For a decade, Google Discover was a feed that happened to you. Google watched your activity, inferred your interests, and assembled a stream of story cards you never asked for — and publishers optimized for that black box the only way they could, by chasing the headline and thumbnail that the algorithm seemed to reward on any given week. On August 20, 2026, Google broke that model. It began rolling out an AI personalization layer that lets a reader tap the three-dot menu on any story and tell Google, in plain language, exactly what they want more or less of — the topics, the specific links, the content types — with a chatbot that asks follow-up questions and then remembers the instruction for every future feed. Discover stopped being a passive recommendation surface and became one each reader actively tunes. That single change rewires the publisher's job. When a reader can say 'more of this topic, less of that,' the surface no longer rewards whoever won the headline lottery this morning; it rewards whoever is reliably, recognizably present on the topics real people ask for, in the formats they ask for them in. And because Discover in 2026 pulls heavily from short-form and social — not just article links — and increasingly groups publishers under two-sentence AI summaries that mediate the click, the response is not a headline tactic. It is a packaging discipline: produce recognizable, on-topic content consistently, across the formats and platforms a tuned feed samples from, on owned channels a reader can deliberately choose. This guide is the practical version of that shift — what the feature actually does, why reader-steered discovery changes what 'optimizing for Discover' means, how the feed's move toward AI summaries and social content changes what earns a click, and how to build a content operation that keeps showing up when a reader tells the feed to bring them more of you.
Google Discover is the personalized feed of story cards that appears in the Google app and on the Android home screen — for years, a stream Google assembled from your activity without you asking for any of it. On August 20, 2026, Google began rolling out an AI personalization layer that changes who is in charge of it. You can now tap the three-dot menu on any story in your feed and tell Google, in your own words, the topics or links you want to see more or less of. A chatbot-style interface takes the request, can ask follow-up questions to refine it, and — the part that matters — remembers the instruction so it shapes every future feed, not just the next refresh. Google framed it as making it 'as easy to tune your Discover feed as it is to text a friend,' rolling out in the coming days in the Google app.
The same announcement bundled two related controls. An embeddable Preferred Sources button lets a reader mark specific publishers to prioritize across Top Stories, AI Overviews, and AI Mode, and publishers can now place that button on their own pages so a reader adds them in one tap; Google said more than 600,000 unique sources had been selected. And in the Google News app on Android, readers can now build custom audio briefings by choosing specific topics, with source attribution and links out. The through-line across all three is a single shift: discovery is becoming something readers dictate — by topic, by content type, and by name — rather than something an algorithm hands them. The news write-up of the launch is in Google's chatbot-tailored Discover feed, and the wider cross-surface rollout in Google's personalization across Search, Discover, and News.
When Discover assembled the feed on its own, publishers optimized the only variable they could observe: the headline and thumbnail. You wrote for the card, watched which framing the black box seemed to favor that week, and chased it. A reader-steered feed retires that game. Once a person can say 'more of this topic, less of that,' the feed is no longer deciding what interests them — they are — and the publisher who benefits is whoever is reliably, recognizably present on the topic that reader just named. The optimization unit stops being a single winning card and becomes a standing relationship with a subject and an audience.
That is a harder thing to fake and a more durable thing to build. A one-off headline optimized to win the old algorithm does almost nothing when the reader has told the feed, in plain language, what they want; a body of consistent, on-topic, recognizable work is exactly what a topic instruction routes toward. This is the same behavioral shift reshaping search — discovery is moving from an inferred, algorithm-owned decision to a stated, reader-owned one — and the strategic consequence is identical: presence and consistency beat the single optimized asset. The search-side version of that argument is in SEO in the age of AI search.
A feed that every reader tunes is not one feed anymore; it is millions of them. When each person steers their own stream, the idea of 'ranking in Discover' as a single achievable state dissolves — there is no shared feed to top. What replaces it is coverage: being present, on-topic, and recognizable across the specific interest slices your audience actually asks for, so that whichever way a given reader tunes, you are in the version of Discover they built. One-size-fits-all Discover optimization matters less; being the dependable source on a well-defined set of topics matters more.
