AI Overviews in 2026: what actually changed in search

The new top of the funnel

The mistake is to treat AI Overviews as another SERP feature. They aren't. They are the first thing a user reads, and for a growing share of queries they are the only thing users read before deciding whether to click, call, or move on.

In our tracked accounts, click-through from queries with an AI Overview present has fallen 30–55% for informational intent, and 8–15% for transactional intent. The interesting number is the second one: transactional queries still convert, but only if you are one of the cited sources — because those are the links the panel exposes as its 'sources' drawer.

So the funnel now looks like this: query → AI Overview → cited sources → your site. If you are not in the citation set, you are not in the market for that query. Rankings still matter, but they no longer guarantee visibility on the queries where AI answers dominate.

For local businesses this hits harder than most realise. When a homeowner in Round Rock searches 'why is my AC blowing warm air,' the AI Overview answers the question and then lists two or three service pages as sources. If the local HVAC shop isn't in that source set, the phone doesn't ring — even if it ranks #2 organically below the panel.

What earns a citation

We reverse-engineered citations across roughly 4,200 queries in Q1 2026. Three signals dominated. First, the source page had to directly answer the exact question — not adjacent to it, not eventually — in the first 60–120 words of body copy. Second, the page had to carry unambiguous entity signals: named authors, Organization schema, sameAs links to well-known profiles, and internal links using clean, descriptive anchor text. Third, the page had to be part of a topical cluster, not a one-off article.

The counter-intuitive finding: domain authority mattered less than expected. Small sites with tight topic authority earned citations against national publishers, provided the answer structure was correct. A twelve-page automotive site outranked a 40,000-page publisher on a diagnostic query because the small site answered the exact question in paragraph one and had five supporting pages linking into the answer. This is why our engagements start with an information-architecture audit before we touch a single word of content.

There's a fourth signal that showed up in the last two months and looks like it's becoming permanent: freshness that is actually visible. A 'Last updated' date in body copy, refreshed statistics, and a genuine content diff seem to boost citation persistence through model updates. Ghost-updating the date field without changing anything does not.

The disappearance problem

The hardest conversation to have with a client is that their #1 organic ranking now sits below a summary that already answered the user. The traffic doesn't fall to zero — it falls to the residual of users who want to verify, compare, or transact.

Two practical responses. One: shift measurement to assisted conversions and branded-search lift rather than raw sessions, because the AI Overview is often the first exposure. If branded search grows 20% while organic sessions dip 8%, you are winning, not losing. Two: rewrite the top-of-page copy to serve the verifier and the transactor, not the researcher — the researcher has already been served upstream by the panel.

The pages that suffer worst are 'ultimate guide' pieces built for the researcher persona. The pages that survive are the ones that assume the reader already knows the basics and now wants specifics: pricing, comparisons, edge cases, local availability, and trust signals. We've rewritten roughly 60 pages this way in the last six months; average conversion rate on those pages is up 34% even where sessions are flat.

Measurement in an opaque channel

You cannot see AI-Overview impressions the way you can see organic impressions. What you can see: query-level referrer patterns, share of citations across a curated query set (we track 200–500 per client), and downstream branded-search lift within 7–14 days of citation.

We publish this in a dedicated tab in the client dashboard alongside classical rankings. The point isn't to replace ranking data; it's to make the second-order effects legible so budget conversations can happen honestly.

The tool stack we use is deliberately boring: Google Search Console for query-level truth, a simple citation-tracker script that hits target queries weekly with a headless browser, and Looker Studio to stitch it together. There are commercial GEO trackers now; most are early and most cost more than the DIY stack while providing similar signal.

A working AEO/GEO playbook

The playbook we ship is five moves. Build an entity graph on the About and Author pages. Restructure top-of-page copy to answer the exact query in the first 100 words. Publish comparison and 'X vs Y' assets in the categories where AI Overviews are most aggressive. Add FAQPage, HowTo, or Product schema where it maps cleanly to the content. Track a curated query set weekly and adjust.

None of this is exotic. What is different is the order and the discipline. Skip the entity graph and the citations don't stick; skip the top-of-page rewrite and the citations you earn don't send traffic; skip the weekly tracking and you find out about a citation loss three months late.

For our Austin local clients we add a sixth move: make sure the Google Business Profile matches the on-site entity data exactly. AI panels increasingly cross-reference GBP data with on-site schema, and mismatched hours, categories, or service areas seem to reduce the odds of local citation. This costs nothing and takes an afternoon.

Rewriting for verifiers, not researchers

The single most valuable exercise you can do this quarter is take your top ten organic-traffic pages and rewrite the first 200 words as if the reader has already read a summary of the topic. Cut the definition. Cut the history. Get to the specifics — the pricing, the timeline, the edge case, the local nuance.

The reason this works is that the users who click through from an AI Overview have already been oriented. They don't need a primer; they need the thing the primer didn't cover. Serving them a primer they've already read is the fastest way to bounce them back to the SERP.

We keep a rewrite checklist taped above the desk: does paragraph one answer a question the summary didn't? Does paragraph two show something a summary can't — a screenshot, a table, a live quote? Does paragraph three make it obvious how to take the next step? If all three answers are yes, the page tends to hold traffic even as AI Overviews expand.

What comes next

Two live experiments we're running: measuring citation persistence across model updates (early data says entity-rich pages hold citations 2.3× longer through model swaps) and testing whether structured comparison tables earn citations more reliably than prose. Early answer to the second: yes, when the table is the answer to the query, not just a supporting artefact.

If you're planning 2026 SEO budget, assume the AI-Overview citation is now the equivalent of the featured snippet in 2018 — a channel most teams underinvest in for a year, then scramble to catch up. The teams that ship the entity graph, the top-of-page rewrites, and the tracking cadence this quarter will be the cited sources on their competitors' queries by Q4.

The teams that treat AI Overviews as somebody else's problem for another six months will be having very different conversations at their year-end planning offsite.