Google AI Overviews in 2026: What Changed for Agencies | Recon
ARTICLE
Google AI Overviews in 2026: What Changed for Agencies
How Google AI Overviews evolved from 2024 to 2026 — citation patterns, traffic-loss data, and what agencies should tell clients about ranking strategy now.
Google AI Overviews launched generally in May 2024, expanded internationally through 2025, and by early 2026 appear on roughly 60% of informational queries in English-language US results. They're no longer an experiment — they're the default search experience for any query Google classifies as informational.
This changes the client conversation about SEO. "Will my page rank #1?" is now followed by "And will it be cited in the AI Overview?" The answers can be different — a #1 organic result that isn't cited often gets less click-through than a #4 result that is.
This post is what's actually known about AI Overviews in 2026: how they pick sources, what's changed since launch, and what agencies should tell clients.
A "show all" expansion revealing additional sources Google considered
A follow-up suggestion bar with related queries the user can ask
Below the AI Overview, traditional organic results continue as before
Roughly 60% of informational queries now show an AI Overview. Transactional queries ("buy X", "X price") and navigational queries ("Facebook login") rarely show one. Local-pack queries usually don't, but the "things to know" or "considerations" cards above the local pack are AI-generated and use similar source-selection logic.
In 2024, AI Overviews typically cited 2-4 sources per answer. In 2026 the average is 3-7, with some "show all" expansions exposing 12-15 considered sources. More citations means more pages have a chance to be cited per query — but it also means each citation drives less individual traffic (the click-through is split across more sources).
Early 2024 AI Overviews cited unreliable sources (Reddit, low-quality content farms) frequently and embarrassingly. Google adjusted aggressively through 2024-2025. In 2026, the source mix is dominated by:
Established publications (NYTimes, The Verge, Wirecutter)
Domain-expert sites (Stack Overflow for programming, MDN for web development, AMA-published medical content)
Government and academic sources
Wikipedia (still the single most-cited source overall)
User-generated content is cited less often. Marketing-page citations are rare. Independent expert blogs are cited consistently when their content is structured for retrieval.
AI Overviews increasingly source content from pages that already answered the related-question variants. A page ranking for "what is CLS" plus "how to fix CLS" plus "CLS threshold" gets cited more often than a page ranking only for "what is CLS" — the multi-question coverage signals comprehensive expertise.
This rewards content depth over content volume. A single 2500-word post answering 6 related questions outperforms 6 separate 500-word posts each answering one question.
The 2024 estimates of "AI Overviews will tank your traffic" turned out to be roughly half-right. The actual data from 12 months of agency tracking:
Pages cited in the AI Overview retain or grow click-through (the citation drives qualified clicks for users who want depth beyond the summary)
Pages ranking #1-3 organically but NOT cited lose 15-30% click-through (the AI Overview answers their query before the user scrolls)
Pages ranking #4+ and not cited see negligible change — the user wasn't going to click that far down anyway
The strategic implication: ranking #1 isn't enough anymore. The real goal is being cited, which often correlates with but isn't identical to ranking high.
If the query is "what is X", the AI Overview will pull from pages that define X clearly in the opening. Storytelling intros, anecdote-led posts, and "Welcome to our blog" preambles get less citation rank.
Comparison and how-to queries pull preferentially from pages with explicit list structure. A 5-step process written as a numbered list is cited more often than the same process written as flowing prose.
For "how to" and "should I" queries, embedded Q&A pairs get cited at unusual rates. The format:
## Should I use WebP or AVIF for hero images?WebP for broader browser support, AVIF for the fastest possible LCP. WebP works in 96% of browsers as of 2026; AVIF works in 89%. The size difference is meaningful — AVIF is roughly 80% smaller than JPEG, WebP is 60-80% smaller — but for a hero image where every millisecond counts and you can use the `<picture>` element to serve both, the answer is "use both."
The H2 question gets cited as the question; the answer paragraph gets cited as the answer. Use sparingly — 3-5 per long article is the bar.
Time-sensitive queries (anything about a current product, current standard, or current data) heavily preference pages with dateModified within the last 12-18 months. Pages without Article or BlogPosting schema, or with stale dates, are systematically downranked for these queries.
Technical, medical, financial, and legal queries strongly preference pages with named authors who have demonstrable credentials. Person schema with worksFor, hasCredential, and alumniOf helps. A bio with bar admission, medical license, or relevant publications matters more than schema alone.
The KPI shift is from organic ranking position to citation presence. Track:
Which queries show an AI Overview
Which queries cite the client's pages
Click-through rate for cited vs. ranked-only pages
Most rank-tracking tools added AI Overview presence and citation tracking in 2024-2025. SEMrush, Ahrefs, and most enterprise rank trackers now report this. If a tool doesn't show it, replace the tool.
New content should be structured for citation from the outset. Briefs should specify:
The primary question the post answers (not just the keyword)
3-5 related questions the post also answers
The opening 200 words (which contain the definitional answer)
The structure (sections, lists, embedded Q&A)
The author and their relevant credentials
The expected citation surface (which queries the post should be cited for)
This is more rigorous than the 2024 brief, which often was just "target this keyword, write 1500 words." The new brief is closer to a structured outline with retrieval architecture baked in.
Heavy AI-Overview-defeating tactics. Pages designed to evade or block AI Overview retrieval (paywalls, anti-scraping headers, JavaScript-rendered content) lose more traffic from being uncited than they gain from forcing clicks.
Anti-AI-Overview blog posts. "Why AI Overviews are killing SEO" content has a 7-day shelf life and signals nothing useful to clients.
Trying to beat Wikipedia. For queries where Wikipedia is the dominant citation, focus on adjacent or specialized topics rather than competing head-on.
Multimodal AI Overviews. Image and video citations are starting to appear in 2026. Pages with strong alt-text, structured image data, and video transcripts will increasingly be cited for visual queries.
Local-aware AI Overviews. Local-intent queries are starting to get AI Overview treatment with citation of local-business sites. Schema-quality and review-velocity signals likely transfer.
Search Generative Experience expansion. The full generative-search experience (interactive follow-ups, conversational refinement) is still rolling out. Pages structured for follow-up coverage (the Q&A pattern) will benefit.
The agency that adapts client strategy to citation-first thinking now is positioned for whichever of these lands first. The agency still pitching "rank #1" by 2027 will have a harder conversation.
