Short answer: Discover now delivers the majority of Google traffic to most news publishers, and it has no queries, no positions and no rank report. You cannot optimise it the way you optimise Search. What you can control is eligibility, image quality, headline honesty and demonstrated expertise on a specific beat — and then measure it as a share of sessions over time rather than as a position.
Most newsrooms still run their SEO process against Search. That process was built when Search was the majority of the traffic. It no longer is, and the surface that replaced it rewards different things.
The shift, in numbers
Google Discover now accounts for roughly two thirds of Google traffic to news publishers, while Search has fallen to around a quarter. Around nine in ten publishers now take more traffic from Discover than from Search.
The direction of travel is not subtle. Research by Define Media Group, reported by Search Engine Land, found organic search clicks down roughly 42% since AI Overviews began expanding — an initial drop of around 16% that deepened as coverage grew. Over the same window that study put breaking-news traffic up 103% and Discover up 30%.
So two things happened at once: the surface publishers optimised for shrank, and the surface they cannot rank-track grew. Most editorial planning has not caught up with either.
What "no queries" actually changes
In Search you have a query, a position and a click-through rate, and you can reason backwards from all three. Discover gives you none of them. There is no keyword to target, no position to improve, and no way to attribute a spike to an intent.
Three practical consequences:
- Rank tracking tells you almost nothing about your largest channel. If your monthly report leads with average position, it is leading with a minority of your traffic.
- You cannot A/B a headline against a query. You can only compare performance across articles over time.
- Volatility is normal and mostly unexplainable at the article level. Judge the trend, not the day. (Diagnosing a genuine collapse is a different job with a different method, and we cover it separately.)
The reporting fix is to measure Discover as a share of sessions alongside Search and direct, tracked monthly, with revenue per thousand pageviews next to it. Traffic mix matters as much as traffic volume, because Discover sessions and search sessions do not monetise identically.
What Google actually requires
Less than most people assume. Google is explicit that no special tags or structured data are required for Discover eligibility — content needs to be indexed and to meet the content policies. The rest is quality signals.
| Requirement | Specification | Why it matters |
|---|---|---|
| Indexed | Standard indexing | Eligibility floor; nothing else applies until this is true |
| Large image | At least 1200px wide, more than 300,000 total pixels, 16x9 | Discover is a visual feed; small images get a smaller card or none |
| Image markup | schema.org markup or the og:image tag | Tells Google which image represents the article |
| Not a logo or a text-heavy graphic | Google advises against both | Generic imagery is explicitly discouraged |
| Honest headline | Titles that "capture the essence of the content" | Clickbait is called out by name in the guidance |
One more control worth knowing: the max-image-preview:large robots directive permits a larger preview image, "up to the width of the viewport", and applies across Google surfaces including Discover. Google does not frame it as a Discover eligibility requirement, so treat it as removing a constraint rather than as a growth lever — but if your site currently restricts preview size, you are competing in a visual feed with one hand tied.
Three eligibility failures we see most
Before optimising anything, rule these out. All three are common, all three are invisible in a rankings report, and each one caps Discover performance regardless of how good the journalism is.
The image is technically ineligible
The 1200px minimum is a floor, not a target, and it applies to the image Google actually selects — which is the one in your og:image or schema markup, not necessarily the one at the top of the article. Publishers routinely ship a beautiful 2000px hero and an og:image still pointing at a 600px thumbnail from an old template. Check what the tag actually resolves to, on a real article, today.
The other frequent miss is the aspect ratio. A 16x9 crop is what the card wants; a square or portrait image gets letterboxed or dropped.
Preview size is restricted site-wide
If a robots meta tag or an HTTP header restricts image previews, you have opted out of the large card in a visual feed. This is usually inherited from a template or a security-hardening pass that nobody connected to traffic. It costs one line to fix and it is worth checking before any content work begins.
New articles are slow to index
Discover is disproportionately a breaking-and-recent surface, so indexing latency translates directly into lost eligibility during the window that matters. If your news URLs take hours to appear, that is an infrastructure problem, not an editorial one — check your sitemap freshness, internal linking from the homepage, and whether anything is rate-limiting the crawler.
Headlines: the tactic that reversed
This is the single biggest behavioural change for most newsrooms.
Google's Discover guidance warns publishers to avoid "clickbait and similar tactics to artificially inflate engagement by using misleading or exaggerated details in preview content", and separately to avoid "sensationalism tactics that manipulate appeal by catering to morbid curiosity, titillation, or outrage". Reporting through 2026 has associated clickbait headline patterns with losses of roughly 30 to 60 percent of Discover traffic.
The withheld-information headline — "You won't believe what happened next" — was a Discover growth tactic for years. It is now a liability, which means the house style that was rewarded is the one that has to change.
The replacement is not a boring headline. It is a specific one. "Council approves 400-home development on green belt site" beats both "Big decision on housing" and "What the council just did will shock residents". Specificity gives the feed something to match against a reader's demonstrated interests, which is the whole mechanism.
Depth on a beat beats breadth
Discover surfaces content to people based on their interests, so the system needs to understand what your publication is reliably about. A site that covers one beat deeply is a clearer signal than a site that covers everything shallowly.
For an editorial team that means resisting the pull toward chasing unrelated traffic spikes. A finance publication that publishes a celebrity story will sometimes get a good day from it, and will muddy the topic signal that produces good months. Publishing consistency on the beat is worth more than occasional reach off it.
Bylines carry part of that signal. A named reporter with a real author page, a biography that states what they cover, and an archive of work on that beat is a stronger expertise signal than a masthead byline or "Staff Writer". If your CMS publishes most articles under a generic account, you are discarding the clearest evidence you have that a specific human covers this subject regularly. Author pages cost nothing and are among the few E-E-A-T signals fully within a newsroom's control.
This is the same logic that governs answer engine optimization and what gets a source cited by AI assistants: systems that have to choose a source prefer one with a demonstrable, narrow claim to the topic.
What evergreen content is for now
Explainer content that used to carry the revenue floor between news spikes is the material most exposed to AI Overviews — impressions hold, sessions fall.
That does not make evergreen worthless; it changes what it should be. The pieces that still return sessions are the ones an overview cannot replace: original data you gathered, first-hand reporting, tooling, and opinion with a named author attached. Reference explainers that restate consensus are now the weakest thing a newsroom can spend a day on.
A working measurement setup
- Split sessions by surface — Discover, Search, direct, social — and report the split, not just the total.
- Put revenue per thousand pageviews next to it, because a rising session count with falling revenue is a real outcome a traffic-only report hides.
- Track article-level Discover performance over rolling windows, not days.
- Log every Google update against the trend line so volatility can be read against a known event rather than guessed at.
- Review headline patterns quarterly against the guidance above, as an editorial standard rather than an SEO task.
None of this is rank tracking, and it should not be presented as such. It is a trend, a mix and a revenue line — which is what a board actually needs to see anyway.
We build this measurement and editorial layer with news websites and digital publishers, usually alongside the search and answer-engine work and the growth programs it feeds. If your Discover share is moving and nobody can explain why, get in touch.
