.

At Alphabet's second-quarter 2026 results on 22 July 2026, Sundar Pichai said AI Mode had passed 1 billion monthly active users since its global expansion the previous October, that Google was "sending billions of clicks to websites every week through AI features in Search", and that AI Overviews and AI Mode had been brought together into one Search experience. Alphabet's results release put Google Search and other revenue at $63.271 billion for the quarter, against $54.190 billion a year earlier.

Read together, Google's position is: search use is growing, the business is growing, clicks in aggregate are stable, but they are going to different sites than before.

What independent measurement shows

Google's aggregate claim cannot be checked from outside, because only Google has the aggregate. What can be checked is behaviour on samples, and the best-documented study we found is from Pew Research Center.

Pew analysed the browsing of 900 US adults who agreed to share it, covering March 2025, and 68,879 unique Google searches. About one in five searches (18%) produced an AI summary. When a summary appeared, users clicked a traditional search result in 8% of visits; when none appeared, they did so in 15%. Users clicked a link inside the AI summary itself in 1% of visits. They were also more likely to end their browsing session after a page with a summary (26%) than after one without (16%). The study was published on 22 July 2025 by Athena Chapekis and Anna Lieb.

That is a sample of Americans, from one month, before AI Mode's global rollout, and it measures click rates, not total clicks. It does not contradict Google's claim that total clicks are stable, because more searches at a lower click rate can still produce similar totals. What it does show is that for any given search where an AI summary appears, a website is roughly half as likely to receive a click.

The crawler side has also been measured. Cloudflare, which carries a large share of web traffic, published figures in August 2025 comparing how many pages each platform crawls with how many visitors it refers. For July 2025 it reported OpenAI at about 1,091 crawls per referred visitor and Anthropic at about 38,066, while Google's ratio was in single figures. It also reported that training drove nearly 80% of AI crawler activity. Our 69-to-1 ratio of GPTBot to Googlebot sits comfortably inside that pattern.

So the independent evidence supports a narrower claim than "SEO is dead". Per search, clicks to websites are falling where AI answers appear. The crawlers that read the web most heavily return very few visitors. Neither finding says search has stopped sending people to websites.

What is dying

Here the owner's framing, and our own experience, deserve a direct answer. Some forms of SEO are dying, and one of them was ours.

Volume pages. The model of publishing thousands of pages, each targeting a slight variant of a search, and letting the aggregate carry the site, is failing. It fails twice. First, AI summaries answer many of those variant searches directly, and Pew's data suggests they cut click rates when they do. Second, Google's rules now name the practice. Its spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users", and give as an example using generative AI tools or similar "to generate many pages without adding value for users". The policy does not care how the pages were made. It cares whether they help anyone.

Thin programmatic pages. We were one. About 24,350 of our live listings, a quarter, carry fewer than 50 words of profile text, and 12,178 companies have none. On 25 September 2026 Google AdSense declined the site for "low-value content". Our 18,333 ranking pages sat at positions 60 to 80. Nothing in our data suggests that making more such pages would have helped.

Borrowed authority. Google's site reputation abuse policy covers third-party content published on a host site "mainly because of that host's already-established ranking signals". Its expired domain abuse policy covers buying an old domain to host low-value content. Both were once standard tactics. Both are now named violations.

Stale catalogues. 9,620 of our live companies were last updated at the original import in December 2020. Of 1,000 company listings drawn at random on 24 September 2026, 171 websites no longer led to the business as listed. Google's helpful content guidance warns against changing dates to seem fresh without real updates. A page that has not changed in six years and describes a business that has closed is not something any ranking system should reward, and ours were not rewarded.

What is not dying

The same evidence points the other way for a different kind of work.

Being the primary source. Our clicks went to pages about one named business or product, at positions 6 to 9. For a search about a specific thing, a page that is actually about that thing, and that says something the summary cannot say without citing it, still gets chosen. Google's own list of the content gaining traffic, first-hand perspectives and original analysis, is a list of primary sources.

Original data. Google's helpful content page asks whether content provides "original information, reporting, research, or analysis" and "substantial value when compared to other pages in search results". A number you counted yourself is the clearest case of that. When our news desk recorded that the median notice period across 28 vendor changes in September 2026 was 31 days, with six at a week or less, nobody else had that figure. An AI answer that wants to state it has to take it from us.

Technical eligibility. Because Google says AI features require only indexing and snippet eligibility, the basics are not obsolete. A page blocked by robots.txt, marked noindex or failing to render is excluded from AI answers as well as from blue links.