This rewards depth and repetition over the scattershot post. A publisher who covers one topic thoroughly and consistently earns a place in the micro-streams of everyone who asked for that topic; a publisher who posts one thing about it and moves on does not register when a reader deliberately requests more. The dynamic mirrors what AI personalization does to reach everywhere it lands — audiences split into narrower, self-selected streams and generic breadth loses to specific, sustained presence. The reach-and-filter-bubble mechanics are worked through in filter bubbles in AI search and content discovery.
Reader tuning is only half the change. Discover in 2026 increasingly groups multiple publishers under short, AI-written summaries with a thumbnail cluster, so a reader can absorb the gist of a story without opening any single source. That intercepts the click a standalone headline used to earn, and it does it on the surface publishers most depend on for discovery traffic. Treating Discover as a reliable click machine — write the headline, collect the visit — is exactly the assumption this breaks.
Which reframes what content is worth making for the feed. If a two-sentence summary can convey the informational value, the summary will, and no click follows. The click survives only for what the summary cannot reconstruct: a distinctive point of view, first-hand reporting or lived experience, a recognizable creator or brand voice, or media the blurb can only point at rather than reproduce. Generic, interchangeable coverage is precisely what an AI summary absorbs; a specific, identity-bound take is what a reader taps through to get. The full argument for why specificity out-earns generic coverage in an AI-mediated feed is in AI SEO and specificity, and the traffic-side reality of the shift is in the publisher traffic collapse.
A third fact changes what 'packaging for Discover' even means: the feed no longer draws only from published articles. Short-form and social content make up a large and growing share of what Discover surfaces, which means a tuned feed samples video and social posts alongside blog links. A publisher present only on its own site is invisible to that portion of the feed — so when a reader asks for more of a topic, the outlets that also live natively on the platforms Discover pulls from have more surfaces the instruction can route back to.
So the response to a reader-steered feed is not just 'publish more articles about the topic.' It is to be recognizable on that topic across the formats and platforms Discover samples: a blog and newsletter you own, plus native presence on the social surfaces the feed increasingly favors. Each format is another way for a topic instruction to find you, and consistency of identity across them is what lets the feed — and the reader — recognize the scattered pieces as one source worth asking for more of. The retrieval-side version of this multi-surface argument is in social content for AI search visibility.
The reason to build owned channels in a reader-steered world is not nostalgia for the direct visit; it is that owned channels are where the two reader-control features actually pay off. The Preferred Sources button lives on your own site — you can only prompt a reader to mark you if they are already on a page you control, engaged enough to tap it. And a blog plus an email list are the surfaces a reader who asked the feed for 'more of you' can be routed to and then kept, off the mediating layer entirely. A tuned feed can still deliver a reader to you; whether that turns into a durable relationship depends on whether you have somewhere of your own to receive them.
This is the same conclusion the rest of Google's 2026 publisher toolkit points to from a different angle: the prominence and measurement controls Google shipped amplify an audience relationship you already have but never create one. The honest read on what those controls can and can't do is in Google's publisher tools for AI traffic loss, and the full set of levers that offset the decline is in AI search traffic loss mitigation for publishers. Discover's chatbot feed is one more input into that strategy — not a substitute for it.
Some Discover habits actively work against you now. Stop optimizing a single headline as if winning one card is the goal — the reader, not the algorithm, increasingly decides what interests them, and a body of consistent on-topic work beats any one framing. Stop treating Discover as a guaranteed click channel; between reader tuning and AI summaries, a standalone headline no longer reliably converts, so a strategy that banks on it is fragile by construction. And stop covering a topic once and moving on: a feed that rewards being asked for 'more' punishes the publisher who has no more to offer on the subject a reader just named.
The subtler trap is publishing interchangeable coverage — the competent, generic article that says what every other article says. That is exactly the content an AI summary absorbs without a click and exactly the content a reader has no reason to request more of by name. Specificity, a recognizable voice, and first-hand substance are what survive both the summary and the tuning. The wider case for why generic content stopped working across every AI-mediated surface is in why AI content stopped working.
