How AI content for Google Discover gets picked in 2026
SearchBlueprint · · 11 min read · scored 95/100
AI content for Google Discover works when the model drafts and a person adds what the feed rewards: original insight, a named author, an image at least 1200 pixels wide and honest packaging. Discover recommends pages to people who never typed a query, so its quality filters run hotter than the results page.
That filter tightened twice this year. The February 2026 core update pushed summarize-only pages down the feed, and since late August readers can tell Discover, in their own words, what they want more or less of. Both changes work against generic output and in favor of sites a reader would pick on purpose. What follows is how the channel selects content now, the technical bar for entry and a production workflow that scales without putting your domain at risk.
How Google Discover decides what to show
Google Discover is the content feed on the Google app home screen and on Chrome's mobile new tab page, and no query is involved anywhere in it. Google predicts what each reader wants from their activity across Search and its apps, then pushes matching pages into the feed. According to Google Search Central, any indexed page that complies with Discover's content policies is automatically eligible, with no special tag and no required structured data. Eligibility is not a promise, though: Google states outright that qualifying content may never appear at all.
The scale is why publishers keep chasing it anyway. Digital Applied's 2026 Discover guide puts the channel at billions of impressions a month, reaching a reported 800 million users, with average click-through listed at 4-6%. A single placement can outdraw weeks of ranking work. The same guide, and Google's own documentation, agree on the cost: the traffic cannot be targeted with keywords, spikes without warning and decays in about three days.
The three Google surfaces now behave differently enough to deserve separate plans:
| Channel | How your page gets chosen | Traffic pattern | What to optimize |
|---|---|---|---|
| Google Search | Matched to a typed query on relevance and intent | Steady, compounds over months | Keyword coverage, depth, internal links |
| Google Discover | Pushed to readers whose interests match the topic, no query involved | Sharp spikes that fade within 48-72 hours | Images, headline honesty, freshness, author trust |
| AI Overviews | Cited by a model synthesizing an answer from sources | Occasional referral clicks from citations | Quotable, factual, well-structured passages |
Discover borrows most of its quality signals from Search, so nothing here replaces normal SEO. Treat it as a second distribution channel your best pages qualify for automatically.
Can AI content earn a place in Google Discover?
Yes, and Google has been consistent about the reasoning: its systems evaluate the page rather than the tool that produced the first draft. The spam policies in Google Search Essentials go after scaled content created to manipulate rankings, however it was produced. A model-drafted article with real substance and a real author behind it sits on the right side of that line. A thousand thin summaries published in a month sit on the other.
What the February 2026 core update changed
The update raised the bar specifically in discovery surfaces. Digital Applied's analysis found original, well-researched content surfacing more often after February, while pages that only summarize existing coverage lost impressions. Sites built on light rewrites of other people's reporting dropped. Publishers with proprietary data, expert commentary and first-hand detail gained ground. E-E-A-T signals, the named author with visible credentials above all, now separate occasional placements from consistent ones, because a feed recommendation is Google vouching for you harder than a blue link ever did.
The information gain test
Before a draft ships, ask what it contains that the model's training data does not. Your own numbers. A customer's story with a name attached. Photos you took. A position you are willing to defend. Something that went wrong and what it cost. If the answer is nothing, Discover has no reason to push the piece at anyone, because a hundred other pages already say the same thing. This is also the practical dividing line between work a model should do and work a person should do; our breakdown of when to use an AI writer and when a human should take over draws it task by task.
The 2026 feed listens to readers directly
On August 20, 2026, Google announced a round of personalization features that changes who your real audience is. Readers can now open the three-dot menu on a Discover card and describe, in plain language, what they want more or less of. The feed adjusts and remembers. Google also released an embeddable Preferred Sources button that lets a reader pin your site so it appears more often in Top Stories, AI Overviews and AI Mode, and Google says readers had already selected more than 600,000 unique sources at launch. The Verge's coverage the same day noted the chatbot-style interface confirms your choices before refreshing the feed.
The consequence is bigger than the features. When readers curate their own feeds, a site nobody would name as a preference loses reach to the sites people actively pin, and generic AI content has never given anyone a reason to pin a source. A distinct voice, consistent beats and original reporting now pay twice: once in the algorithm's quality scoring, and again in reader-declared loyalty.
The technical bar, straight from Google's documentation
Google's Discover documentation is unusually specific about images, because the feed is a stack of picture cards. Miss the specs and your content is close to invisible there regardless of quality.
| Requirement | Google's spec | Why it matters in the feed |
|---|---|---|
| Image width | At least 1200 pixels wide | Smaller images disqualify the page from the large card format that earns clicks |
| Total resolution | More than 300,000 pixels; a 1280x720 image is 921,600 and passes | The image has to survive cropping across device sizes |
| Aspect ratio | 16:9, cropped to horizontal yourself rather than trusting auto-crop | Auto-cropping a portrait image can cut the subject out of the card |
| Preview permission | The max-image-preview:large robots setting, or AMP | Without it Google is not allowed to show the large image at all |
| Declared image | Set og:image or schema.org image markup; never a logo or a text-heavy graphic | Google picks the thumbnail from what you declare |
Two more items round out the checklist. Core Web Vitals apply because Discover is almost entirely mobile: the thresholds to clear are LCP under 2.5 seconds, INP under 200 milliseconds and CLS under 0.1. Article or NewsArticle schema is recommended rather than required; it hands Google clean data about the author, the publish date and the imagery, which supports the trust signals the February update rewards.
