Type a question into Google now and there’s a decent chance you never leave the results page. An AI-generated summary sits above everything else, already answering you, already citing sources you didn’t ask for. Pew Research Center found that when an AI Overview appears, users click through to a website far less often than when it doesn’t, based on tracking hundreds of thousands of real searches through mid-2025. That’s not a marginal dip. It’s a structural change in how information travels from the open web to a person’s screen.
For years, ‘best of’ content worked on a simple bargain. You wrote the list, you ranked the options, Google sent you the traffic, and the reader trusted you because you showed up first. That bargain is being renegotiated in real time, and most publishers haven’t noticed yet.
AI Overviews Don’t Link Out. They Absorb.
Here’s the mechanic that changes everything. Traditional search results are a menu. AI Overviews are a meal already plated. Google’s summary pulls fragments from several sources, stitches them into a single answer, and presents it as if no single author wrote it.
Futurism’s reporting on the fallout describes traffic collapses at outlets that used to rely on being the definitive answer to a query. Wired and The Verge, sites with real editorial infrastructure, still got hit. If established newsrooms are losing clicks, a thin affiliate page with a template and a table isn’t surviving on vibes.
The part that should worry anyone running a comparison site: AI Overviews don’t cite the freshest page. They cite the page their model finds most extractable. Clear structure. Direct answers near the top. Verifiable claims. Sloppy, keyword-stuffed lists get skipped over even when they used to rank fine the old-fashioned way.
There’s also a reliability problem baked into the system itself. A widely shared example from The Independent showed Google’s AI Overview behaving inconsistently across two nearly identical political queries, surfacing an answer for one and blocking it for the other. If the summarization layer can’t apply its own rules consistently on a topic with enormous scrutiny, nobody should assume it’s being careful with a niche comparison page nobody’s fact-checking.
Why ‘Best Of’ Pages Are the Canary in This Coal Mine
Comparison and ranking content is exactly the format generative search is best at compressing, and exactly the format that’s easiest to fake. A generic AI Overview can summarize five products from five review sites in one paragraph. It doesn’t need your click if your page just restates specs anyone could find. That’s the trap.
The pages that survive this shift share a pattern. They go deeper than a spec sheet. They compare terms, not just headlines. They update on a real schedule instead of quietly backdating a timestamp. And crucially, they show the kind of comparison depth that a language model can’t fully replicate by skimming ten competing pages at once.
Financial and promotional content is a useful stress test here because the stakes for accuracy are higher than a gadget roundup. A page recommending the best casino bonuses has to account for wagering requirements, expiry windows, game weighting, and jurisdictional rules, details that shift often enough that a lazy or stale list becomes actively misleading within weeks. Pages built to survive AI-driven scrutiny in that category tend to break down real terms side by side rather than just listing headline offer amounts, which is precisely the kind of structured, verifiable comparison that both readers and extraction models reward over generic copy. (For context on how quickly UK-facing gambling content ages out, the responsible gambling guidance from BeGambleAware.org is worth checking alongside any offer page, since terms and player protections both shift regularly.)
The same logic applies whether you’re comparing credit cards, VPNs, or wireless earbuds. If the underlying comparison can be reconstructed from three competitor pages in ten minutes, it’s already vulnerable. If it requires actual testing, actual terms parsing, or access to data a summarizer can’t scrape cleanly, it holds up.
The Content Farm Problem, Turbocharged
Before generative AI, mass-producing thin SEO content took teams of low-paid writers and a lot of templates. That entire model, what’s long been described as a content farm, just got a productivity upgrade. Now one person with an API key can generate five hundred near-identical ‘best of’ pages in an afternoon.
Google knows this. It’s the whole reason AI Overviews exist in the current form: to give users an answer without sending them into a swamp of interchangeable affiliate pages competing for the same fifteen keywords. The irony is that the fix punishes good and bad content in similar ways, at least at first. Traffic drops for the thin page and the thoughtful one. The difference shows up over months, not days, as the model starts preferring sources it can verify and re-cite reliably.
The Register’s coverage of the Pew data put it plainly: AI summaries are suppressing independent web traffic broadly, not selectively. That should reframe how publishers think about ranking pages. The goal isn’t beating a competitor’s list anymore. It’s building something durable enough that a summarization model, and a skeptical human, both trust it on sight.
What Actually Holds Up Under This Kind of Scrutiny
A few patterns keep showing up in the pages that are still pulling traffic through 2026’s search environment.
Specificity survives. A page that says ‘clears in under ten minutes’ beats one that says ‘fast withdrawals.’ Named entities survive. A citation to a real analyst, a real dated report, a real regulatory update reads as verifiable in a way that vague attribution doesn’t. Structural clarity survives. Headers that answer a specific question, tables that let you compare terms directly, FAQ blocks that resolve a query in one paragraph.
None of this is really new advice. It’s what good editorial always looked like. What’s changed is the cost of skipping it. A thin page used to rank fine for years before anyone noticed the quality gap. Now the gap gets exposed within a search cycle or two, because the model doing the comparing is faster and less forgiving than a human skimming ten tabs.
Frequently Asked Questions
Does an AI Overview replace the need to rank on page one? Not entirely, but it changes what ranking is worth. If your page appears in an Overview’s citations, you can still get clicks from readers who want more depth. If you rank page one but never get cited, you may see traffic fall even at a strong position.
Can smaller publishers compete with AI-generated summaries at all? Yes, particularly with content that requires original testing, direct sourcing, or data a model can’t easily reconstruct. Generic explainer content is the most exposed. Original comparison work, first-hand reporting, and updated data hold up far better.
How often should comparison or ‘best of’ pages be updated to stay trustworthy? Quarterly at minimum for anything involving pricing, terms, or promotional offers. Faster-moving categories, like financial products or bonus terms, often need monthly checks since a single outdated figure can undermine the whole page’s credibility.
Is thin AI-generated content actually getting penalized by Google directly? Google’s public stance focuses on quality signals rather than detecting AI authorship specifically. In practice, thin AI-generated pages tend to underperform because they lack the depth, sourcing, and update cadence that both readers and ranking systems now expect.
The List Format Isn’t Dying, It’s Getting Audited
Comparison content isn’t going away. If anything, demand for clear, structured answers is higher than it’s ever been, it’s just that the bar for what counts as a trustworthy answer moved. A list built on genuine comparison work, real sourcing, and regular revision is more valuable now than it was two years ago, not less. A list built to game a keyword is the one quietly disappearing from view.
The publishers who treat this as an editorial problem, not just a technical one, are the ones still getting cited months from now.
