Threads tests ‘Dear Algo’ feature for feed personalization

Threads is testing a simple, yet brilliant, approach to feed control: speak directly to the algorithm. The social media app is experimenting with a feature where users can write a public post that starts “Dear Algo” and tweak the stuff that turns up in their recommendation feeds for just a little while.

Conor Hayes, who runs Threads, called it a way to use AI to personalize without diving into settings. Request more cooking tips or fewer crypto hot takes, and the system recalibrates your feed in that direction for around three days, ample time to render some new content, but not enough to permanently transform your preferences.

How talking to the algorithm works on Threads

Functionally, it’s lightweight. Slap “Dear Algo” onto a post, tell the system you want more or less of something, and congratulations: your request is now a signal to the recommendation system. The feed then amps or suppresses related content for some time. Engagement during that trial — likes, follows, time spent — helps shape longer-term ranking, serving as feedback that can stick.

There is one catch worth knowing: these posts are public for accounts set to public. That is to say, your intention signal really is a social broadcast. Friends, followers, and creators can all see requests, riff on them, and even repost them. That visibility could help build discovery communities (“Dear Algo, show me more translated fiction”), but it also encourages performative asks and trend chasing.

Why Meta is introducing this conversational control now

Two pressures are overlapping: the demand for user autonomy and regulatory scrutiny of opaque recommendation engines. People crave a fast, intuitive interface that makes a difference. In places with platform accountability laws, regulators have advocated for transparency and a meaningful choice about ranking systems. And there’s a conversational mode that fits both bills — it’s easy for users to explain, but also something you can go and audit.

It is also consistent with Threads’ growth story. The app surpassed 100 million monthly active users in 2024, Meta has said, and keeping those users engaged depends on relevance. Lightweight prompts for them could also help new casual users “cold start” their feeds without having to search for or toggle on buried filters for accounts to follow.

Advantages and disadvantages of public preference signals

The good part is speed and clarity. Instead of hitting “not interested” dozens of times, you tell it a goal once and the system adjusts. And because the prompt is public, creators can jump in and answer questions with recommendations that correspond to the ask, adding human curation to algorithmic tweaks.

The trade-offs are real. Public preference posts might become the next growth hack, with creators gaming trending “Dear Algo” themes and spinning cozy coverage to drive placements. And since the shift is temporary, some users might mistake it for a permanent overhaul. Clear labeling and a visible timer would cut down on confusion, and potentially limit the risk that people feel whiplash as their feeds flip back.

Then there is the question of privacy: letting the world know your interests can divulge a delicate preference. The defaults and education of the feature count. If people can keep some requests secret while still shaping their feeds, adoption may be more widespread.

How it stacks up against competing controls

Most platforms give us reverse controls or categorical ones — “not interested,” keyword mutes, or topic toggles. TikTok introduced a “refresh” button to reset the For You feed. YouTube enables you to mark videos as not relevant, but a Mozilla Foundation investigation of these tools found that they were less effective than hoped, noting that they had success only 12 to 43 percent of the time in tests.

Threads’ method is unusual because it’s proactive and involves natural language. Instead of constantly subtracting single posts, they state some short-term programming bias and then turn the system on to let it fill in. If it does work, we can expect competitors to give similar conversational interfaces a try — particularly because generative AI is making understanding and handling freeform user intent increasingly affordable.

Design details that will make or break the feature

Precision: The model needs to understand subtle requests (“more beginner-friendly running plans, less ultramarathon posts”) without wandering into irrelevant space. Providing light confirmations — recommended topics in response to a query — might also improve accuracy.

Safety: If users ask for sensitive or marginal categories, the system must provide strong guardrails. Safety classifiers and policy-aligned topic constraints should be tested right away.

Abuse resistance: Public cues are an alluring coordination device. Slowing down the flood of content will be crucial, through rate limits, diversity caps, or spam detection to avoid brigading and low-quality floods.

Transparency: People want to know how long the effect lasts, how it affects ranking, and how to turn it off. Simple “active for 3 days” badges and a history of past requests can make it less mysterious.

What to watch next as Threads refines conversational controls

Meta leaders said they had the feature idea because organic “dear algorithm” posts were already floating around, so there is latent demand for this behavior. If testing shows improved session quality and retention — particularly among newer users — anticipate wider rollout and iterations such as private requests, prefilled templates, and topic bundles.

A broader picture is a move toward conversational control of feeds. If saying your preferences works — and it’s an effective and safe way for people to personalize what they see — it could become the new default for algorithmic personalization: fewer settings-scrolling, more telling your app what you want, seeing if you get it.