Google just confirmed it: featured snippets didn't die. They became AI Overviews' backup generator.

Executive summary

For two years, the SEO industry treated "featured snippets are dying" as a symptom of AI Overviews taking over. The data says something more specific happened: featured snippets got absorbed. Ahrefs' tracking of 1,000,000 U.S. desktop SERPs found featured-snippet visibility collapsing from 15.41% of tracked queries in January 2025 to 5.53% by June 2025 — a 64% relative drop — while AI Overview visibility rose from 3.93% to 27.43% across the exact same window, a correlation Ahrefs itself describes as "very strong" (Confidence: High — directly fetched primary Ahrefs source, all headline figures confirmed verbatim). By July 2026, a Semrush Sensor-based analysis put U.S. desktop featured-snippet share at just 0.22%, down from 3.39% fifteen months earlier — a 93.5% collapse (Confidence: Medium — third-party blog citing Semrush Sensor, no disclosed methodology).

The part the industry mostly missed: on January 22, 2026, Google's own VP of Engineering for Search, Rajan Patel, posted on X that the company "fall[s] back to featured snippets when we are unable to generate an AI Overview," adding that "this experience isn't as clear as it should be" and that he'd asked his team to fix it (Confidence: High — verbatim post text located directly, matching Search Engine Roundtable's independent reporting word-for-word). That's not industry speculation. That's Google confirming featured snippets never disappeared — they got demoted to invisible plumbing, rendered only when the AI Overview generation pipeline fails.

Meanwhile, the AI Overview rollout itself was not the smooth monotonic climb most coverage implies. Semrush's own primary tracking of 10M+ keywords shows AIO prevalence rising from 6.49% of queries in January 2025 to a peak of 24.61% in July 2025 — then falling back to 15.69% by November 2025 (Confidence: High — directly fetched Semrush's own blog, figures matched exactly). That pullback lines up with mounting pressure: a Pew Research study published July 22, 2025 found users click a traditional result only 8% of the time when an AI summary appears (versus 15% without one) (Confidence: High — primary Pew methodology and figures confirmed directly), and Penske Media Corporation's live antitrust suit against Google over AI Overviews cannibalizing publisher traffic was still active in February 2026 (Confidence: High — corroborated across three independent outlets plus a real CourtListener docket, though the filing itself was relayed via secondary synthesis). Whether that pressure caused the mid-2025 throttle is not something Google has confirmed — it's a plausible read of parallel timelines, not an established causal claim.

A timeline / comparison table

Part 1 — The great swallowing: what Ahrefs' 1,000,000-SERP dataset actually shows

Ahrefs' analysis, drawn from tracking 15 distinct SERP features across a million randomly sampled U.S. desktop results, is the single most rigorous dataset available on this shift. Featured-snippet visibility ran at 15.41% of tracked queries in January 2025. By June 2025 — five months later — it was 5.53%. That's a 64% relative decline (Confidence: High — directly fetched primary source, all figures confirmed verbatim). Over the identical window, AI Overview visibility moved from 3.93% to 27.43% of queries, a roughly sevenfold increase. Ahrefs describes the relationship between the two curves as "very strong," reporting a correlation of approximately 0.9 in magnitude; the commonly cited four-decimal figure (-0.9053) does not actually appear in Ahrefs' own text and should not be treated as confirmed (Confidence: High for the ~0.9 magnitude; the precise decimal is unverified and excluded).

The same dataset shows knowledge panels — the entity-fact boxes on the right rail for people, places, and organizations — sliding more modestly, from 14.09% to 12.95% of SERPs, an 8% decline (Confidence: High). A separate, smaller feature Ahrefs calls "Knowledge Cards" actually grew, from 1.50% to 2.30% of SERPs, a 54% increase (Confidence: High). March 2025 stands out as the clear inflection point in the data: AI Overview appearances jumped 116% that single month, with featured snippets falling in rough lockstep (Confidence: High).

Diagram note for design: A dual-axis line chart plotting featured-snippet visibility (falling, 15.41% → 5.53%) against AI Overview visibility (rising, 3.93% → 27.43%) on the same January–June 2025 x-axis, with March 2025 marked as a vertical annotation line, would be the single most persuasive visual in this piece — it makes the inverse relationship immediate without requiring the reader to hold two numbers in their head.

