Google Discover Optimization: Cards Worth a Tap
Google Discover optimization is not keyword work. There is no query to match, no rank to hold and no page one to defend. Instead a reader scrolls a personalised feed, sees a card with one image and one headline, and decides in under a second whether your publication is worth opening. That decision is shaped by things you can control: a clear topical identity, a crop-safe image, a headline that tells the truth, and a byline a stranger can verify.
- Topical identity audit: the subjects your site can plausibly own
- Card image system with crop-safe 16:9 artwork and real alt text
- Headline patterns tested against the mobile truncation point
- Author pages, credentials and publisher entity consistency
- A cadence plan that keeps fresh work flowing without thinning quality
No one can guarantee placement in Discover. We remove the reasons a card gets scrolled past.
Mock feed column. Cards, artwork and headlines are illustrative.
Different surface, different rules
Discover is not a smaller search results page
Teams often budget for Discover as if it were a second SEO channel, then wonder why the same pages behave so differently. The mechanics are not the same, and the difference changes what you write.
Search
- A query exists, so intent can be read from words.
- A page competes for a position that is broadly stable.
- Keyword coverage, links and page experience drive eligibility.
- The click is the answer to a need the user already named.
- Reporting includes the query, which makes diagnosis possible.
Discover
- No query is typed; interest is inferred from behaviour.
- One card competes for attention with everything else in a feed.
- Topical clarity, imagery, freshness and trust drive the tap.
- The click is a bet on a headline, made in a fraction of a second.
- Reporting gives page-level data, so cohorts replace queries.
The practical consequence is simple. On a search page you optimise for a phrase. In a feed you optimise a topic, a picture and a sentence. A site with sharp topical edges and consistent publishing gets more from Discover than a broader site that publishes once a quarter.
Eligibility signals
What tends to get surfaced, and what gets skipped
Nobody outside the platform can describe the feed ranking exactly, and anyone who claims to is guessing. What we can see, across publisher audits, is a set of conditions that appear again and again on content that gets distributed.
One subject, repeated with depth
A site that covers home energy in detail beats a general news site on home energy cards. Depth is visible in the archive, the internal links and the number of pages that answer adjacent questions.
An image that survives the crop
The card image is the largest element a reader sees. Low-contrast stock photos, text baked into artwork and busy collages all lose to one clear subject on a simple background.
Visible dates and real updates
Updated pages need updated facts, not a changed date stamp. A visible publish date, a genuine revision note and a current example are what separate a refresh from a rewrite.
A person behind the byline
Author pages with role, experience and a linked archive make a publisher legible. Anonymous bylines are common on sites that plateau and then decline.
The card promises, the page delivers
Feed readers arrive with no context. Pages that open with a definition, a result or a clear answer hold attention. Pages that open with a newsletter prompt lose it.
What happens after the tap
Scroll depth, returning readers and subscription behaviour feed back into how a publisher is treated over time. Manufactured clicks without reading do not help.
None of this is a checklist that guarantees distribution. It is a list of reasons cards fail, and most publishers can fix three of them in a month without writing a single new article.
Imagery
Image specs and the crop that decides the tap
Card images are rendered at several sizes and ratios across devices, and the exact requirements shift. Treat the numbers below as a working range and confirm against the current official documentation before a template change.
| Slot | Aspect ratio | Working pixel range | Practical note |
|---|---|---|---|
| Feed card image | 16 : 9 | 1200 × 675 to 1920 × 1080 | Keep the subject inside the central 80 percent so a tighter crop still works. |
| Article hero | 4 : 3 | 1200 × 900 | Used when a card falls back to the in-article lead image. |
| Square thumbnail | 1 : 1 | 800 × 800 and up | Needed for related modules and some app surfaces. |
| Publisher mark | 1 : 1 | 512 × 512 | Same mark as the favicon, the about page and the social profile. |
| Syndication fallback | 1.91 : 1 | 1200 × 630 | Open Graph image; keep the headline out of the artwork. |
Two rules matter more than the numbers. First, never bake text into a card image; it is unreadable at phone width and it duplicates the headline. Second, build a reusable art direction, three to five visual treatments with consistent colour, so a reader recognises your cards before reading the publisher name. Our SEO Content Writing Services work covers the technical side: file naming, compression, responsive sources, alt text and the metadata that keeps one image consistent across web, feed and social.
Send two URLs: one article that gets feed traffic, one that never has. We will compare the image, headline and byline, and tell you which change to make first.
Headlines
Titles, bylines and the source line under the card
The headline is the only argument you get. It has to be specific enough to promise something real, short enough to survive truncation, and honest enough that the page keeps the promise. Curiosity without substance produces a tap and a bounce, and the bounce is the expensive part.
Our working pattern is: subject first, benefit second, qualifier last. "Heat pump running costs in a cold climate: what the first winter shows" tells a reader what they get and who it is for. "You will not believe what this heat pump did" tells them nothing and attracts a reader who leaves in four seconds.
Under the headline, the source line does quiet work. The publisher name must match the logo, the favicon and the about page; inconsistency here reads as a syndicated copy. A visible date helps on news and hurts on evergreen pieces that have been genuinely updated, so we set the display rule per content type rather than globally.
Headline work is production work, not inspiration. We keep a tested library of 40 to 60 patterns per publisher, each tagged by content type, and we retire patterns when their click-through falls below the publication average for two consecutive months.
