Google’s Discover documentation names one requirement you can check in 30 seconds: images at least 1200 pixels wide, served with the max-image-preview:large setting. In the audits we run at Instant Press, that single check fails on more than half of the sites that ask us why Discover sends them nothing. The feed that drives billions of visits across the mobile web ignores them over a robots directive most SEOs have never read.
That is the pattern with Discover. The channel looks like a black box, yet most of the levers sit in plain sight. This guide walks through the seven signals that decide whether you get on Google Discover, in the order you should fix them.
What Google Discover rewards
Discover is not search. Nobody types a query. Google assembles a feed for each person based on their activity across Search, Chrome, YouTube, and Android, then matches that interest graph against fresh content from the open web. Your page does not rank for a keyword. It gets matched to a person.
That inversion changes what wins. In classic search, you target demand that already exists and the title needs to satisfy an explicit question. In Discover, you target curiosity. The card has one job: stop a thumb mid-scroll. Google’s own guidance says to write titles that capture the essence of the content without resorting to clickbait, and its 2021 core update documentation made clear that withholding information to bait clicks gets demoted in the feed.
The practical reading: Discover rewards pages that a defined audience would tap on without ever having searched for them. A piece called “Property Tax Deadlines in Ohio” satisfies search intent. A piece called “The Property Tax Mistake That Costs Ohio Sellers $4,000” earns Discover taps. Same research, different framing, different channel.
It helps to picture the matching engine on the other side. Google holds an interest profile for each user, built from what they searched, watched, and followed, and it holds a topical profile for your site, built from what you publish and who cites you. A card gets dealt when the two profiles intersect and your page is the freshest, most tappable candidate at that intersection. Everything in this guide either widens your site’s topical profile, sharpens the card, or raises Google’s confidence that dealing you into the feed will not embarrass it.
One more behavioral note before the signals: Discover sessions are leisure sessions. People scroll the feed standing in line and lying on the couch, which is why emotional and curiosity-driven framings outperform utilitarian ones, and why business sites with strictly transactional content struggle here. The feed is a magazine rack, not a phone book, and pages need to dress for the venue.
The Discover eligibility stack
We score client sites against what we call the Discover eligibility stack, a four-layer model that organizes the seven signals. The base layer is indexability: the page must be indexed, mobile-friendly, and free of content policy violations. Signal one lives here, and it is binary. If Search Console shows indexing problems, Discover is off the table.
The second layer is presentation. Signal two is the 1200 pixel image rule covered above. Signal three is the headline itself, which must promise something specific without baiting. The third layer is content character. Signal four is freshness, because the feed favors recent publication or substantial updates. Signal five is topical depth, meaning your site has a cluster of related pieces rather than one orphan post.
The top layer is trust. Signal six is entity clarity: a named author, a real organization, an About page that AI systems and quality raters can verify. Signal seven is engagement history, the feedback loop where past tap-through and satisfaction shape how often Google risks showing your next card. You climb the stack from the bottom. Fixing headlines on a site with broken indexing wastes everyone’s time.
The stack also explains why advice about Discover sounds contradictory. One publisher swears headlines made the difference; another saw nothing move until they fixed images; a third reached the feed after adding author pages. All three are right about their own site, because each was missing a different layer. Diagnose before you optimize: walk the seven signals in order against your own site and mark each one pass or fail. Most sites fail two or three, and the failures cluster at whichever layer nobody on the team owns.
Audit your images before anything else
Of the seven signals, images give the fastest win. Discover is a visual feed. Large-card placements, the ones with the full-width image, pull a multiple of the clicks that small thumbnails get. Google grants the large card when the image is at least 1200 pixels wide and you have opted in via max-image-preview:large.
Run this check today: open the page source of your five most recent posts and search for max-image-preview. If your CMS or SEO plugin has not set it to large, that one line of robots meta is the highest-value edit on this page. Then check the actual hero files. A 1200 pixel rule means the source file, not the display size. Plenty of themes downscale uploads to 800 pixels and quietly disqualify every post on the site.
While you are in there, kill stock photos that say nothing. A feed card competes against family photos and sports highlights for attention. Generic handshake imagery loses that fight. Photos with faces, motion, specificity, or tension earn the tap. The image is half the pitch, and on Discover it renders larger than your headline.