Search as a channel. The engines are not shrinking. Alphabet reports Search revenue up about 17% year on year in the second quarter of 2026, and query growth driven by AI features. Other answer engines crawl heavily enough that ours appear in our logs every minute. People are asking more questions, not fewer. They are just being answered in more places.

So our answer to "is SEO dying?" is no, not as a whole. The practice of earning visibility for pages that deserve it is changing. The practice of earning visibility for pages that do not deserve it is dying, and it deserves to.

What changes in practice

If the goal shifts from "rank many pages" to "be the source an answer depends on", several habits change.

• Measure clicks per page type, not impressions per site. Impressions include scraper queries and results on page eight. Clicks by page type tell you what is working.
• Read your server logs. Analytics tools are built to count people. Logs show who is actually reading, and on our site that is mostly machines.
• Decide on crawler access deliberately. OpenAI documents three separate agents: GPTBot for training, OAI-SearchBot for surfacing sites in ChatGPT search, and ChatGPT-User for live user requests. Blocking one does not block the others. Google documents Google-Extended as a separate control for some uses of content outside Search. Choose on purpose rather than by default.
• Write fewer pages, each of which could be cited. Ask of every page: is there a sentence here that another site would have to credit us for?
• Treat staleness as a defect. Dates, prices and websites decay. Build checks for them.

A method you can apply

This is the method we are now following, reduced to steps anyone can use.

1. Export 90 days of Search Console data by page. Group pages by type (product, category, article, location, and so on). For each type, total impressions, clicks, click-through rate and average position.
2. Remove noise from the query list. Delete queries that only a scraper would make, such as a name plus "phone" or "email". Delete queries where your position is worse than 50; they are not reaching people.
3. Find the type that earns clicks. In our case, single-entity pages at 1.6% against 0.03% for lists. Your winner is where your effort goes.
4. Count thin pages. Pick a threshold, such as 50 words of original text, and count pages under it. If it is more than a small share, fix or remove those pages before creating new ones.
5. Sample your outbound facts. Take 1,000 random pages, follow their key links and check their key facts (prices, dates, addresses). Record the failure rate.
6. Rebuild from primary sources. For each page you keep, read the subject's own material and write in your own words what it does and what can be checked. Leave out anything you cannot support.
7. Retire and remove visibly. If a subject has closed, say so. If you cannot find it at all, remove the page.
8. Publish one original number a month. It can be small. It must be counted by you, with the method stated.
9. Read one night of logs. Count requests by user agent and referrer. Decide which crawlers you want and set robots.txt accordingly.
10. Review in a quarter. Compare clicks per page type with the baseline from step 1. If nothing moved, say so and change the plan.

We have done steps 1 to 5 and 9, and started 6 to 8. We will publish the step 10 comparison when a quarter has passed.

Sources

• TrustList edge proxy logs, 21:19 UTC 24 September to 06:39 UTC 25 September 2026: page requests by user agent and referrer. TrustList internal data, one overnight window, a sample not a month.
• TrustList Google Search Console, 92 days to 18 September 2026: impressions, clicks, positions by page type, and query gaps. TrustList internal data, read 25 September 2026.
• TrustList catalogue and news desk counts, 20 to 25 September 2026: thin listings, last-updated dates, dead-website sample, vendor notice periods. TrustList internal data.
• AI features and your website: eligibility for AI Overviews and AI Mode, no special optimisation, controls, query fan-out, Search Console reporting. Google Search Central, last updated 10 December 2025.
• Spam policies for Google web search: scaled content abuse, site reputation abuse and expired domain abuse. Google Search Central, last updated 28 August 2026.
• Creating helpful, reliable, people-first content: self-assessment questions on original information and substantial value; warning about changing dates. Google Search Central, last updated 10 December 2025.
• AI in Search is driving more queries and higher quality clicks: Google's statement on stable total click volume and shifting traffic. Liz Reid, Google, 6 August 2025.
• Alphabet Q2 2026 earnings call: Sundar Pichai remarks: AI Mode users, clicks through AI features, combined AI Overviews and AI Mode. Google, 22 July 2026.
• Alphabet Announces Second Quarter 2026 Results: Google Search and other revenue, Q2 2026 and Q2 2025. Alphabet, Form 8-K exhibit 99.1, 22 July 2026.
• Google users are less likely to click on links when an AI summary appears in the results: click rates with and without AI summaries, session endings, share of searches with a summary. Pew Research Center, 22 July 2025.
• The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals: crawl-to-referral ratios by platform and the training share of AI crawling, January to July 2025. Cloudflare, 29 August 2025.
• Overview of OpenAI crawlers: the purposes of GPTBot, OAI-SearchBot and ChatGPT-User. OpenAI (read 25 September 2026).