Google Discover is not a search result you rank for — it is a personalized feed of cards Google assembles from indexed web content and creator video, matched to what it thinks each person is interested in right now. There is no query to optimize against and no position to climb to; a card either gets surfaced to a given user or it does not, and Google says outright that appearing in Discover is less predictable than ranking in Search. That makes "video optimization for Discover" a genuinely different discipline from video SEO, and it got sharper in September 2026 when Google VP Robby Stein announced a test of AI-generated topic overviews that starts on video cards — a summary layer that sits in front of the feed and links out to related stories. This guide is the practitioner's read on how a video actually reaches Discover in the first place (the two real paths: native YouTube and Shorts, and a video embedded on an indexed watch page you own), the signals the feed rewards that ordinary search does not weight as heavily — freshness, a strong 16:9 hero image, a headline that earns curiosity without clickbait, E-E-A-T, and topical consistency — the technical requirements Google publishes for video eligibility, what the topic-overviews test means for whether your video gets seen or summarized past, and the production reality that decides whether you can package video to Discover's spec often enough to matter.
The most common mistake in "Google Discover video optimization" is importing the mental model of search. In Search there is a query, a ranking, and a position you try to occupy; you win by being the best answer to a term someone typed. Discover has none of that. It is a personalized feed that Google assembles from indexed web content and creator video, matched to what it infers each individual user is interested in at that moment. Nobody searched. There is no keyword to target and no rank to climb — a card is either surfaced to a given person or it is not, and the same video can be shown to one user and never to another with different interests.
Google is unusually candid about the consequence: it states in its own documentation that Discover traffic is less predictable and less dependable than keyword-driven search, and that fluctuations come from shifting user interests, content-type adjustments, and Search updates. That is not a warning to ignore; it is the defining property of the surface. You cannot engineer a guaranteed placement, because there is no placement to engineer. What you can do is make a video maximally eligible and maximally matchable — meet the technical bar, then package it so the recommender has every reason to show it to the right person. This is the same distribution-by-relevance logic that now governs the social feeds, covered in social media discoverability beyond followers; Discover is Google's version of it.
A video reaches Discover in one of two concrete ways, and they call for different work. The first is native platform video — overwhelmingly YouTube, including YouTube Shorts, which Google surfaces in Discover as video cards and dedicated Shorts carousels tuned to a user's interests. If your video lives on YouTube, its Discover eligibility rides on YouTube's own signals and on Google understanding the topic; you are not optimizing a web page, you are producing a strong, on-topic video with a clean title and thumbnail and letting Google's cross-surface recommendation pick it up. Google has also said it is expanding the kinds of creator content in Discover, including posts from other platforms, so the native-video door is widening rather than narrowing.
The second path is a video embedded on a page you own and control — a blog post, an article, or a dedicated watch page with the video in it. Google finds these videos through standard HTML: a video, iframe, embed, or object element in the rendered HTML. For this path the page itself has to earn Discover eligibility (indexed, meeting content policies, good page experience) and the video has to meet Google's video requirements — the watch page should feature that single video as its primary content, the video must not be paywalled, and it needs a stable, high-quality thumbnail. This path gives you more control over the packaging Discover rewards (the hero image, the headline, the surrounding text) because it is your page, which is why it matters even for creators whose primary video home is YouTube.
Because Discover matches interest rather than answering a query, its important levers are tilted away from classic SEO. Freshness is the clearest example. Discover is organized around current interests, so it leans toward recently published content far more than evergreen search does — Google's guidance describes qualifying content as timely for current interests, and in practice a strong recent piece routinely outperforms an older, more authoritative one that no longer maps to what people care about today. This does not bar evergreen video, and refreshing a good older piece can help, but it does make publishing cadence a genuine ranking-adjacent lever: the more recent, on-topic candidates you give the feed, the more chances it has to surface one.
The visual is the second big divergence. Every Discover card is an image card, so the hero image is not decoration — it is a primary factor in whether the card surfaces at all. Google's published advice is specific: use a high-quality image at least 1200 pixels wide, enable max-image-preview:large so Google can render a large preview, and prefer a 16:9 aspect ratio that matches the card format. Generic stock imagery and text-heavy graphics underperform original, compelling images that read at a glance. For video specifically, the thumbnail carries the same weight, and Google gives you multiple ways to set a stable one — the poster attribute, VideoObject's thumbnailUrl, a video sitemap, or og:video:image. A video with weak or missing imagery is one the feed tends to skip regardless of how good the content is.
The third divergence is the headline. In Search the title competes on a results page against other results for a known query; in Discover it has to stop a thumb mid-scroll for someone who was not looking for it. Google's guidance threads a specific needle here: titles and headlines should capture the essence of the content and spark interest, but must not use clickbait, sensationalism, or exaggerated details to manufacture engagement. Curiosity that the content pays off is rewarded; a headline that over-promises is exactly the pattern Google's Discover policies target. Underneath all of this sits E-E-A-T and topical authority — Google evaluates whether the content is credible and whether it clearly understands what your site or channel is about, because a clear topic is what lets it match you to the right interest cluster. The AI-search version of this credibility problem is covered in SEO in the age of AI search.
