AI overview optimisation and the quiet arithmetic of a badge

What Google Preferred Sources actually changed#

Two lines of code landed on Google's own site on 20 August 2026, and by teatime every SEO desk in the northern hemisphere had already renamed the news cycle. The Google Preferred Sources button lets a reader tell Google, in one tap, which publisher it should keep favouring: a script tag and a single div, embedded on a publisher's own pages, asking a reader to mark the site as preferred for Top Stories and, for anyone who has opted into it, AI Mode and AI Overviews.

For a newsroom that has spent two years watching the click drain quietly out of search, that is a genuinely useful lever. It is also, on its own, nowhere near enough ai overview optimisation to bank a strategy on, and this piece would rather say that plainly on day one than let a client discover it the hard way in month three. Real ai overview optimisation is a discipline, not a checkbox, and the rest of this piece treats it that way.

The mechanics are almost insultingly simple, and news of the button moved through the SEO hedgerow faster than Google's own changelog did. TechCrunch reported that Google shipped the embeddable button on the Thursday, following its earlier May rollout of source selection inside AI Mode and AI Overviews themselves, and that as of that May launch people across the web had already selected more than 345,000 unique sources through the tool. By the week of this article, coverage of the 20 August announcement was already citing a fresher figure again: PYMNTS reported more than 600,000 sources claimed, and repeated Google's own line that readers are twice as likely to click through to a source once they have marked it preferred.

Not every page qualifies, and the fine print matters more than the fanfare. Google's own developer documentation is specific: only domain-level and subdomain-level sites are eligible to appear in the source preferences tool, so a subdirectory such as a client's own /blog/ cannot be claimed on its own, only the domain or subdomain that hosts it. That single sentence quietly settles an argument every content team eventually has: the badge belongs to the brand, not to any one section of it, so whoever owns the root domain owns the button.

It is worth being precise about what the badge changes and what it does not, because the two get conflated constantly. Being claimed as a preferred source does not touch eligibility for citation inside an AI Overview at all: per Google's own AI features guidance, publishers do not need to create new machine readable files, AI text files or markup to appear in these features, and there is no special schema.org structured data the button requires either. The badge is a popularity signal layered on top of ordinary eligibility, not a backstage pass around it.

Why the badge exists: the Pew arithmetic underneath it#

Google Preferred Sources did not appear in a vacuum, it appeared as damage control for exactly the gap ai overview optimisation exists to close, and the sharpest scent of the damage comes from outside Google altogether. The Pew Research Center tracked 900 US adults through a browsing app on its KnowledgePanel across March 2025, capturing 68,879 unique Google searches, of which 12,593 produced an AI summary, then independently collected the matching search results between 7 and 17 April 2025 to see exactly what each searcher had actually been shown.

The headline numbers are stark and worth sitting with. Searchers clicked a traditional link in just 8% of visits when an AI summary appeared, against 15% when it did not, roughly half the rate. Only 1% of visits clicked a link inside the summary itself, and a searcher was markedly more likely to end the whole session there and then: 26% of AI-summary visits ended the browsing session outright, against 16% without one. Some 18% of all searches in Pew's sample produced a summary at all, rising to 60% for question-phrased searches, close to the shape a B2B buyer's own research journey tends to take.

Pew's numbers are not an outlier. A randomised field experiment run by researchers at the Indian School of Business and Carnegie Mellon University's Heinz College, posted to SSRN and revised in June 2026, tracked 1,065 US desktop searchers through a custom browser extension and found that outbound organic clicks fell by 39.8% whenever an AI Overview appeared, while zero-click searches, the ones that end without a single outbound click anywhere, rose by 34.5%. Two independent methods, a survey panel and a randomised experiment, land on the same story: the summary is not stealing a slice of the click, it is eating most of the meal.