Part 2 — Google says the quiet part out loud

The industry spent most of 2025 debating whether featured snippets were being "replaced" by AI Overviews or merely "cannibalized" by them — a distinction that matters for how publishers should structure content. On January 22, 2026, Google VP of Engineering for Search Rajan Patel closed that debate from Google's own side. Posting on X, he wrote: "On occasion we fall back to featured snippets when we are unable to generate an AI Overview. This experience isn't as clear as it should be though, I asked the team to improve it." (Confidence: High — verbatim text located directly, matching Search Engine Roundtable's independent reporting word-for-word.)

That single sentence reframes the entire dataset in Part 1. Featured snippets were never simply losing a popularity contest to AI Overviews — they became a dependency of AI Overviews, rendered on the rare query where Google's generation pipeline can't produce a summary. The visibility drop Ahrefs measured isn't snippets dying; it's snippets going invisible-by-default, surfacing only in AIO's failure cases. Patel's own admission that the fallback "isn't as clear as it should be" suggests even Google engineering isn't confident users understand which experience they're looking at when a snippet does appear.

This reframing has a second, quieter data point behind it: Google's formal deprecation of the FAQ rich result. Per Google's own Search Central documentation, FAQ rich results stopped appearing in Search starting May 7, 2026, and Google removed the documentation for the feature entirely by June 15, 2026 (Confidence: High — directly fetched primary Google documentation, dates confirmed). Unlike featured snippets, which Google kept alive as connective tissue, FAQ markup got cut outright — a sign that Google is actively pruning legacy SERP real estate that AI Overviews can absorb natively, rather than maintaining every structured-data-driven feature as a parallel track.

Part 3 — AI Overviews' rollout was a rollercoaster, not a ramp

Most coverage of AI Overviews implies a straight line: launched, then grew, then dominated. Semrush's own primary tracking of over 10 million keywords contradicts that. AIO prevalence moved from 6.49% of tracked queries in January 2025 to a peak of 24.61% in July 2025 — and then fell back to 15.69% by November 2025 (Confidence: High — directly fetched Semrush's own blog, not secondary coverage; figures matched exactly). That's a genuine pullback of roughly a third from peak, not noise.

Two other numbers from the same Semrush dataset complicate the "AI Overviews kill clicks and print money for Google" narrative in different directions. First, the zero-click rate on keywords tracked before versus after they gained an AI Overview actually fell slightly, from 33.75% to 31.53% (Confidence: High) — a counterintuitive result worth sitting with, since it cuts against the assumption that AIOs mechanically increase zero-click behavior on a per-keyword basis. Second, Google Ads co-occurring with AI Overview SERPs grew sharply, from under 1% when Semrush began tracking in March 2025 to 25% of AIO-bearing SERPs by November 2025 (Confidence: High). Note: an earlier commonly cited version of this figure claims a jump from "roughly 3% in January 2025 to 40% by November 2025" — that specific framing does not appear anywhere in Semrush's own primary post and has been corrected here; the verified numbers are under 1% (March 2025) to 25% (November 2025).

A reasonable counter-argument: the July-to-November 2025 pullback in AIO prevalence could simply reflect Google tuning trigger thresholds for quality reasons, unrelated to the publisher and regulatory pressure documented in Part 5. Semrush's data shows the pullback happened; it does not show why. Treating the Pew study and the Penske Media suit as the explanation is a plausible inference from parallel timing, not a causal finding Google has confirmed.

Part 4 — Who actually survives contact with AI Overviews

Not every legacy SERP feature followed featured snippets downward. People Also Ask (PAA) — the expandable question boxes — did the opposite: it co-occurred with AI Overviews in 90.03% of AIO-bearing SERPs by November 2025, and Related Searches co-occurred in 95.32% (Confidence: High — Semrush's own primary data). A separate Semrush Sensor-based analysis covering April 2025 to July 2026 found PAA share growing +15.95% on desktop and +12.83% on mobile, reaching 68.3% and 70.2% of SERPs respectively (Confidence: Medium — third-party blog citing the Semrush Sensor tool, no disclosed sample size or methodology). Reviews-format results grew even faster in that same window: +119% desktop, +32% mobile (Confidence: Medium, same caveat).

