Oct 29, 2026: Google is testing “query refinement” bubbles at the top of AI Overviews, but the click doesn’t behave like a normal refinement. Instead of adjusting the search terms, it routes the user into AI Mode for the same query set.
That sounds like a tiny UI test. It isn’t. It’s a traffic re-routing mechanism. If this sticks, more people will stay inside Google’s AI experience, and fewer will reach your site—unless you’re one of the sources the AI Mode response chooses to cite.
What changed (and why this is different)
AI Overviews already reduce clicks when they answer the question outright. The new part is the “refinement” affordance. Users see bubble suggestions that imply “tap to narrow the query,” which historically meant: new SERP, new blue links, new chance for you to win the click.
In this test, the bubble is more like an on-ramp. Click it and you’re in AI Mode—an experience designed to keep the session going inside Google. The biggest implication: the “second query” (the refinement) becomes another AI-first interaction, not a fresh web-results moment.
Primary source: Search Engine Roundtable’s report on AI Overview query refinement bubbles.
What it actually means for business owners
Expect more zero-click behaviour, but in a slightly different shape than the first wave of AI Overviews.
Here’s the shift we care about: the “browse loop” used to bounce people between related queries and web pages. The new loop bounces people between related queries and AI Mode responses. If your acquisition relies on informational content (how-tos, comparisons, “best X for Y,” troubleshooting, definitions), you’re now competing with an interface that can satisfy curiosity without ever leaving Google.
There’s a second-order effect most teams will miss: your analytics won’t just show fewer visits. It will show fewer assist visits—those early research sessions that later become a branded search, a demo request, or a store visit. If the top-of-funnel is siphoned into AI Mode, your brand has fewer chances to be “seen” during consideration.
Winners, losers, and the new premium on being cited
Winners: brands with clear entity signals (real organisation pages, consistent authorship, strong About/Contact, reviews), and pages that are easy for AI systems to quote: specific, structured, and unambiguous. If you already earn citations in AI features, this can consolidate your advantage.
Losers: content farms, generic “ultimate guides,” and thin affiliate pages. Also: legitimate SMEs with good services but weak digital PR footprints. If your brand isn’t well understood, you’re less likely to be selected as a supporting source when Google answers inside AI Mode.
Google has also been telegraphing that quick-win AI spam tactics won’t hold. John Mueller’s warning is basically: enjoy it while it lasts. Reference: Mueller on Google eventually catching AI spam and “tricks”. Our translation: don’t build a strategy on formatting gimmicks that exist purely to bait AI citations.
Our point of view: this is UX, not “SEO news”
Most commentary will frame this as another AI Overviews experiment. We see it as a SERP UX test that changes user intent mid-stream.
Those bubbles look like a promise: “refine your query.” The actual result is: “stay in AI Mode.” That’s a psychological nudge, not an algorithm change. And UX nudges often ship faster than ranking systems do.
One more wrinkle: Google has openly said AI Mode ads monetize worse than classic search ads. That doesn’t mean AI Mode goes away; it means Google has an incentive to make AI Mode sessions longer and more valuable over time. See: Google’s statement that AI Mode ads earn less than traditional search ads. If the company needs AI Mode to pay for itself, expect more experiments that increase engagement without requiring extra queries.
What this means for your site (actions for this month)
- Audit pages that currently “earn the click” from informational queries. If your top landing pages are glossary-style or generic explainers, rewrite them to include decisive, quote-ready specifics: thresholds, steps, caveats, and named tools. AI Mode likes content it can confidently cite.
- Build a citation-targeted section on priority pages. We usually recommend a tight block near the top: a 2–4 sentence answer, followed by a short table or numbered steps. Not for “featured snippet” nostalgia—because it increases the chance the AI response pulls your wording or references your page.
- Strengthen entity and trust signals across the whole domain. Make your About page and author pages real. Add a physical location if applicable, and ensure contact methods match GBP and socials. AI Mode selection is less forgiving of anonymous or inconsistent brands.
- Invest in digital PR that creates independent references. If AI Mode is the new middle layer, third-party mentions become more valuable. This is where PR and SEO stop being separate line items.
- Track “citation visibility,” not just rankings. Rankings won’t tell you whether you’re being used as a source in AI interfaces. Start monitoring AI Overview/AI Mode appearances for your head terms and documenting which competitors get cited.
What we’re changing for clients this week (effort vs impact)
1) Rebuild 5–10 money pages to be citation-friendly (1–2 weeks): not a full replatform, not a blog binge. We take the pages that already convert and improve how easily they can be quoted: tighter definitions, clearer claims, original diagrams where it helps, and a “how to choose” section that’s specific to your market.
2) Create an internal linking spine (2–4 days): AI systems and humans both need a clear map. We add a small number of highly intentional links between the “definition” content and the “transactional” content so that if you do earn the visit, it doesn’t dead-end.
3) Technical cleanup that reduces ambiguity (3–7 days): canonicals, duplicate near-identical location pages, inconsistent schema, mixed author bylines. None of this is sexy. It’s the stuff that stops you being treated like a content mill.
If you want a reference point for the kind of execution we mean, start from our latest posts and briefs and compare the structure to the average competitor page in your niche. Most are still writing for 2019.
If you do nothing for 90 days
Expect two things: (1) a slow drop in non-branded organic sessions on “research” queries, and (2) a weaker pipeline later because fewer people encountered your brand during consideration.
You might not notice immediately if paid is propping up demand. But when your CPCs rise or budgets tighten, you’ll look for “free” search traffic and find the SERP is now an AI-first funnel where only a few sources get referenced.
This isn’t panic-worthy. It is priority-worthy.
Where BrandJunction fits (and what to review)
If your growth plan depends on organic search, this is a good month to tighten your fundamentals and stop publishing content that merely “ranks” but doesn’t deserve to be cited. Our teams do this work day-to-day across US SEO retainers and UK SEO programmes, and it increasingly overlaps with information architecture and performance work on modern sites.
If your site is slow, messy, or difficult to parse, fixing content alone won’t cut it. That’s where our US web builds for marketing teams come in—because AI-first SERPs reward clarity and speed, not CMS chaos.
Want the other tests we’re watching that could change click behaviour? Check the BrandJunction trend watch.
Next step: pick 10 queries where you used to rely on “refinement” clicks (people searching, then narrowing). We’ll tell you which pages should be rebuilt first to compete for AI Mode citations.
Primary source: Search Engine Roundtable
