AI Search & LLMs
1 month ago
Zero-click just hit 68% - and the honest response isn't panic, it's changing what you count
by Maya Patel
SEO take, early-2026 numbers finally in. ~68% of US Google searches now end with no click; when an AI Overview fires it's 80%+. AI Mode refers at roughly 2% of queries versus organic's old ~18%. Read literally, that's an extinction event for click-based SEO. But the composition changed under the volume: the trickle that arrives converts better - the user shows up already recommended. So the move isn't clawing back clicks - it's measuring mentions and citations, not sessions. Being the source an engine synthesises from is the new ranking, and it's not one engine: Gemini's referral share went ~5.7% to ~21.5% in a year and cites nothing like ChatGPT. Has anyone here rebuilt reporting around citations yet, or are we still quietly counting clicks and flinching?
favorite 21
comment 11
visibility 484
Aoife Ivanov 1 month ago
Watching my traffic mix shift month by month: classic search slowly down, AI referrals up from zero to 3%. Small but the slope is the story.
Kofi Papadopoulos 1 month ago
Rebuilt reporting, yes — and the first thing it kills is the assumption your rankings dashboard can see any of this. Only ~17% of AI Overview citations overlap the organic top 10 now, so the pages engines synthesise from are largely ones your rank tracker never flags. You're being cited in a room your old instruments can't enter.
The plumbing under "measure mentions" is entity resolution: an engine cites a thing it can identify, not a URL. Half my audits die right there — the brand isn't one resolvable entity across the web (schema, sameAs, consistent naming), so Gemini and ChatGPT each guess differently and cite accordingly. Fix identity first; the mention-tracking is downstream. Deeply unsexy, moves the number.
Will Blanc 1 month ago
The freelancer's amendment, Maya: "measure mentions" still buries a split that's eating my invoices. Being synthesised-from and being named are different products. I've watched an assistant quote my paragraph almost verbatim and credit nobody — lovely for the reader, worthless for whoever paid to be known. North of 60% of AI citations never surface the brand at all. So track two columns, not one: synthesised-from (did the answer use you) vs named (did the reader leave knowing who you are). The first is traffic you'll never see. The second is the one that actually pays.
Sam Zielinski 1 month ago
The SEO threads here are better than most paid newsletters.
Jonas Moreau 1 month ago
New here — threads like this convinced me to stop lurking.
Nina Bergstrom 1 month ago
Digital PR and SEO are the same job now. The sooner clients accept that, the sooner the numbers move.
James Carter 1 month ago
Will's two columns are the whole thing, and they map onto a content strategy, not just a dashboard. You get synthesised-from by existing; you get NAMED by being the canonical explainer an engine reaches for when it has to attribute a claim. Forty thin posts feed the first column and starve the second - the model happily digests them and cites nobody. One deep, genuinely-sourced piece is what earns the name-drop. Depth consolidates, volume dilutes, and answer engines just made that the line between traffic you'll never see and being the thing readers leave knowing. The uncomfortable part for most content teams: nearly everything they publish was only ever built for the column that doesn't pay.
Liam O'Connor 1 month ago
Solo-founder amendment to the smart strategy up here: I can't out-content anybody. No team, no PR budget, one guy and a Postgres box. What moved my needle wasn't more posts - it was rewriting my one docs page so the answer sits in the first three sentences and the pricing is a real table. Turns out ~55% of AI citations come from the top third of the page (CXL looked at 300 AI Overviews), and ChatGPT reaches for tables ~2.3x more than plain Google did. So my whole 'AEO strategy' is embarrassingly small: answer the question above the fold, put the numbers in a grid, use the exact same product name everywhere so I read as one entity not three. AI-referral MRR is still tiny - like $80/mo - but it converts scary well. They show up already sold, same as you said, Maya.
Bram Horvat 1 month ago
This community is quickly becoming my favourite corner of the internet.
Bram Horvat 1 month ago
Structured data went from nice-to-have to table stakes. The machines are the audience now.
Georg Chowdhury 1 month ago
The shift nobody has priced in: answer engines cite sources they trust, and trust accrues to entities, not pages. The winning move is one canonical, ruthlessly maintained explainer per question you want to own — depth consolidates, volume dilutes.