The same Sensor-based dataset shows featured snippets falling from 3.39% to 0.22% of desktop SERPs (down 93.5%) and 2.18% to 0.16% on mobile (down 92.7%) over that fifteen-month span, while AI Overviews rose to 42.09% desktop and 50.27% mobile — roughly a 400% increase — and knowledge panels fell 23% desktop and 37% mobile (Confidence: Medium, same third-party caveat). Read against the Ahrefs data in Part 1, the direction is consistent even though the two datasets use different tools, windows, and methodologies — which is itself a form of corroboration, even if neither dataset alone is definitive.

The pattern across both datasets: features that ask a question (PAA) or provide social proof (Reviews) got pulled into the AI Overview experience as supporting material. Features that answered a question directly and completely (featured snippets, FAQ rich results) got replaced outright, because they compete with the AI Overview for the same job.

Part 5 — The cost: fewer clicks, medical safety gaps, and a live antitrust docket

Pew Research Center's July 2025 study is the most methodologically transparent public data on what AI Overviews cost publishers in click volume. Tracking 900 U.S. adults' browsing behavior across 68,879 unique Google searches in March 2025, with search results independently scraped April 7–17, 2025, Pew found users clicked a traditional search-result link in just 8% of visits where an AI summary appeared, versus 15% when it didn't — roughly half the click-through rate. Only 1% of visits clicked a link cited inside the AI summary itself (Confidence: High — primary source, methodology and figures confirmed exactly).

An academic audit published on arXiv in November 2025 (and revised through March 2026) examined 1,508 real baby-care and pregnancy queries and found AI Overviews appearing in 84% of results (1,272 queries) versus featured snippets in just 32.5% (490 queries), with both co-occurring in 22% of cases. Where both appeared on the same SERP, the information was inconsistent between the AI Overview and the featured snippet in 33% of cases — and medical safety disclaimers were present in only 11% of AI Overviews and 7% of featured snippets (Confidence: High — directly fetched full paper, all figures confirmed verbatim). That gap matters specifically for a health-adjacent query category where inconsistent, under-caveated information carries real downstream risk — this is a place where "efficacy or diagnostic claims" concerns apply to Google's own output, not just to health-tech vendors.

On the legal front, Penske Media Corporation — parent of Rolling Stone, Variety, Deadline, The Hollywood Reporter, and Billboard — filed a court memorandum around February 12, 2026 opposing Google's motion to dismiss its antitrust suit, which alleges AI Overviews cannibalize publisher traffic. The case is active under docket 1:25-cv-03192 (Confidence: High — corroborated across three independent named outlets plus a real CourtListener docket entry, though the filing details were relayed via secondary synthesis rather than a directly fetched primary document). This is a real, ongoing case with a named plaintiff and a real docket number — not industry rumor.

What everyone is missing

The "featured snippets are dying" narrative measured the wrong death. Every tracking tool in this piece measured visibility — how often a feature renders on a SERP — and every one of them shows that number collapsing. None of them, until Patel's January 2026 post, could measure whether the underlying mechanism (extracting a direct, structured answer from a well-marked-up page) had actually gone away or had just gone underground. It hadn't gone away. It got demoted to a fallback path that only activates when Google's generative pipeline fails to produce output — which means content once optimized to win a featured snippet is still being evaluated by the same extraction logic, just less visibly and less often.

People Also Ask's growth alongside AI Overviews — not despite them — is the more important signal for content strategy than the snippet collapse. Ninety percent co-occurrence with AI Overviews (Semrush) plus double-digit share growth on both desktop and mobile (structureddata.co.uk/Semrush Sensor) means Google is actively using PAA as a companion surface to the AI Overview, likely because question-format content maps cleanly onto the sub-questions an AIO needs to answer. Content built around clearly delineated question-and-answer structure is being rewarded twice: once inside the AI Overview's synthesis, once in the PAA box beside it.

The Semrush pullback from 24.61% to 15.69% is the most underreported data point in this entire trend, and it should change how teams plan for 2026-2027. An industry that treated AI Overview coverage as an inevitable, one-way ratchet toward 100% of queries built its content strategy on a false assumption. Google throttled its own rollout by roughly a third from peak within four months — for reasons it has not disclosed, though the timing lines up suspiciously well with Pew's click-rate study and Penske Media's antitrust filing. Teams that assumed steady linear AIO growth need to revisit that assumption; the actual pattern is closer to "aggressive rollout, partial regulatory-and-backlash-driven retreat, renewed push" than "monotonic takeover."