Google's badge versus Perplexity's bargain#

Google is not the only platform rebuilding its relationship with publishers this year, and the contrast is instructive. Where Google offers a free badge and a promise of more clicks, Digiday reported that Perplexity's Comet Plus tier pays publishers 80% of a pooled $42.5 million fund, split across three channels: direct visits through the Comet browser, citations inside search answers, and content used to complete a task by Comet's own AI agent. Blavity, Der Spiegel, Fortune, Gannett, The Independent and Time are named participants in the wider publisher programme.

That difference is the whole of perplexity seo in miniature: a discipline built on getting paid per use rather than getting clicked more often. Google's badge is a popularity contest with no purse; Perplexity's programme chases the same quarry down a different trail, a royalty statement with no popularity contest attached. Neither replaces the other, and a publisher chasing both is not being greedy, it is correctly reading that visibility now runs on at least two separate currencies.

Both are still small next to the scale of the problem described above, but a publisher building deliberately for both is playing a longer, steadier game than one that claims a single badge and calls the quarter's ai overview optimisation work done.

The unglamorous work of ai overview optimisation#

Here is the uncomfortable arithmetic worth putting in front of every content-heavy B2B client this week: claim the badge, because it costs ten minutes and nothing else, and then stop treating it as the strategy. Ten minutes of embed work cannot cut a path through the thicket an 8-versus-15 gap represents. The actual work of ai overview optimisation is slower, duller, and considerably harder to screenshot for a slide.

How do you optimise website for ai search, not just Google's version of it#

Building the boring half of this is what the SEO and GEO service line at folkfox has spent the past year doing, and none of it is exotic. Google's own helpful content guidance still asks the same question it always has: was the page written for a person or for a crawler? An AI summary lifts a sentence, not a section, so every claim on a page needs to survive being read entirely alone, stripped of the paragraph built up around it.

The habit that actually moves an ai overview citation is unglamorous: turning a claim that sounds confident into a claim a machine can safely repeat. A summary lifts a sentence, not a paragraph of throat-clearing around it, so the sentence has to survive completely alone. That same discipline is what wins a citation in ai mode search too, not just in a classic AI Overview: the surface changes, the sentence-level test does not.

None of this is exotic, but all five moves compound into a working ai overview optimisation programme: folkfox's five-move framework for AI Overviews from earlier this month sets out the mechanics in full, and the same logic applies whether the citing engine is Google's or somebody else's.

Measuring citation, and folkfox's honest read#

The scale of what the badge is up against is worth stating without flinching. Condé Nast's chief executive Roger Lynch told Search Engine Journal that in each of the last three years his teams budgeted for a search traffic decline, and every year the real fall came in worse than they had forecast; he now expects search to settle at a single-digit share of total traffic. AdExchanger's reporting puts numbers to smaller newsrooms too: Business Insider lost 55% of its organic search traffic between April 2022 and 2025, HuffPost lost half its search referrals over the same stretch, and the New York Times saw search's share of its own traffic fall from 44% to 37%.

None of that collapses because a badge exists. What changes is what gets measured. Search Console's own performance report still separates impressions from clicks, and that split is the whole diagnostic: if impressions hold while clicks fall, the page is being summarised, not demoted, and no badge fixes a demotion that never happened.

This is where folkfox lands on it, plainly. Claim the Google Preferred Sources badge, because it is free, it takes ten minutes, and there is no honest reason to skip it. But do not mistake claiming a badge for having a strategy: Pew's own arithmetic, 8% against 15%, is the size of the problem a two-line embed is being asked to solve, and it cannot.

The actual fix is the unglamorous kind, and it is what ai overview optimisation looks like sentence by sentence: structured data on every page that deserves it, claims built to survive alone, sources named where a machine can find them. Every one of those habits eventually leads back to the den, your own site, your own schema, your own sourced sentences, not somebody else's button.

Healthcare brands read this the same way fintech and cybersecurity brands do: the citation, not the click, is now the unit that matters, a theme traced across answer engine optimisation for healthcare and, more broadly, across how five different instruments measure ai search traffic without ever quite agreeing on the number. Add a fresh spam update to the same month, and the honest answer is that nobody gets to coast on one clever button in 2026.

Frequently asked questions#