Key takeaways
- AI Overviews don't just rank pages — they select passages. Optimize self-contained sections that answer one question cleanly, not only whole pages.
- Google's AI breaks a query into many sub-questions ("query fan-out") and rewards content deep enough to answer several at once.
- One page is rarely enough. AI Overviews favour claims corroborated across your own site and the wider web — consensus beats a single confident assertion.
- Being the named citation is what protects your click when the answer sits above your link. Sites summarized but not cited are the ones that lose traffic.
- Track presence deliberately. Appearing in an Overview isn't the same as ranking #1, and citations shift week to week.
AI Overviews vs traditional rankings: what changed
For two decades, Google SEO was a contest to be one of ten blue links. AI Overviews — the generated answer that now sits at the top of many results — and the fuller conversational AI Mode changed the unit of competition. Google no longer just ranks pages and hands you the click. It reads the strongest results, extracts the parts that answer the question, writes a synthesized answer, and cites a handful of sources beside it. You're no longer only competing to rank a page. You're competing to have a passage of your page pulled into the answer.
The two goals overlap but aren't identical. A page can sit fourth in the organic list and still be the source Google quotes in the Overview, because that one section answered the sub-question better than the pages above it. That's the shift worth internalizing: relevance is increasingly judged at the passage level, not only the page level, and the winner is often the clearest few sentences rather than the highest-authority domain.
The query fan-out model and why it rewards depth
AI Mode doesn't treat your search as a single query. It quietly expands it into a cluster of related sub-questions — a process Google has described as "query fan-out" — runs several searches in the background, and assembles the answer from whatever sources best cover each thread. Someone typing "best video production company in Toronto" implicitly triggers sub-searches about pricing, turnaround, portfolio quality, in-house versus freelance teams, and reviews. The AI resolves all of them at once.
This is why thin, single-angle content struggles while genuinely thorough pages win. A page that answers one narrow question can only satisfy one branch of the fan-out. A page that anticipates the natural follow-ups — cost, timeline, comparison, objections, local specifics — can be cited across several branches of the same answer. Depth here isn't padding; it's coverage of more sub-questions.
How to write for the fan-out
- Map the real follow-up questions a buyer asks after the headline one, and give each its own clearly labelled section.
- Cover the awkward questions competitors avoid — price ranges, what's not included, when your service is the wrong fit.
- Add the local and the specific: figures in CAD, GTA context, 2026 timelines. Specifics get quoted; generalities get skipped.
- Keep one idea per section so the fan-out can grab exactly the branch it needs.
Passage-level optimization: winning the snippet inside the answer
Because selection often happens at the passage level, the practical craft is writing sections that still make sense when lifted out of the page entirely. If a paragraph only reads correctly after three paragraphs of setup, it's a poor candidate for extraction. The goal is self-contained answers.
What a quotable passage looks like
- Answer-first structure. Lead with the direct answer, then explain. Don't bury the payoff under context.
- A heading that mirrors the question. A heading phrased the way a person asks ("How much does a website cost in Toronto?") signals exactly which sub-question the section resolves.
- Concrete numbers and ranges. "Most SMB websites in the GTA land between CAD $5,000 and $25,000" is extractable; "it depends on your needs" is not.
- Tight lists and short definitional sentences. The AI often lifts a clean one- or two-sentence definition close to verbatim.
If you want the deeper mechanics of writing for machine readers rather than only human ones, our primer on what generative engine optimization actually is goes further, and it pairs well with the technical discipline our web development and SEO team applies to structuring pages for extraction.
Corroboration and consensus: why one page is never enough
Generative answers are cautious by design. Google would rather cite a claim that several independent sources agree on than a bold assertion sitting on a single page. So even a well-written passage can be passed over if nothing else on the web — or on your own site — backs it up. Corroboration functions as a selection signal, not just a nicety.
In practice this means two things. First, your own site should reinforce its core claims across multiple pages: a service page, a case study, a pricing page, and a blog post that all describe what you do and what it costs the same way create an internal consensus the AI can trust. Contradicting yourself — three different price ranges across three pages — actively hurts you. Second, off-site consensus still matters: reviews, directory listings, mentions on other reputable sites, and consistent business details all feed the picture.
The AI isn't hunting for the most confident page. It's looking for the claim it can stand behind — the one echoed by other sources it already trusts.
This is also the clearest line between the old game and the new one. For the fuller comparison, our breakdown of GEO vs. SEO covers where the two disciplines overlap and where corroboration changes the strategy.