On September 18, 2026, Robby Stein, Google's VP of Product for Search, announced that Google is testing a "Dive deeper" button in Discover that opens an AI-generated topic overview: a short summary paired with prominent links to related stories, community reactions, and original reporting, carrying a label noting it was generated with AI. Crucially, Google said it would test the experience on videos first and experiment with multiple designs in the following weeks. It extends the AI previews Google began showing on trending Discover topics in October 2025, turning a card-level preview into an on-demand, topic-level summary. The full detail of the announcement is in Google's Discover topic-overviews test.
For anyone optimizing video for Discover, the test cuts two ways. The risk is familiar from AI Overviews in Search: a summary that sits in front of the feed can satisfy an interested reader who then never taps through to any single video — the same click-through erosion publishers already flag. The opportunity is that the overview links out to related stories, community reactions, and original reporting, so a video that is one of a topic's credible, well-covered pieces can earn exposure inside the summary rather than being bypassed. That rewards owning a topic with real depth across formats rather than posting one clip and stopping. It is an early, limited test with no rollout timeline and multiple designs in flight, so plan for the direction — a summary layer in front of your video — without rebuilding a strategy around a single screenshot.
For the watch-page path, the eligibility work is concrete and worth doing carefully. Make sure Google can actually find the video: it must appear in the rendered HTML through a video, iframe, embed, or object element, not hidden behind a click or a fragment that only loads on interaction — confirm it with the URL inspection tool. The watch page should feature that single video as its primary content rather than burying it among unrelated text, and it must not be behind a paywall. Provide VideoObject structured data with a unique name, description, and thumbnailUrl, keep that metadata consistent with the actual video, and let Google fetch the video bytes from a stable URL so features like key moments and previews can work.
Then layer on the Discover-specific packaging. Enable max-image-preview:large sitewide, supply a hero image at least 1200 pixels wide in a 16:9 frame and declare it through og:image or schema, and give the video a stable, high-quality thumbnail in a supported format. Get page experience right — fast mobile load and healthy Core Web Vitals are part of Discover eligibility, not just search. None of this guarantees appearance; Google is explicit that eligibility is not appearance. What it does is remove every technical reason the feed would have to pass your video over, which is the only part of the equation you fully control. For the YouTube-native path, the equivalent work is a strong title and thumbnail and a channel with a clear, consistent topic — see YouTube AI video understanding for how Google now parses video content itself.
Read the levers back to back and a pattern emerges. Freshness rewards regular publishing. The image factor demands an original, compelling 1200px hero for every piece. The headline has to earn a tap without crossing into clickbait, every time. Topical authority requires a consistent subject so Google can classify you. The topic-overviews test rewards covering a subject across video and supporting written formats rather than posting one clip. Every one of those is satisfied by the same underlying capability: producing a steady stream of on-topic, well-packaged video with strong original imagery and supporting text — and starving on the same constraint, which is that a solo creator or small team cannot manufacture that package often enough by hand to keep the feed supplied with fresh, matchable candidates.
This is the honest reframe of "how do I optimize video for Discover." You do not tune a single asset until it ranks, because there is no ranking. You build a practice that repeatedly ships video Discover is eligible to surface — recent, on a coherent topic, with a hero image and thumbnail that meet Google's spec, headlined to earn a tap, on pages that pass the technical bar — and you give the feed enough of those candidates that being matched stops depending on luck. The measurable failure mode is not a bad video; it is an empty pipeline, where you package one piece perfectly, it may or may not get picked up, and then nothing new arrives for weeks.
Kompozy is an AI content generation and multi-platform publishing engine, and its fit for Discover is precise: the feed rewards a specific package — fresh, on-topic video, a strong 16:9 hero image, honest headlines, supporting written depth, published consistently — and that package is exactly what a small team cannot assemble repeatedly by hand. Kompozy manufactures the whole package from one input. It generates the video Discover's test is starting on — Clipped Shorts cut from long-form and captioned Persona Shorts fronted by a face-locked avatar for the native video path — and it generates the original imagery the image-card factor demands: Photo Posts and Infographic Photo scenes you can use as the 1200px 16:9 hero and the video thumbnail, rendered on-brand rather than pulled from generic stock, which is precisely the difference between a card that surfaces and one that does not.
The watch-page path is where the written outputs earn their place. Because a video embedded on an indexed page you own is a real route into Discover — and the topic-overviews test rewards covering a subject across formats — Kompozy also produces the Blog Articles that give a video a proper indexed home with surrounding text and a declared hero image, plus the newsletters and social posts that build the topical footprint an overview links out to. The Persona Brief governs voice and subject so a week of output reads as one coherent topic, which is the account-level signal that lets Google classify you and match you to the right interest cluster, and HyperFrames keeps every hero image, card, and thumbnail pixel-consistent so a high volume of Discover-spec assets looks deliberate rather than scattered. The full range of what it publishes across the output buckets reaches the channels you own as well as the feeds you rent.
