A new AI personalization layer lets readers describe the topics, links, and content types they want in Discover, in plain language — reshaping the feed publishers depend on for discovery traffic.
2026-08-21 · by Moe Ameen
On August 20, 2026, Google rolled out AI-powered personalization to Discover, alongside related updates to Search and Google News. The headline change: readers can now customize the Discover feed by describing what they want in plain language, through a chatbot-style interface reached by tapping the three-dot menu on a story in the Google app. You tell it the topics, links, or types of content you want more or less of; the system confirms your preferences, shows what it will prioritize, lets you refine, and remembers the instruction so future feeds reflect it. Google said the feature is rolling out in the coming days.
This is a shift in how Discover decides what to show. Discover has always been an interest feed that Google assembles from your activity — you didn't ask for any of it. Now the reader can steer it directly by conversation, which turns a passive recommendation surface into one each user actively tunes. The same update added custom audio briefings in the Google News app on Android, where you pick specific topics for a daily rundown with sources and links.
Google paired the personalization with a publisher-facing control it had been building out: an embeddable "Preferred Sources" button a publisher can place on its own site so readers add that outlet as a preferred source across Search, Discover, and Google News. Google said more than 600,000 unique sources had been selected as of this update. Preferred Sources first launched in 2025 and expanded globally in 2026; the embeddable button (launched August 20) and now the conversational feed controls are the newer layers on top.
The backdrop is that Discover is becoming more of an AI-first product. Google has also been placing AI-written summaries in Discover that group several publishers' stories beneath roughly two sentences of AI text, with a cluster of source thumbnails — so a topic that a dozen outlets covered can surface as one summary with the individual links a tap away. Combined with a feed users can now reshape by telling an AI what they care about, the effect is that Discover's referral traffic is increasingly mediated by AI rather than by a straightforward stream of publisher headlines. Treat the specifics as an early snapshot of a fast-moving surface and confirm current behavior in Google's own documentation.
Read this as a change in what wins discovery: not the single best-optimized page, but a recognizable, consistent presence on the exact topics readers now tell an AI they want more of. A feed a user tunes by conversation will keep routing them to sources it has learned they engage with on that subject — so the play is topical depth produced reliably across every format and platform Discover samples, not one headline tuned for an algorithm that readers are now overriding by hand. That production problem is what [Kompozy](/) exists to solve.
Point Kompozy at one idea on a topic you want to own — a transcript, a post, a product note — and it generates a full spread around it: captioned [Persona Shorts](/glossary/persona-shorts) and avatar video for the short-form surfaces Discover increasingly pulls from, brand-exact [Carousels](/glossary/hyperframes) and Photo Posts, a Blog Article, and an Email Newsletter, every asset held to one [Persona Brief](/glossary/persona-brief) so the voice and face read as the same source each time. Kompozy's per-post review pipeline and [Autopilot](/glossary/autopilot) then schedule and fan the set across the eight social platforms plus blog and email, so when a reader tells Discover "more of this," you already occupy the topic wherever the tuned feed and the platforms behind it can find you. For the related publisher-controls story, see [Google's new anti-traffic-loss controls](/news/google-publisher-controls-ai-traffic-loss); for why social presence now matters to Google discovery, see [AI Overviews citing Facebook, Instagram, and TikTok](/news/google-ai-overviews-facebook-instagram-tiktok).
It is an AI personalization layer, rolling out from August 20, 2026, that lets readers customize the Discover feed by describing in plain language what they want more or less of. You reach it through a chatbot-style interface in the three-dot menu on a story in the Google app; it confirms your preferences, shows what it will prioritize, and remembers the instruction for future feeds.
It makes discovery more mediated by AI. Readers steer the feed by conversation instead of receiving a straightforward stream of headlines, and Google separately groups publishers under short AI-written summaries. Both reduce the guarantee that a standalone headline earns a click, which is why building owned, cross-platform presence — rather than relying on one Google surface — is the more durable response.
No, they are separate but related controls shipped together. The AI chatbot feed lets a reader describe the topics and content types they want; Preferred Sources lets a reader mark specific publishers to prioritize across Search, Discover, and Google News, and publishers can embed a button on their own site to prompt it. Google said more than 600,000 unique sources had been selected as of this update.
Build recognizable, consistent presence on the topics you want to own, across the formats and platforms Discover samples from. A tuned feed keeps routing readers to sources they engage with on a subject, so depth and consistency win. A tool like Kompozy turns one input into captioned shorts, carousels, images, a blog, and a newsletter in your brand voice, then schedules and publishes them across the eight social platforms plus blog and email.