Content formats AI Overviews pull from most often
Not all content is equally liftable. Across the GTA client sites we work on, a few formats show up in AI Overviews more often than their share of total pages would suggest — because their structure maps neatly onto how the AI wants to quote.
- Direct-answer FAQ blocks. A real question as a heading, followed by a two-to-four-sentence answer, is close to the ideal extractable unit. This is the single highest-leverage format.
- Comparison and "X vs. Y" sections. Fan-out queries constantly ask which option is better; structured comparisons get pulled straight in.
- Step-by-step how-tos with numbered steps. Ordered lists are quoted closely for procedural questions.
- Definition-style openers. A crisp "[Term] is …" sentence at the top of a section is prime citation material.
- Pricing breakdowns with real ranges. Concrete cost bands answer a sub-question most competitors dodge, so they tend to get cited disproportionately.
The through-line: structure that isolates one answer per block. Long, flowing narrative can be lovely to read, but it gives the AI nothing clean to grab.
Technical signals: crawlability, freshness and structured data
None of the above helps if Google can't cleanly read the page. The technical baseline for AI Overviews is the same solid ground that supports ordinary SEO — it just matters more here, because the AI is parsing your content, not merely indexing it.
The signals that carry weight
- Crawlability and clean HTML. Content rendered only after heavy JavaScript, or buried in tabs and accordions a crawler skips, may never be seen. Server-rendered, semantic HTML wins.
- Structured data (schema). FAQPage, Article, Product, and LocalBusiness markup help Google understand what each block is. It doesn't guarantee a citation, but it removes ambiguity and makes eligible content easier to select.
- Freshness. AI answers lean toward recently updated sources for anything time-sensitive. A visible "updated 2026" date and genuinely current figures help; stale pages quietly fall out of Overviews.
- Page speed and Core Web Vitals. Slow, unstable pages are read less favourably across the board.
This is unglamorous plumbing, but it's where a lot of otherwise-good content loses. A well-structured site with clean markup is, in effect, pre-formatted for extraction.
Protecting traffic when the answer appears above your link
The honest worry: if Google answers the question at the top, does anyone still click? Sometimes no — and that's real. But the picture is more nuanced than "AI Overviews killed traffic." Being the cited source is precisely what protects you, because a named citation earns the click an uncredited summary would have absorbed. The businesses that get hurt are the ones summarized but never named.
How to stay clickable
- Give people a reason to go deeper. Put the summary-level answer in the Overview-friendly passage, but keep the tools, calculators, full examples, templates, and hard specifics on the page. The AI hands over the gist; you own the detail.
- Compete for commercial and navigational queries, not just informational ones. "Best web design company near me" ends in a decision, and those clicks still convert.
- Strengthen brand. When your name is the one cited, searchers increasingly click through to vet the source they'll actually buy from.
- Treat the Overview as top-of-funnel. Its job is to introduce you; your page's job is to close.
The wrong move is to withhold content out of fear it'll be summarized. Thin pages don't get cited at all — they lose both the Overview and the ranking.
Tracking your presence in AI Overviews
You can't manage what you don't measure, and AI Overviews are genuinely harder to track than blue-link rankings. Presence shifts week to week, varies by phrasing, and often won't surface in standard rank-tracking dashboards. A practical approach in 2026 looks like this:
- Manual spot-checks. Run your priority queries in a logged-out or incognito session and record whether an Overview appears, whether you're cited, and which passage was pulled. Do it on a fixed cadence — monthly at minimum.
- AI-visibility tracking tools. A growing set of platforms now monitor citations across AI Overviews and assistants. They're imperfect, but far faster than checking by hand at scale.
- Search Console context. Watch for pages with steady or rising impressions but softening click-through — often the fingerprint of an Overview answering the query above your link.
- Referral traffic from assistants. Segment visits arriving from AI surfaces to see what's actually converting.
Be realistic about the goalposts. Appearing in an Overview isn't the same as ranking #1, citations come and go as Google reshuffles sources, and a disappearance often just means a competitor got corroborated more strongly that week — not that you did anything wrong.
If this is starting to feel like a second discipline layered on top of SEO, that's a fair read — it largely is. It's the work our team folds into every engagement: structuring content for passage-level extraction, building the on-site and off-site corroboration the AI trusts, and tracking where you actually show up. If you'd rather have a partner own that end to end, that's exactly the kind of thing Arctec AI is built to handle.