Trust
Author pages and the E-E-A-T layer
Feed distribution amplifies whatever trust signals a publisher already has, including the weak ones. A reader who has never heard of your site will check the byline, the about page and the contact page in that order, usually within the first minute.
- Author pages that say something. Role, years in the subject, what they have built or tested, and a list of their articles. A photo helps; a generic avatar does not.
- An about page for the publication. Who writes, how the site is funded, what it covers and what it deliberately does not cover.
- Contact details that resolve. A form plus a real address or phone number. Pages that hide contact behind a chat widget read as lead traps.
- Consistent entity data. The same organisation name in schema, bylines, footer and social profiles. Mismatches split the entity and weaken every page.
- Primary sources and corrections. Link to the study, the standard or the filing. Publish a corrections note when something changes, instead of silently editing.
- Structured data that tells the truth. Article or NewsArticle markup, Person markup for authors, Organization markup for the publisher. No markup for claims the page cannot support.
This layer is unglamorous and it is where most Discover programmes stall. The pages that get shared are usually the ones where a reader can see, quickly, that a named person with relevant experience wrote the piece.
Cadence
A publishing rhythm the team can actually hold
Feeds reward consistency more than volume. A publisher who ships two strong pieces every week for a quarter outperforms one who ships twenty in a month and then nothing. The plan below is a template we adapt to the team size, not a rule.
The rhythm assumes two writers and one editor. Smaller teams should cut the data lane first, because an explainer series keeps the archive coherent while a half-finished data project produces neither freshness nor depth. Writers for the explainer lane come from our Image SEO Services desk when the internal team is at capacity; the topic map and the art direction stay with us either way.
Measurement
How to read Discover traffic without misreading it
Discover reporting is coarser than search reporting, and the numbers swing hard. A single card can produce a week of traffic and then stop, with no query data to explain why. We report in cohorts so a normal dip is not mistaken for a penalty.
- Impressions and clicks by page. The Discover report gives page-level data. Group pages by topic and content type, then compare four-week windows rather than single days.
- Click-through by card cohort. Track the average CTR of cards published under each art treatment and headline pattern. This is the number that improves when presentation work is done properly.
- On-site engagement. Engaged sessions, scroll depth and second-page clicks separate a curious tap from a reader who stays.
- Return behaviour. Direct and branded traffic in the weeks after a spike shows whether a feed visit turned into familiarity.
- Business outcomes, loosely coupled. Newsletter signups, trial starts and quote requests from Discover sessions. Attribution is imperfect, so we report direction and volume, not precise revenue per card.
The one thing we refuse to do is present a spike as a strategy. Feed traffic is rented, not owned. The value of the work is a publisher that a reader recognises the second time, which is why the trust and cadence layers get as much attention as the image crop.
Scope
What the engagement includes
Quoted by scope: number of topic lanes, article templates, author pages and the size of the existing archive to be reworked. No prices on this page.
Topical identity audit
Archive review, competitor card inventory, and a short list of lanes the site can defend with real depth.
Card image system
Art direction, crop-safe templates, alt text rules and a handover pack a designer can run without us.
Headline library
Tested patterns by content type, with truncation checks at mobile width and a retirement rule.
Author and publisher layer
Author page template, entity consistency check, schema review and a corrections policy.
Cadence plan
A four-week publishing rhythm mapped to team capacity, with a refresh queue and a distribution lane.
Reporting dashboard
Page cohorts, CTR by treatment, engagement and return behaviour, reviewed every four weeks.
Evidence
What the work has moved
Illustrative outcomes, anonymised. Feed traffic is volatile, so we report ranges and the period they cover rather than a single best week.
A home improvement publisher, 5 months: feed impressions 0.9M to 3.4M per month, with newsletter signups from feed sessions up 22 percent.
Home improvement · US
A B2B equipment supplier, 6 months: two explainer hubs averaging 41k feed impressions per month, demo requests up 9 per month.
Industrial equipment · UK
A regional food publication, 3 months: average card click-through 1.8 to 3.1 percent after the image and headline system shipped.
Food publishing · Canada
FAQ
Questions publishers ask about Discover
Can you guarantee our articles appear in Discover?
No. Distribution is decided by the platform and varies by reader, device and week. We can improve the conditions that make a card eligible and worth tapping, then track what happens. Any agency promising a fixed number of cards is guessing.
Does Discover traffic convert?
It converts differently. Feed readers arrive with less intent than searchers, so direct sales are rarer and newsletter signups, return visits and brand searches are stronger. Pages that assume prior knowledge convert worst.
How often should we publish?
As often as you can hold quality. Two strong pieces a week with a refresh lane beats a burst of twenty followed by silence. Consistency in topic matters more than raw volume, because the feed rewards a recognisable subject.
Do we need special markup or a separate feed?
No separate feed. Standard crawlable HTML, clean structured data, fast pages and an image that meets the card requirements are the baseline. There is no submission form and no allow-list to join.
Why did feed traffic drop to almost nothing?
Usually one of four things: the topic moved out of season, a template change broke the card image, a competitor published a stronger treatment, or engagement on the landing page fell. We check those in that order before assuming a penalty.
Which images should we stop using?
Text baked into artwork, low-contrast stock photos, collages with no focal point, and the same image reused across a dozen posts. All four waste the largest element on the card and train readers to scroll past your publisher mark.