Treat the image-headline pair as one creative unit and test it like one. The strongest cards pair a concrete photo with a headline that explains why the photo matters, so each element makes the other more interesting. A picture of an ordinary kitchen above “The Renovation Choice That Cuts Resale Value 12%” works because the mundane image gains tension from the words. Mismatched pairs, dramatic photo with a dry headline or vice versa, leak taps at the exact moment the thumb hesitates.
Why does Discover skip evergreen content?
The freshness signal frustrates publishers who built libraries of timeless guides. Discover leans hard toward content published or meaningfully updated in the past few days. Google has said evergreen content can appear when it matches a fresh spike in personal interest, and it does happen, but the distribution curve concentrates on recency.
You do not need to abandon evergreen strategy to get on Google Discover. You need a refresh rhythm. Take the guide that earned links for two years, update its data, add a section on what changed this quarter, revise the publish date with an honest updatedDate, and reframe the headline around the new angle. The underlying asset stays evergreen. The Discover-facing surface stays fresh.
This is also where the cluster signal compounds. A site that publishes one piece on a topic looks like a tourist. A site with twelve interlinked pieces, three of them updated this month, looks like a destination. Google’s systems read that density as topical authority, and the feed pulls more confidently from sites it has already classified as authorities on the interest it is trying to serve.
A practical refresh cadence for a small team: pick your ten strongest evergreen assets, assign each a quarter, and give the assigned piece a genuine update during its quarter, new data, a current example, a revised takeaway. Pair each refresh with one net-new piece on the same cluster so freshness and depth move together. Avoid the lazy version of this, where someone bumps the date without changing the page; Google compares versions, and a cosmetic date change with identical content reads as the manipulation it is.
Build entity trust before you chase volume
Discover has a defined content policy layer that search does not enforce as strictly: misleading content, manipulated media, and anonymous medical or financial claims get filtered before ranking signals even apply. The fix is entity clarity. Every post carries a named author with a bio that links somewhere verifiable. The organization behind the site has a Knowledge-Panel-grade footprint: consistent name, real address, third-party coverage, a Wikipedia or Wikidata presence if you can earn one.
This matters more in 2026 than it did three years ago because the same entity infrastructure now feeds AI answer engines. The work that helps you get on Google Discover also makes your brand citable in AI Overviews, ChatGPT, and Perplexity. One investment, two distribution channels. Sites that skipped the entity work are finding they are invisible in both.
A test worth running: search your brand name plus your lead author’s name. If the results cannot prove either one exists beyond your own domain, you have an entity problem, and no amount of headline optimization will outwork it.
The infrastructure fixes are concrete. Give every author a real page with a photo, credentials, and links to their external profiles, then connect it with Person schema and sameAs properties. Add Organization schema to the site root with your logo, founding date, and social profiles. Earn at least a handful of third-party mentions, local press, trade publications, podcast appearances, that corroborate what your site claims about itself. None of this is Discover-specific, which is the point: you are building the verification layer that every Google surface, and every AI engine, checks before it promotes you.
Measure Discover or you’re guessing
Search Console ships a dedicated Discover report, separate from the Performance report most teams live in. It appears once your site has meaningful Discover impressions, which creates an awkward cold-start period where you are flying blind. During that period, watch mobile direct and mobile Google referral traffic in your analytics for spike patterns that do not match query data. Those spikes are usually Discover.
Once the report exists, study which pieces earned cards and what the tap-through rate looked like. Discover CTR runs far higher than search CTR when the image and headline work together, and far lower when they fight. Compare your spiking posts against the eligibility stack: what did the winners share? Image size, framing, topic, author? That pattern is your editorial brief for the next quarter.
Set expectations with whoever reads your reports, because Discover volatility breaks dashboards built for search. A post can pull 80,000 impressions in three days and zero the week after, and that is the channel working as designed, not a penalty. Judge the program on rolling 90-day totals and on the ratio of posts published to posts that earned cards. A site converting one post in five into a Discover appearance has a system. A site converting one in fifty has a lottery habit, and the eligibility stack will show you which layer is buying the tickets.
The publishers who treat Discover as a system, image compliance, fresh angles, entity depth, and measurement on a loop, will keep pulling from a feed most of their competitors still call random.