Put the four shifts together — reader-tuned topics, fragmented micro-streams, mediating AI summaries, a feed that samples social — and the requirement is clear: recognizable, on-topic, identity-bound content, produced consistently, across the formats and platforms a tuned feed pulls from, anchored to owned channels. That is a volume-and-consistency demand, and it is where most publishers stall, because covering a set of topics deeply across blog, newsletter, social, and video by hand does not scale to the cadence a reader-steered feed rewards. That gap is what Kompozy is built to close.
Kompozy is a content generation and publishing engine, not a repurposing add-on. Point it at your topic and it generates the range a tuned Discover feed can sample: blog articles and email newsletters for the surfaces you own, text posts and carousels and quote or infographic images for social, and persona or avatar video for the short-form Discover increasingly favors. A Persona Brief governs the voice and a face-locked persona keeps a recurring identity consistent, so the same topic sounds and looks like you whether it lands as a blog post, a LinkedIn update, or a short — which is precisely the recognizability that lets a reader, and the feed, treat the scattered pieces as one source worth asking for more of.
On the publishing side, the same engine schedules and fans that output across the eight social platforms plus blog and email, with autopilot and a per-post review pipeline. That turns 'be reliably, recognizably present on this topic everywhere the feed looks' from a staffing problem into a running cadence: one idea becomes a blog piece, a set of native social restatements, and a video, each published to a surface Discover samples, on a schedule you don't hand-manage. Owning your own blog and list — where the Preferred Sources button lives and where a routed reader can be kept — is the anchor; a generation-and-publishing engine is how you feed enough recognizable, on-topic content into enough surfaces that a reader-steered feed keeps bringing people back to you. For the broader distribution framing, see personal-brand-led content strategy.
It is an AI personalization layer Google began rolling out on August 20, 2026 that lets a reader customize the Discover feed by describing, in plain language, what they want. You tap the three-dot menu on a story in the Google app and tell it the topics, links, or content types you want more or less of; a chatbot can ask follow-up questions to refine the request, and it remembers your preferences for future feeds. It turns Discover from a feed Google assembles for you into one you actively steer.
It moves the target from a headline to a relationship. When Discover assembled the feed alone, publishers optimized the one variable they could see — the title and thumbnail that seemed to win that week. Now a reader tells the feed 'more of this topic,' and the source that benefits is whoever is consistently, recognizably present on that topic. Optimization shifts from winning a single card to being the reliable, on-topic name a reader deliberately asks the feed to bring them more of.
It makes discovery more mediated, which pressures the guaranteed click two ways. A reader steering the feed by conversation gets a stream tuned to them rather than a straightforward run of every publisher’s headline, and Google separately groups stories under short AI-written summaries with thumbnail clusters. Both mean a standalone headline no longer reliably earns the click it once did. The durable response is owned, cross-platform presence rather than dependence on one Google surface.
No, but they shipped together on August 20, 2026 and work in tandem. The chatbot feed lets a reader describe topics and content types they want; Preferred Sources lets a reader mark specific publishers to prioritize across Top Stories, AI Overviews, and AI Mode, and publishers can embed a one-tap button on their own site to prompt it. Google said more than 600,000 unique sources had been selected. Together they make discovery something readers dictate, by topic and by name.
Content a two-sentence summary can’t replace. When Google clusters publishers under an AI-written blurb, the informational value is delivered before the click, so the click goes to whatever the summary can’t reconstruct — a distinctive point of view, first-hand reporting or experience, a recognizable creator or brand voice, or media the summary only references. Generic, interchangeable coverage of a topic is exactly what an AI summary absorbs; a specific, identity-bound take is what a reader taps through to get.
Both, and increasingly social. Discover in 2026 surfaces a large share of short-form and social content alongside traditional article links, so a publisher present only on its own blog is invisible to the part of a tuned feed that samples video and social. Being recognizable across the formats and platforms Discover pulls from — not just publishing articles — is what gives a reader’s topic instruction more surfaces it can route back to you.
Google Discover's AI chatbot-tuned feed, which Google began rolling out on August 20, 2026, lets a reader tap a story's three-dot menu and tell Google in plain language what topics and content types they want more or less of, then remembers it. For publishers, that moves discovery from winning a single headline to being the recognizable, on-topic source a reader deliberately asks for — consistently, across the formats and platforms the feed now samples from.
Get started → · ← All guides · Compare Kompozy vs other tools