How to grow on Threads in 2026

Meta's own Transparency Center is unusually direct about this. The Threads feed is “selected, ranked, and delivered by an artificial intelligence (AI) system” where multiple machine-learning models work together [2]. No mystery, no vibes. It's a scoring machine.

There are two feeds, and they behave completely differently:

• For You is AI-ranked. It blends accounts you follow with cold recommendations based on your likes, replies, and reposts [2].
• Following is strictly reverse-chronological. No algorithm at all [2].

Almost all discovery happens in For You. So that's the one you optimize for.

The ranking runs a three-step loop [2]:

1. 1Gather inventory. Collect a pool of eligible public posts.
2. 2Leverage signals. Read input signals about each post and about you, the viewer.
3. 3Rank. Score every candidate by predicted value, then order the feed highest-first.

The signals that actually move the score [2]:

• Reply-click prediction. How likely you are to open the replies. Driven by reply counts, recency, and the author's engagement history.
• Like prediction. Weaker than replies, but it counts.
• Follow prediction. How likely you are to follow the author after seeing the post.
• Profile-click prediction. How likely you are to tap through to the profile.
• Scroll-past prediction. A negative signal. If people scroll past without engaging, you get buried.

One line matters more than the rest. Replies are weighted more heavily than likes, and Mosseri has repeatedly framed reply-driven conversation as the strongest engagement signal [2]. He's said the sum of your replies is roughly as valuable as the sum of your posts [3]. Read that twice. Your comments on other people's threads count about as much as your own posts.

Two more things changed the math in 2025 and 2026. Meta rebalanced the feed to surface more from accounts you already follow and fewer cold recommendations [4]. And ranking shifted hard toward timeliness, topical relevance, and surfacing people who are actively driving live conversation, not recycling old content [4]. Engagement velocity, how fast a post earns replies in the first hour, now beats cumulative day-long totals [2].

There's also the Instagram tie. Your IG activity, including profile views, feeds Threads recommendations directly [2]. They're not separate apps under the hood. They're one graph.

One caveat I'll flag honestly. You'll see blog posts claiming “original content gets 40 to 60% more reach” and “X links lose 30 to 50% reach.” Those numbers appear only in secondary marketing blogs with no primary Meta source [4]. I won't repeat them as fact. The direction is real; the exact percentages are made up.

This is the part I actually live by. Not theory. What I do most days.

1. 1Engineer the post to start a conversation, not collect likes. Open questions. Mild-contrarian takes. Relatable confessions with specific texture. Replies are the strongest signal [2], so write the post as bait for a reply, not a clap. My breakout post on Threads was absurdist relatable humor and it hit a 3.0% engagement rate. But the posts that pull the right audience are concrete practitioner value. A “5 SEO fixes” style post lands with the people I actually want.
2. 2Treat replying as a primary growth channel. This is the move most people skip. Spend daily time replying thoughtfully inside other people's threads in your niche. It surfaces your profile to audiences who've never seen you, and it builds recognition faster than standalone posts [3]. Since your replies count roughly as much as your posts, an hour of good replies can outperform an hour of writing.
3. 3Win the first line. Threads truncates fast. Put the hook in the first line, always. A question or a bold claim that promises value. From my own data: posts that don't hook in the first hour get throttled under roughly 600 views. The For You feed surfaces or buries you on a single question, do viewers engage, or do they scroll past [2]. Weak hook, high scroll-past prediction, dead post.
4. 4Re-engage your own thread fast. When a post gets early replies, respond promptly. Keep the velocity high while the post is fresh. Early reply velocity matters more than day-long totals [2].
5. 5Add a visual. Image, screenshot, GIF, or short video plus a tight caption tends to beat plain text [4]. A designed photo card beats a wall of plain stat text. Multi-format wins.
6. 6Lean on your Instagram graph. Keep IG active and cross-pollinate. IG profile views and engagement feed Threads recommendations directly [2]. Free leverage if you already have an Instagram.
7. 7Ride timely conversations. 2025 to 2026 ranking rewards timeliness and being an active driver of live discussion [4]. My biggest reply counts came from fresh AI-news takes that provoke debate. One take on why 2026 is structurally unique pulled around 130 replies. Un-angled news? Dead on arrival.
8. 8Repurpose, don't invent from scratch. The fastest growth isn't conjuring ideas from nothing. It's taking a proven idea, yours or a competitor's, and re-angling it in your voice on your platforms. Creators who aren't even on X grow on Threads by repurposing viral X threads. Use one topic tag per post, the single closest match, not a stack of hashtags [4].

A workable 2026 rhythm: roughly 3 to 5 thoughtful original posts per week, some practitioners push for at least one a day, plus daily replies inside relevant conversations [4]. I run 1 to 3 posts a day and treat replies as non-negotiable.

Don't flood. Posting 1 to 2 times a day is fine; cramming more can cannibalize your own distribution [4]. And don't go dark, because feed recency and momentum matter, a dry feed loses returning viewers.

Formats that earn replies [4]:

• Open-ended question posts that invite an answer.
• Bold or mildly contrarian one-liners with room to debate.
• Personal observations and confessions with concrete texture.
• Image, screenshot, or GIF plus a short caption.
• Short native video.
• Designed cards over plain stat text.
• Timely hot-topic posts on a live conversation.
• Reply chains where you follow up on your own thread.

Consistency beats the occasional viral swing. Organic is a flywheel, not a lottery. Small, frequent, on-topic. The wins stack while you sleep.

I've made most of these. Here's the list, grounded:

• Optimizing for likes instead of replies. Likes are a weaker signal; a post with many likes and few replies underperforms a conversation [2].
• Reposting or screenshotting with no commentary. Meta limits distribution of unoriginal content [4]. Always add your own take.
• Engagement-bait phrasing. “Comment YES,” “like if you agree,” “reply 1 for the link” all get actively down-ranked [4].
• Hashtag stuffing. Threads discovery uses single topic tags, not stacks [4]. Piling on tags signals low quality.
• Duplicate replies across many threads. Posting the same near-identical reply everywhere gets flagged as spam [2].
• Aggressive bulk follow/unfollow. User reports cite roughly 60 to 200 follows per hour as a flag threshold, though Meta hasn't confirmed it [4].
• Over-posting on a brand-new account. Stricter thresholds apply in the first weeks [4]. Quality and consistency beat volume early.
• Letting viewers scroll past. Weak hooks raise scroll-past prediction, which directly suppresses you [2].
• High-volume political content. Tolerated occasionally, flagged at scale [4].

One nuance on links. They no longer carry the automatic blanket penalty they did at launch [4]. But putting an external link in every single post, especially to low-trust domains, still hurts. Don't make every post a link drop.

Honest version, no pitch.

I built Skedlark because I wanted my agent to do the boring half of this without ever putting words in my mouth. So here's the actual loop. I'm in Cursor or Claude, an idea hits, and the agent captures it through MCP. It drafts a Threads post in my voice, grounded in my real analytics, what actually got replies, when my audience is awake, which formats landed. Then it queues it.

Nothing publishes without me hitting approve. No autopilot. That's the whole point.

The rule I'm proudest of is the anti-fabrication one. The agent may only use proof I actually have. My numbers, my wins, never an invented stat. In a feed drowning in AI slop, sounding like a real person is the entire differentiation, and the fastest way to sound fake is to let a model make up a result. Skedlark is EU-hosted and it posts to Threads among other channels. The agent drafts. I decide. The numbers stay real.