Dear Algo puts Threads reach in users' hands

In February 2026 Meta announced Dear Algo: a public Threads post that starts with those two words, followed by what you want more or less of, and the feed adjusts for three days. You can also repost someone else's Dear Algo request to apply their preferences to your own feed. In June 2026, as Threads crossed 500 million monthly users, Meta shipped Your Algo as a private complement — the same more-or-less request, lasting one, three, or seven days, visible only to you, managed in one hub with Dear Algo. Both rollouts have been staged by country; read the newsroom posts for the current markets rather than assuming every account has the controls.

Most ranking changes are invisible and arrive as a rumour. This one is a product surface the audience interacts with on purpose, and it moves a variable out of the ranker's hands and into theirs. That rewards a different kind of content. Product surfaces like this move fast, so treat the rest of this page as a posture, not a spec.

A user instruction is not a ranking signal

A ranking signal is a prediction. Instagram's published ranking explanation is a list of them: how likely you are to reshare a Reel, watch it through, like it, tap the audio page. The system guesses what you will do and orders the feed accordingly. You can influence a prediction by producing the behaviour it predicts.

A user control is a stated preference. It does not model what someone will do, it records what they said. Unlike a prediction, it is not something you can produce behaviour to change. For as long as that request is active — three days on Dear Algo, one to seven on Your Algo — hook craft in the topic they asked to see less of does not reopen the door.

That asymmetry is the whole strategic content of this feature. Ranking signals are things you compete for. Stated preferences are things you either match or you don't, for the window the user set.

Pinterest already ran this experiment

If you want to see where user-side controls land, look at Pinterest, which shipped the same idea narrowed to AI content.

Pinterest auto-applies an "AI modified" label from IPTC metadata, owner disclosure, and classifiers that detect generative AI even when the content carries no obvious markers. It then gave users a "see fewer" control for categories it describes as highly prone to AI modification or generation — beauty, art, fashion and home decor among them — with the list expanding over time.

Pinterest applies the label automatically, including to content the creator never disclosed, and then lets users see fewer labelled Pins in those verticals as a standing preference they can change later. If you sell in those categories, an unknown share of your addressable audience has removed itself from your inventory. Pinterest publishes no figure for the size of the loss.

Threads' controls are broader than AI and are framed as preference rather than suppression. They are also temporary topic requests, not a persistent opt-out of a labelled pool. Same class of lever — the viewer, not the ranker, decides a topic is out — with a much shorter window.

Being requestable

A control that works on categories rewards content that is legibly of a category.

Think about what someone actually types after "Dear Algo," or into Your Algo. They ask for more of a thing they can name. Cooking. Football tactics. Book recommendations. Small business finance. Woodworking. Local news. They do not ask for more of "content from brands with a friendly conversational register," because that is not a thing anyone thinks in.

Content that is hard to categorise has always suffered mild distribution problems, because recommendation systems struggle to place it. Under an explicit preference control, that penalty becomes categorical rather than probabilistic. You cannot be asked for by name if you do not have a name.

So the work is to become nameable.

1. Pick a noun. Not a positioning statement, a noun. What would a stranger call your account in one word if they were describing it to a friend? If the honest answer is "marketing," you are competing in a category most people ask for less of, not more.
2. Say it in the account. Bio, pinned post, the first line of most posts. Not as branding, as classification. You are helping both the system and the human place you.
3. Hold the category still for a quarter. An account that posts recipes, then hiring takes, then a product launch, then a meme is not in a category. It is in four, badly. Consistency is the cheapest thing you can do here and the one most brands refuse to do.
4. Make the category narrower than feels comfortable. "Food" is not a request anyone makes. "Fermentation" is. Narrow categories have smaller ceilings and much higher request rates, and on a preference-driven surface the request rate is what matters.

The recurring-format advice in Threads for brands is the same insight from the production side: a weekly bit with a consistent look compounds because it is recognisable. Under Dear Algo, recognisable and requestable are close to the same property.

What this means for AI-generated content

Two implications, and they pull in opposite directions.

The risk. If users can ask for less of something, "AI-generated video" is an obvious thing to ask for less of, and Threads inherits Meta's labelling stack, described in Meta's announcement on labelling AI images across Facebook, Instagram and Threads. Meta concedes in its transparency documentation that the methodology is still evolving and may not capture labels on some AI-edited content, so labelling is inconsistent by the platform's own account. Building a strategy around whether your label appears is building on a detection gap, which is not a permission and not a plan.

The defence. Category is not method. Nobody asks for more "human-shot handheld video." They ask for more fermentation. If your account is a fermentation account that happens to use generated b-roll, the request that matters is one you satisfy. The failure mode is the account whose only identifiable characteristic is that it is AI-generated, because then the method is the category, and the method is exactly what a "less of" request is for.

Answer production and positioning separately. Generated visuals in service of a subject are subject-first. Generated visuals in service of volume are method-first, and method-first accounts are the ones exposed here. Our note on profile-level versus post-level AI disclosure covers how to be straightforward about method without making it your identity.

A working checklist

Before you post anything to Threads through the rest of 2026:

• Can a stranger name your category from three consecutive posts? If not, fix that before optimising anything else.
• Is the category one people ask for more of, or one they tolerate? Be honest. Most brand accounts are in the second group and are competing for indifference.
• Does each post stand alone? The feed mixes followed and recommended content, so a large share of your readers arrive with no context. A post that only makes sense to a regular is invisible to the people who did not already follow you, and that is the group that decides whether you grow.
• Is your visual style consistent enough to be recognised in a scroll? A brand kit and a fixed style preset do more for requestability than any individual post. If you run a recurring generated character, character consistency across episodes is the same argument at the asset level.
• Would removing the method change the category? If the answer is yes, the method is your positioning, and that is fragile on a platform where users can ask for less of a method.

FAQ

Does Dear Algo let users block AI content on Threads specifically?

Meta's February announcement describes a three-day, topic-level more-or-less request, not an AI opt-out. Pinterest, not Meta, is the platform that has shipped a standing "see fewer" control framed around generative AI. For what the Threads controls cover, and for how long a request lasts, read Meta's newsroom posts rather than a summary of them.

If users control the feed, does ranking still matter?

Yes. The controls sit on top of a ranking system, and Meta's own Threads feed ranking page still describes a machine-learned mix of followed and recommended content. Preferences narrow the pool the ranker draws from. Getting excluded from that pool is worse than ranking badly inside it, which is why category clarity comes first.

Should I stop posting AI-generated video on Threads?

There is nothing published that says labelled content is suppressed on Threads. The argument here is about positioning, not compliance: build an account whose identity is a subject rather than a production method, and the method question stops being the one that decides your reach. The broader picture on labels is in AI content detection and labels.

How narrow should my category be?

Narrow enough that you could write the request yourself and it would sound like something a real person would type. "More short videos about restoring old tools" is a request. "More engaging content from innovative brands" is not. If your category cannot be phrased as a sentence a stranger would type unprompted, keep narrowing.