**TL;DR** — Across 28 client sites in March-April 2026 we logged how often Google AI Overviews surface YouTube video chips inside the answer card. Video chips appear on 38% of commercial-intent queries and 61% of how-to queries. Click-through to the cited video is small (~2.4%), but the citation itself reshapes the surrounding answer text in ways that matter for brand presence. Three structural choices determine whether your video gets pulled in: a chaptered transcript with timestamps, a clear question-answer pair inside the first 30 seconds, and a YouTube description that mirrors the schema of a companion blog post. None of this is news to YouTube SEO veterans — what is new is that AI Overviews now reads all of it.
Why video chips started mattering this quarter
Through 2025 the AI Overviews video chip felt like a beta. Chips appeared inconsistently, the cited videos were often unrelated to the query, and most teams reasonably ignored them. That changed in February 2026 when Google quietly raised the prevalence — across our basket of 60 commercial queries, video-chip presence went from 9% in January to 38% in April. The composer is now treating YouTube as a peer source rather than a fallback, and the chip has stabilised enough that the question "does my video get cited?" has a real answer instead of a coin flip.
The interesting part is not the chip itself — it is that the surrounding answer text now references what the video says. We diffed 200 AI Overviews answers before and after the video chip appeared on the same query, and the textual answer changed substantively in 71% of cases. The composer is reading the transcript, lifting a phrase or fact, and weaving it into the prose answer. That means even a video that nobody clicks is shaping how your category gets described in the answer card. For a category leader that is leverage; for a challenger whose video is missing, it is silent erosion.
How we ran the audit
Twenty-eight client sites — split roughly evenly across SaaS, DTC, B2B services and one media client — each with at least one active YouTube channel. The basket was 60 queries per client, weighted 40% commercial, 30% how-to, 30% comparison. We ran the queries weekly through Google Search with a clean Chrome profile and AI Overviews opt-in, captured the answer card with a Playwright script, and parsed the video chip URL plus the surrounding 800 characters of text. The capture takes about 12 minutes per client per week; we keep the raw HTML in S3 because the answer text changes from week to week and the diff is the analysis.
Two normalisation moves made the data trustworthy. First, we ran each query from three geographies (US-East, EU-West, AP-South) because video chip prevalence varies by locale — the US numbers are roughly 10 percentage points higher than the AP ones, and reporting only the US figure would over-state the global picture. Second, we discounted any query where the cited video was older than 24 months — the composer increasingly prefers fresh video, and an old citation is a different signal from a current one. The numbers above are after both filters.
What gets pulled in
The first structural lever is a chaptered transcript. Videos with publisher-provided chapters that cover the query intent inside the first three chapters were cited 4.1× more often than videos with auto-generated transcripts and no chapters. The composer reads the chapter title as a headline and the chapter transcript as the body — exactly the same shape it reads HTML for. Writing chapters in question form ("How does pricing work?", "What integrations are supported?") matched cited-video patterns roughly 60% of the time in our sample; writing them as nouns ("Pricing", "Integrations") matched 22%.
The second lever is a clear question-answer pair inside the first 30 seconds of the video. We pulled the cited timestamp range for every video chip in the audit and 78% of citations landed in the first 90 seconds; 41% landed in the first 30 seconds. The composer is reading early — if the answer to the query is buried at minute 4, even a chaptered video tends to get skipped in favour of one that gets to the point fast. The pattern matches the same passage-level citability we have written about for HTML content; the composer does not care that it is a video, it just wants the answer near the top.
The third lever is description-to-schema mirroring. Videos whose descriptions repeat the same key facts and link to a companion blog post with matching Article schema were cited at roughly twice the rate of videos with sparse descriptions. The composer is using the description as a confidence signal — if the same fact appears in the video transcript, the description, and the linked blog post's schema, the citation likelihood compounds. None of those individual signals is decisive, but the stack is.
The click-through is small — and that is not the point
Across the audit, video chip CTR averaged 2.4% — well below text-link CTR for the same queries. Treating that as a reason to ignore the chip is the same mistake teams made about Knowledge Panels in 2018: the value is not the click, it is the entity presence inside the answer. When your video gets cited, your brand name shows up in the answer card; the composer often paraphrases your phrasing in the surrounding text; competitor Knowledge-Panel-adjacent prose gets pushed below the fold. We measured a 9% lift in branded search impressions on weeks where a client's video was cited in at least 25% of the basket queries — branded search is the downstream pickup for being the source the AI summarised from.
The harder pattern to see in dashboards is the answer-text change we mentioned earlier. The composer often lifts a phrase from the cited video transcript verbatim — and once that phrase is in the answer card, it is what users read on every subsequent visit, regardless of whether they click. We have started tracking "answer-text share" as a quarterly metric for video-heavy clients: how many of the basket queries return an answer card whose text quotes or paraphrases something from a client video. It is the AI-era analog of brand mention share, and the click-through underweights it badly.
What changed in our YouTube checklist
Three additions. We require publisher-provided chapters on every commercial-intent video, written as questions matching the keyword cluster — auto-chapters are not enough because YouTube's auto-chapter titles tend toward generic nouns and lose the question-form match. We require the answer to the implicit query in the first 30 seconds of every video, even at the cost of a less-elegant intro — the engagement metric that matters is no longer watch time, it is composer reachability. And we require the YouTube description to repeat the three highest-leverage facts from a companion blog post that has Article schema — the description is the bridge that lets the composer treat video and post as the same entity.
We also dropped one habit. Through 2024 we treated the YouTube chapter title as a place to be clever; in 2026 we treat it as a place to be literal. "How does Slack pricing work in 2026" beats "Slack pricing, decoded" by every metric we can measure. The cleverness costs citations, and citations are now the channel.
- 01Add publisher-provided chapters to every commercial-intent video, written as questions matching your keyword cluster. Auto-chapters cite at roughly a quarter of the rate.
- 02Make sure the answer to the implicit query lives inside the first 30 seconds of the video. 41% of cited timestamps landed there in our sample.
- 03Mirror the three highest-leverage facts between the YouTube description, the transcript, and the schema of a companion blog post. The composer reads the stack as an entity confidence signal.
- 04Track "answer-text share" — how many basket queries return an answer card that quotes or paraphrases your video — alongside CTR. CTR understates the value by an order of magnitude.
Where this argument breaks
For sites with no YouTube footprint and no near-term capacity to build one, the work above is still optional — the absolute click-through is small enough that an under-resourced video program will lose to under-resourced HTML work. For categories where Google has not yet enabled video chips (most legal queries, most regulated finance, most healthcare), the prevalence numbers are zero or near-zero and the audit does not pay yet. The Chinese-language picture is different again: 百度 video integration into 文心 answers is on a different schedule and the structural levers do not transfer cleanly. Outside those carve-outs, video chips are now a load-bearing slot inside AI Overviews — and the team that treats YouTube as a 2024-era distribution channel rather than a 2026-era citation source will keep being surprised by how often their category gets summarised in someone else's voice.