How the Threads algorithm works in 2026

How the Threads algorithm works: the short answer

The Threads algorithm is a set of machine learning models that ranks the For You feed for one viewer at a time. For each post it predicts whether you will like it, reply, tap into it or the author's profile, how long you will look at it, and whether you will scroll past. Posts that start a conversation quickly, from accounts people already engage with, tend to win.

Meta documents the inputs in its Threads Feed AI system card (last updated March 7, 2025), but not the weights. Every claim below is tied to that card or another named source, and unconfirmed claims are labeled as such.

How the For You feed is built

Threads has two home feeds. Following shows only accounts you follow, newest first, with no recommended posts. For You is the ranked one.

The system card describes three steps. The system gathers a slice of public posts plus everything from accounts you follow that passes Meta's "quality and integrity rules", reads signals about those posts and about you, and ranks them by the value it predicts each one will have for you.

That means a post has no single score. The same post about invoicing can rank high for a freelance designer who replies to money threads every week and never reach someone who mostly likes sports clips.

The ranking signals Meta lists

Look at what is missing. The card lists no prediction for reposts, quotes or follows. Those actions may matter somewhere else in the system, but Meta hasn't said so in public.

Why replies carry so much weight

The reply prediction is fed by activity under the post over the last hour and six hours, down to replies to replies. A post that starts a conversation keeps feeding the model fresh input; a post that collects a few likes and goes quiet doesn't.

Meta's own people point the same way. Adam Mosseri, in a 2024 interview with Platformer: "If you're really trying to grow your presence, you should reply much more than you post. And the sum of all your replies is about as valuable as the sum of all the value of all your posts." In October 2024 Business Insider interviewed Emily Dalton Smith, then head of Threads, and summed up Meta's position: the company was "sticking by comments" and by posts that "drive conversation" as a way to decide what gets recommended.

Buffer's analysis of 128,000 Threads posts found that posts where the author answered comments got about 42% more engagement than that account's other posts, with a positive effect for roughly two thirds of profiles. Buffer calls it correlation: authors may simply reply more under posts that were already doing well.

There is a limit. In October 2024 Mosseri posted that Threads had seen "an increase in engagement-bait" and, replying under that post, wrote that "not all comments or replies are good". Replies count, and Meta is working against people who fake them.

Compare two replies under a post about a client who vanished after the first invoice. "So true! 🔥" adds nothing and gets scrolled past. "Happened to me last spring. Now I send a two-line recap after every call, so the scope is in writing before money comes up" gives the author something to answer.

The slow part is finding enough threads where your reply belongs. That is what Replyvero, a Chrome extension for replying on Threads, does: it reads the feed or a keyword search, keeps posts in your niche and drafts a short reply. By default each draft waits for your approval before anything is sent. The reply guy strategy guide covers the manual version.

Followers first, and a short shelf life

On November 21, 2024, Mosseri announced that Threads was "rebalancing ranking to prioritize content from people you follow" and told creators to expect "unconnected reach" to fall and "connected reach" to rise. In October 2026 Connor Hayes, head of Threads, posted about another change that "could increase the posts you see from those you follow", which for creators should mean more reach to followers on average.

If your reach among strangers dropped, check this first. A smaller audience that actually replies is worth more now than in early 2024.

Timing matters as well. Dalton Smith told Business Insider that "most of the content will be distributed in just a few hours over the course of the day". A summary of a February 2026 interview with Hayes says new posts are first tested on a small audience and widened if they perform, and that posts older than three days are no longer recommended to non-followers. That is a third-party summary, so treat it as reported, not documented.

What reduces reach

Links

Meta has repeated that it doesn't downrank links: in February 2024, in November 2024 with the caveat "we don't place much value on p(click) and people don't like and comment on links much", and in June 2025, when Mosseri said links were now "ranked properly". Engagement is another matter: in Buffer's 2026 format study, link posts had the lowest median engagement rate on Threads, 2.34%. Write posts that are worth reading without the click, and use the five bio links Threads now allows.

Engagement bait

Meta's spam policy bans like/share-gating: requiring likes, shares, follows "or any other public-facing form of engagement" to get access to content. "Comment GUIDE and I'll DM it to you" fits that description uncomfortably well. How hard ranking punishes it is not public.

Recycled posts and comment pods

Instagram's recommendation guidelines exclude content "largely repurposed from another source" without added value, and "coordinated comment networks intended to artificially drive engagement". They are written for Instagram, but the official Threads account linked to them in October 2024 when it named two things that limit distribution on Threads: content that goes against those guidelines, including clickbait, and unoriginal content. An engagement pod is a coordinated comment network.