Cadence — the freshness lever Discover weights so heavily — is Autopilot, which schedules and publishes the whole set across the eight social platforms plus blog and email from one queue behind a per-post review gate, so the feed is fed a steady supply of recent, on-topic, well-packaged candidates instead of one hand-built piece every few weeks. The boundary is worth stating plainly, because Discover invites overclaiming: Kompozy cannot submit a video to Discover, cannot guarantee the recommender surfaces it, and does not control Google's indexing or the topic-overviews test — no tool can, because there is no slot to buy and Google says eligibility is not appearance. What it removes is the production tax that makes consistent, Discover-spec video packaging impossible to sustain by hand, which is the specific reason most creators optimize one asset perfectly and then let the pipeline run dry. For the format mechanics that turn one source into that full package, see content repurposing.
Two paths. The first is native platform video — most commonly YouTube and YouTube Shorts, which Google surfaces in Discover as video cards and Shorts carousels matched to a user's interests. The second is a video embedded on an indexed page on your own site: Google finds it through the standard HTML elements (a video, iframe, embed, or object tag), and if the watch page is indexed and meets Discover's content policies it can appear as a card. In both cases there is no submission and no ranking position — Discover is a recommendation feed, so a video becomes eligible and then is surfaced, or not, to each user based on predicted interest. Google states plainly that Discover traffic is less predictable than keyword search.
Yes, importantly. Video SEO is about ranking a watch page for queries — you target keywords, earn a position, and capture the people who search that term. Discover has no query. It is a feed Google assembles from a user's inferred interests, so you are not climbing a ranking, you are being matched to a person who did not search for anything. That flips the weighting: freshness matters far more than it does in evergreen search, the hero image is a first-class factor rather than a nicety, the headline has to earn a tap in a scroll rather than a click on a results page, and topical consistency (does Google clearly understand what your site or channel is about?) drives whether you get matched to the right interest cluster. The technical eligibility overlaps with video SEO, but the levers that decide whether a card is surfaced are closer to those of a recommendation feed than a search engine.
On September 18, 2026, Google's VP of Product for Search, Robby Stein, announced a test of a "Dive deeper" button in Discover that opens an AI-generated topic overview — a short summary with prominent links to related stories, community reactions, and original reporting, labeled as generated with AI. Google said it would test the experience on videos first and experiment with multiple designs. For video, it means two things: an interested reader can now get a topic summary in-feed and may not tap through to any single video, and the overview links out to sources, so a video that is one of a topic's credible, well-covered pieces can earn exposure inside the summary. It is an early, limited test with no rollout timeline, so treat it as direction, not a shipped ranking factor.
For a video embedded on your own watch page, Google's published guidance is concrete: use a high-quality image at least 1200 pixels wide, enable the max-image-preview:large meta setting so Google can show a large preview, prefer a 16:9 aspect ratio that matches how Discover cards render, and avoid generic or heavily text-laden images. Specify the preferred image through og:image or schema markup. Videos themselves also need a stable, high-quality thumbnail, which you can set via the video element's poster attribute, VideoObject structured data's thumbnailUrl, a video sitemap, or og:video:image. The image is not a finishing touch in Discover — a card without a strong large image often simply does not surface, regardless of how good the content is.
Substantially, yes. Discover is built around current interests, so it leans toward recently published content in a way evergreen search does not — timely, of-the-moment pieces frequently outperform older, more authoritative ones because the feed is trying to match what a person cares about now. Google's own guidance describes Discover content as timely for current interests. This does not mean only breaking news qualifies; evergreen video can and does appear, and refreshing a strong older piece can help. But it does mean cadence is a real lever: a channel or site that publishes on-topic video regularly gives Discover many recent, relevant candidates to surface, while one that posts occasionally gives it few.
No, and any tool or agency that promises it is misleading you. Discover has no submission, no ranking you can occupy, and no query to target. Google is explicit that eligibility does not guarantee appearance and that Discover traffic is less predictable and less dependable than search. What you can do is maximize eligibility and fit: meet the technical requirements, publish genuinely useful video on a consistent topic with strong hero imagery and honest headlines, maintain good page experience, and keep a regular cadence so the feed has fresh, relevant candidates. You are improving the odds that the recommender picks you up, not buying a slot.
Google Discover is a personalized feed, not a search result, so a video is not ranked — it is matched to users by predicted interest and either surfaced or not. Videos reach Discover two ways: natively via YouTube and Shorts, or embedded on an indexed watch page you own. The levers Discover weights heavily are freshness, a strong 1200px 16:9 hero image, a curiosity-not-clickbait headline, E-E-A-T, and topical consistency. Google's September 2026 topic-overviews test adds an AI summary layer that starts on video cards.
Get started → · ← All guides · Compare Kompozy vs other tools