Every B2B site I have been handed in the last decade has the same page on it. The title is some version of “X vs Y”. It runs about 1,800 words. There is a table in the middle with ticks and crosses, and the ticks are arranged so that the company who wrote the page wins.
That page was the best converting page on most of those sites for about fifteen years. It was the middle of the funnel, written down.
I have not watched one of those pages earn a click in months. It still ranks. It is still being read. It is being read by a model, and the model is not sending anyone.
The thesis
Zero-click is not the death of search. It is the death of one stage of the funnel, and it is the middle one. Buyers still start with a question and they still end with a purchase. What they no longer do is visit you while they are deciding between you and three other people.
The decline is not spread evenly, and that is the whole story
Start with the number everyone quotes. SparkToro, using Similarweb clickstream data for January to April 2026, found 68.01% of United States Google searches ended without a click. That was 60.45% in 2024 and 49% in 2019. Rand Fishkin also put AI Overviews in more than 20% of searches, cutting clickthrough by roughly 60% where they appear. Ahrefs, tracking Google referral share across more than 75,000 domains, measured an eight percentage point fall between June 2025 and May 2026, about a 22% drop.
Most people read those numbers as a flat tax. Everything goes down a fifth, so write more content and hold on.
Then look at where the answer box actually went. Semrush tracked more than 600,000 keywords across ten industries on the United States desktop database from November 2025 to April 2026. AI Overview presence on commercial-intent queries grew 71% over those six months. On transactional-intent queries it fell 5%. In finance, commercial-intent presence grew 231.25%.
Commercial intent is not a technical term you can wave past. It is the query class where somebody types “best”, “vs”, “alternatives”, “compared”, “review”. Transactional intent is “buy”, “pricing”, “demo”, “login”. One of those is your middle funnel. The other is your bottom.
So the answer box is not taxing your site evenly. It is expanding fastest at exactly the point where a buyer is choosing between vendors, and retreating at the point where they have already chosen and want to transact. If you average the decline across your whole site you will conclude you have an SEO problem. You have a stage problem, and the stage that took nearly all of the damage is the one that used to produce your pipeline.
The shortlist is now built before you know the account exists
6sense surveyed more than 4,000 B2B buyers across North America, EMEA and APAC for its 2025 Buyer Experience Report. The split between independent research and seller engagement moved from 70/30 to 60/40. Then the finding that should stop you: 94% of buying groups had already ranked a preferred vendor before they made contact with any vendor at all. 77% of purchases went to that preliminary favourite. The pre-engagement favourite wins 80% of deals. And 94% of buyers said they used a large language model to synthesise and organise their research.
Put the two datasets next to each other. The stage where the shortlist gets built is the stage where the click disappeared, and the tool doing the building is a model that never visits.
And the model is not reading the page that ranks
This is the part most marketing teams have not absorbed. Ahrefs ran 15,000 long-tail queries through Brand Radar and checked where the cited URLs sat in Google. Across ChatGPT, Gemini and Copilot, an average of 12% of AI-cited URLs ranked in Google’s top 10 for the same query. ChatGPT in-text citations: 8.0%. Gemini: 8.6%. Copilot: 8.2%. Perplexity was the outlier at 28.6%. Roughly 80% of citations did not rank anywhere for the query they answered.
Your comparison page won the middle of the funnel by ranking. Ranking is now a poor predictor of being quoted. Those are two different retrieval systems with two different tastes, and you have been optimising for one of them for fifteen years.
I stopped asking for the rankings report about a year ago. When I sit down with a marketing team now, the first artefact I ask for is the list of domains cited underneath the answers their buyers get, and almost nobody has one. I have asked for it often enough to know the reaction: a pause, then someone offers me the Search Console export instead. That export measures a channel that has stopped being the one that decides.
What replaces the middle of the funnel
Not more content. Three changes, and they are uncomfortable in a way that content calendars are not.
Comparison content has to stop being self-serving. A model asked to compare four vendors will not lift a table where one vendor wins every row, because that table is not useful for the job it was given. It lifts the source that concedes something. The page that says “if you need X, buy the other one” is the page that gets quoted, and being quoted is now the thing you wanted from ranking.
Your presence has to exist off your own domain. Given that 80% of citations rank nowhere, the sources being pulled are review sites, forums, documentation, analyst write-ups and third-party comparisons. You cannot publish your way to a citation on a domain you control. You can be the vendor those sources have something specific to say about.
The metric has to change from sessions to mentions. If your dashboard counts visits, the middle of your funnel now looks like it has stopped existing. It has not. It has become invisible to the instrument you are holding.
Where this breaks
Four honest limits, and I would not publish this without them.
The 68% figure comes from a clickstream panel, United States only, four months. Panels over-represent some users and under-represent others. The direction has held across seven years of this study, so I trust the trend. I would not build a board slide on the second decimal place.
The Ahrefs overlap result is long-tail queries specifically. Head terms almost certainly overlap more, because head terms are where the well-known sources sit. If your category is three words wide, your mileage differs.
The 6sense numbers are buyers self-reporting after the fact, and people reconstruct their own decisions to look more orderly than they were. Nobody remembers being influenced. Everybody remembers deciding.
And the honest one: Semrush also reports that a visitor arriving from a language model is worth 4.4 times an organic search visitor, which is the statistic every AI search vendor has been quoting all year. It has no published sample size or method, and it has an obvious selection effect. Someone who arrives from a model arrives late, already informed, already close to buying. Of course they convert better. That does not mean the model created the demand, and treating it as proof of channel value is the same error as claiming credit for branded search.
If you sell into formal procurement, or you sell a commodity, or your revenue is renewals, your middle funnel was never on the open internet and none of this touches you.
What I would do on Monday
Tag every landing page as informational, commercial or transactional, then pull twelve months of traffic by tag. If the loss is concentrated in commercial, you have this problem. If it is flat across all three, you have a different one and this post is not it.
Write down the five questions a buyer asks in the four weeks before they shortlist. Ask them in ChatGPT, Gemini, Copilot and Perplexity. Record whether you were named and what was said about you. That is under an hour and it is a baseline.
Read the citations under those answers, not your rankings. Most will not be you and most will not rank. That list is your actual distribution channel now.
Take one comparison page and concede a real case where a competitor is the better choice. Name it. Then watch whether it starts getting quoted.
Put named-in-answer rate on the dashboard next to sessions. A number nobody owns is a number nobody moves.
The page is still there
That “X vs Y” page is still sitting on the site, still ranking, still being read more than it ever was. It is doing the job it was built for. It is comparing two products for a buyer who is deciding.
The buyer just never arrives to say thank you.
I spend most of my week inside companies working out why the pipeline thinned before the traffic did. If that is the month you are having, book a call.