Sensitive topics

The same guidelines keep some categories out of recommendations. Since January 2025 political content can be recommended again, and each user picks how much of it they see (Engadget).

Penalties on the account

Threads has an Account Status screen that shows removed posts, demoted posts and content that can't be recommended. If you suspect a "shadowban", look there before you rewrite your strategy.

Topic tags and communities

Each post can carry one topic tag. Mosseri explained in 2023 that the single tag was deliberate, to focus tags "more on communities and less on engagement hacking". Meta's March 2025 update says posts that include a topic "generally receive more views than those that don't".

Since October 2025 Threads has had Communities, public spaces built around popular topics, more than 200 of them by December 2025. You post into one by adding its topic. Members who are followed by other members, post regularly and "keep conversations going in the replies" can get a community champion badge, and Meta says it is working on ranking inside communities and in For You.

In practice, pick the tag that matches an existing community in your niche, spelled the way people search for it. "Freelance Design" will find readers; "random thoughts" won't.

What viewers control

Per the system card, viewers can unfollow, mute, report, switch to Following, or hide a post, which "helps to limit similar content". Since February 2026, people in the US, UK, Australia and New Zealand (the countries Meta listed at launch) can also write a public "Dear Algo" post asking for more or less of a topic, which adjusts their feed for three days. In June 2026 Meta added a private version called Your Algo, rolling out in the same countries plus Canada.

For you this means two things. Hides and scroll-pasts count against a post, and when someone asks for "more posts about Figma plugins", a post that names Figma plugins outright is easier to match than one that says "this tool". Name the niche, the tool and the problem.

Format, length and posting frequency

The hard limits come from Meta's API documentation: 500 characters of main text, one topic tag, up to five links. Longer pieces can go into a text attachment of up to 10,000 characters.

On format, the best public data is Buffer's: median engagement rate on Threads was 5.55% for video, 4.55% for images, 2.79% for text and 2.34% for links. The sample is accounts that schedule through Buffer, so a personal text account may see something different.

On length, neither Meta nor any large study has published an optimal number, so be suspicious of precise "ideal character counts".

On frequency, Meta told Business Insider in 2024 that it recommends two to five posts a week and that frequent posting is "linked to higher impressions per post". Timing data conflicts: Meta said weekend posts "may help drive higher engagement", while Buffer's study of 2.5 million posts found weekday mornings worked best. Your own Insights show when your followers are online, and that beats both.

Spam and automation rules

Threads content falls under Meta's Community Standards. The spam policy bans "posting, sharing, engaging with content... either manually or automatically, at very high frequencies", and says Meta "may place restrictions on accounts that are acting at lower frequencies" when there are other signs of spam, such as repetitive content. The inauthentic behavior policy covers fake and deceptive accounts.

Frequency is only part of it. Threads' Terms of Use restrict automated access to the service, so any automation, Replyvero included, carries a risk of restrictions at any pace. Meta also publishes no numeric thresholds, so no tool can promise you're safe.

What lowers the risk is the opposite of what the policy describes: a human pace, no repeated replies, no comment rings between your own accounts, and reading what goes out under your name. If you use a reply tool, keep drafts on and the volume modest. Replyvero, for example, caps sending at 10 replies an hour and 100 a day with random pauses, and stops for 24 hours if Threads shows a warning. The limits lower the risk; they don't remove it.

What creators believe but Meta never confirmed

Many "Threads algorithm" articles, especially on tool vendors' blogs, mix the system card with folklore. These claims keep coming up, and none of them comes from Meta:

Vista Social ran a similar check and also found no public Meta source for the "30 reactions in an hour" and "three times more than likes" claims. Some might be true, but nobody has shown it in a way you can check.

A practical checklist

1. Put the point in the first line, so readers stop instead of scrolling past.
2. End with something people can answer: a specific question or a claim they can disagree with.
3. Reply to the people who comment, with substance, in the first few hours.
4. Spend more time replying in other people's threads than posting your own.
5. Add one topic tag that matches an existing community.
6. Name your niche, tools and problems in plain words.
7. Make every post worth reading without its link; keep links in the bio or a text attachment.
8. Write about what is happening now.
9. Skip bait: no "comment YES", no gated freebies, no pods.
10. When reach drops, check Account Status first and your audience hours in Insights second.
11. If you automate any part of this, keep it slow, varied and reviewed by you.

To turn replies into paying work, read how to get clients on Threads; for a broader plan, see how to grow on Threads. If you want help with the reply part, Replyvero's plans start with a free one-hour trial.