A production workflow for AI content for Google Discover
The five steps below assume a small team: one editor, a drafting model and a site with some topical history.
1. Plan inside topics you already own
Discover surfaces sites with visible depth, so pick three to five clusters where you have standing and stay there. A trending story inside your cluster is worth chasing within hours. The same story outside it teaches the feed nothing about you. If your calendar is improvised, build the cluster map first: our walkthrough of an AI-powered content plan built in five steps shows how to pick clusters from data instead of instinct.
2. Brief the model with what it cannot know
By the time the model writes, the research should already be done. Give it your data and your customer language, and set the angle before it drafts a word. A prompt that only names the topic returns the average of everything published on it, which is exactly what the February update demotes.
3. Add the information gain by hand
Insert the material from the test above: figures, names, photos, the opinion. This is the step teams skip under deadline pressure, and it is the difference between a page the feed can recommend and a page it has already seen a hundred times.
4. Put a named author on it
A real person, a real bio, credentials that check out, linked profiles. Discover is a proactive recommendation, and Google is visibly more careful about what it pushes than about what it merely ranks. Anonymous bylines read as risk to a system that has to vouch for whatever it recommends.
5. Package the card honestly, then keep the cadence
The headline must state what the piece delivers. Google's Discover policies name withheld information, morbid curiosity, titillation and manufactured outrage as grounds for removal, and a pattern of misleading cards can draw a manual action, visible in Search Console. Pair the headline with an original image at spec and accurate Open Graph tags. Then hold the schedule: 2-3 substantial pieces a week, no gaps longer than a week, and refresh proven winners with new data instead of always starting from zero.
How much automation is safe?
| Approach | What it produces | Discover outcome | Where it fits |
|---|---|---|---|
| Full automation, no review | High volume, no original input, no accountable author | Demoted since the February 2026 update, and scaled publishing invites spam scrutiny | Nowhere that Discover matters |
| AI-assisted with human editing | Model drafts at scale while people add data, judgment and a byline | Eligible and competitive when each piece passes the information gain test | Most teams, most topics |
| Fully manual | Original reporting and investigation | Strongest trust signals, slowest output | Flagship pieces that anchor a cluster |
The middle row is where the economics work. The failure mode is drift: an AI-assisted pipeline quietly becomes a full automation pipeline the first month nobody has time to edit. Build the human steps into the process itself, because good intentions are the first thing a busy month deletes. If a checklist item can be skipped silently, eventually it will be.
Measure Discover as the bonus it is
Discover reporting lives in its own tab in Google Search Console, and the tab only appears once your site clears a minimum impressions threshold. The report covers 16 months of impressions, clicks and CTR, and it counts Discover activity in Chrome as well. A missing tab is itself a finding: nothing has surfaced yet.
Read the numbers with the channel's shape in mind. A placement that sends five figures of traffic in two days and nothing on day four is Discover working normally. Watch the pattern across placements instead: click-through against that reported 4-6% average, and whether readers stay or bounce straight back to the feed. Engagement drives the recommendation loop, so a card that overpromises hurts the next card's chances. Keep Discover out of your revenue baseline too, since Google describes the traffic as complementary to query-based Search precisely because it cannot be relied on month to month.
Mistakes that keep AI content out of the Discover feed
- Shipping the og:image as a logo, a text-heavy graphic or an undersized stock photo. Google's documentation warns against all three.
- Writing curiosity-gap headlines. The payoff you withheld is the exact clickbait signal Discover's policies target.
- Publishing under a house byline or no byline. The February update made anonymous authorship expensive.
- Bursting: eight posts one week, silence for three. Cadence is itself a signal that the site is alive.
- Chasing every trend regardless of fit. Placements outside your clusters dilute what the feed knows you for.
- Declaring victory off one spike. One placement is luck; placements every month are a system.
FAQs
Does Google penalize AI content in Google Discover?
No. Google's spam policies go after scaled, low-value pages however they are produced, and say nothing about the drafting tool. An AI-drafted page with original input, a named author and honest packaging is eligible for Discover like any other indexed page.
How long does Google Discover traffic last?
A few days in most reported cases. Digital Applied's 2026 guide describes spikes arriving within 24-48 hours of a placement and gone within 48-72. Treat each spike as a bonus and build durable traffic through Search.
Do I need a news site to appear in Discover?
No. The feed carries sports, health, lifestyle and niche interest content from across the web, and Google notes that older pieces can surface when they match a reader's interests. Timeliness helps, and a focused site with depth qualifies just as well as a newsroom.
What image size does Google Discover require?
At least 1200 pixels wide and more than 300,000 pixels in total, with large previews enabled through the max-image-preview:large robots setting or AMP. A 1280x720 image passes the resolution test at 921,600 pixels.
Where do I see Discover performance for my site?
In Google Search Console's Discover tab, which appears once your content passes a minimum impressions threshold. It reports impressions, clicks and CTR for the past 16 months, and Chrome's Discover surfaces are counted in the totals.
See where your site stands first
Most sites fail the Discover image and page experience bar before content quality ever gets judged, so check the plumbing before you scale AI content for Google Discover. Open a sample SEO report to see exactly what a full site audit measures, from image markup to Core Web Vitals to content depth. Then talk to the team about running one on your own domain.
Resources / References
Sources
About the author
SearchBlueprint researched, wrote and scored this article with the content pipeline our customers run on their own sites. The score in the byline is what our content engine gave this page before it published. Nothing here is reviewed by hand, which is the point: it is the product's output, published where anyone can check it.