The medical-content finding from the baby-care audit is a preview of a coming compliance problem, not a one-off study result. An 11% disclaimer-presence rate inside AI Overviews for pregnancy and infant-care queries, alongside a 33% rate of direct factual inconsistency between the AIO and the featured snippet on the same page, describes a system that is currently under-caveated for at least one health-adjacent vertical. If that pattern holds across other health, legal, or financial query categories — which this dataset alone cannot confirm — it becomes a live regulatory exposure for Google well beyond the Penske Media antitrust matter, and a reason for any brand in a regulated vertical to treat AI Overview visibility as a liability surface, not just an opportunity surface.

Future predictions

• Google will continue narrowing the featured-snippet fallback experience rather than reviving it as a standalone feature — expect further UI changes to make the fallback state visually distinct, following through on Patel's stated intent to "improve" clarity. (Confidence: Medium — based on a single public statement of intent, not a shipped change.)
• More legacy structured-data-dependent rich results will follow FAQ rich results into deprecation over 2026-2027 as Google consolidates presentation logic into AI Overviews. (Confidence: Medium — one confirmed deprecation precedent, extrapolated forward.)
• Expect further non-monotonic swings in AI Overview prevalence rather than a steady climb toward saturation, driven by a mix of quality tuning and external pressure (litigation, regulatory scrutiny, publisher backlash). (Confidence: Medium — based on one confirmed peak-and-pullback cycle; a single data point is a pattern candidate, not a proven cycle.)
• People Also Ask and Related Searches will keep growing as AI Overview companion surfaces rather than declining, since they appear structurally integrated into AIO's answer-generation rather than in competition with it. (Confidence: Medium-High — consistent across two independent datasets, Semrush and Semrush Sensor via structureddata.co.uk.)
• The Penske Media antitrust suit (docket 1:25-cv-03192) is likely to produce discovery that clarifies, for the first time, Google's internal reasoning for the mid-2025 AIO pullback — information currently unavailable publicly. (Confidence: Low-Medium — speculative about litigation outcomes and discovery scope.)
• Health, legal, and financial verticals should expect increased regulatory attention to AI Overview disclaimer and accuracy rates, following the pattern the baby-care audit identified. (Confidence: Low — extrapolated from a single-vertical academic study; no regulatory action has been confirmed as of this writing.)

Practical takeaways

1. Stop optimizing for featured-snippet capture as a primary goal — it is now a fallback state Google renders only when AI Overview generation fails, meaning it's a symptom of AIO absence rather than a target of its own. Structure content so it can be extracted by either mechanism; this is core Technical Foundations work, not a snippet-specific tactic.
2. Treat People Also Ask optimization as a durable, growing priority rather than a legacy tactic — it co-occurs with AI Overviews in 90%+ of cases and is one of the few classic features still expanding. Build question-and-answer content architecture deliberately, which is a core piece of Search Experience Design.
3. Do not assume AI Overview coverage will keep climbing linearly for your query set — Semrush's own data shows a documented pullback from peak. Build tracking that captures month-over-month AIO presence per keyword cluster rather than relying on a single historical snapshot; this is exactly what LLM Visibility Monitoring is built to catch.
4. If you operate in health, legal, or financial content, audit your own pages for factual consistency between what would render in a featured snippet and what an AI Overview might synthesize from the same content — the baby-care study found 33% inconsistency where both appeared together, and that gap is a real content-integrity liability, not just a traffic issue.
5. Reassess FAQ schema markup investment now that Google has formally deprecated the FAQ rich result as of May 2026 — the markup itself may still help entity understanding but will not produce the visual SERP feature it once did. This is a conversation for Entity & Knowledge Architecture, since the underlying structured data still has value beyond the deprecated rich result.
6. Monitor Google statements and Search Central changelogs directly rather than relying on secondary SEO commentary — Patel's fallback confirmation and the FAQ deprecation notice both came from primary channels (X and Google's own docs) months before most industry analysis caught up.