Twitter engagement is down in 2026. Here is what the data actually shows.

Updated 9 November 2026 with a link to our guide on spam labels. On 21 October 2026: every benchmark number was rechecked against its source, and a few were corrected or removed. New sections cover what the open-source ranking code shows can lower reach (the 48-hour AgeFilter, author diversity, the out-of-network discount, negative signals and visibility labels) and how to tell a sudden drop from a gradual one.

Why is Twitter engagement down in 2026? Three things are happening at once.

• The benchmarks are falling. The median brand engagement rate on Twitter went from 0.035% (RivalIQ, 2023) to 0.029% (2024) to 0.015% (Sprout Social, 2025).

• The feed is built to limit any one post. Twitter’s published ranking code drops posts older than 48 hours from For You, discounts each extra post from the same author in one feed load, discounts posts from accounts the viewer does not follow, and subtracts heavily when the model expects a mute, block or report.

• The audience changed. Attention is split across more platforms, readers scroll past AI-shaped posts faster, and the user mix has shifted since 2022.

If your reach fell off in a week rather than a year, that is a different problem. Check Twitter’s Under the Hood report and run a free Twitter shadowban test before you blame the trend. The section on sudden drops below walks through it.

This piece keeps two things apart throughout. Data means a published benchmark or a line of Twitter’s open-source code, cited. Our read marks our interpretation, which you can disagree with. The companion piece on AI content is the state of AI content on Twitter in 2026.

Is Twitter engagement down in 2026? What the benchmarks say

“Engagement is down” hides four questions: are counts per post falling, is the rate falling, is reply quality falling, and is the value downstream (subscribers, customers, conversations) falling? The public benchmarks only answer the second one, and only for business accounts. Here is what they say, checked against each source on 21 October 2026.

• RivalIQ, 2024 report. RivalIQ puts the median Twitter engagement rate at 0.029%, down from 0.035% the year before. The top 25% of brands sit at 0.08%. Category medians run from 0.07% for sports teams to 0.04% for nonprofits. RivalIQ counts likes, retweets and replies divided by follower count.

• Sprout Social, 2025. Sprout Social reports the median falling to 0.015% in 2025, from 0.029% in 2024. Its 2025 Content Benchmarks report gives a higher cross-brand average of 0.16%; averages are pulled up by a few large accounts, so the median is the better guide for a typical account.

• Buffer, October 2025. Buffer’s study of 18.8 million posts from 71,000 accounts (August 2024 to August 2025) found the median engagement rate for non-Premium accounts reached 0% by March 2025, while Premium accounts rose to about 0.4% by mid-2025. For text posts, non-Premium accounts fell to about 0.25% and Premium accounts climbed close to 0.9%.

• The “good” line. Sprout calls a rate above 0.5% solid and above 1% excellent for most brands. Sprout does not give a source for those thresholds, so treat them as a rule of thumb.

Median Twitter engagement rate per post across consecutive annual benchmark reads. It has roughly halved each year, and the axis is a fraction of one percent throughout, so the decline compounds off an already-low base. Sources: RivalIQ (2023, 2024) and Sprout Social (2024, 2025).

Twitter engagement-rate benchmarks, rechecked 21 October 2026. Medians unless marked. Definitions differ by publisher, so compare within a row’s source, not across sources.

Two cautions. First, these are business-weighted samples measured with each vendor’s own definition, so the year-on-year change inside one report means more than any comparison across reports. Second, the base was already tiny. A move from 0.035% to 0.015% happens inside a fraction of one percent, which is why Buffer’s “median non-Premium account at 0%” lands harder than the percentages.

Our read on the Premium gap: Buffer measured a real difference, but the ranking code below has no Premium or verified term in its scoring formula. The gap may come from somewhere the code does not show (reply placement, who subscribes, how often they post). We cover the decision in X Premium: is it worth it at all?.

Why are impressions down on Twitter? What the open-source code shows

Twitter publishes its For You ranking code at xai-org/x-algorithm. The weights and the visibility-filtering code arrived in the 13 August 2026 release; everything below was read from the repository on 21 October 2026. Five parts of it can lower a post’s reach without anything being “wrong” with your account.

1. The 48-hour AgeFilter

Data: a filter called AgeFilter drops any For You candidate older than 48 hours (MAX_POST_AGE = 48 * 60 * 60 seconds). The small-account boost has a shorter window still: a post must be 2 hours old or less. The popular-posts list looks back 24 hours.

Our read: a post earns almost all of its For You reach in its first two days, so a week where you posted less is a week of fewer impressions, full stop. The detail is in the 48-hour rule in the ranking code.

2. Author diversity

Data: after scoring, each extra post from the same author in one feed load is multiplied down. With the code defaults (decay 0.5, floor 0.25) the first post keeps x1.0, the second x0.625, the third x0.4375, then it heads toward x0.25. These defaults sit in the vm-ranker service and are not in the file synced from production, so the live values are not confirmed. Separately, the in-network source takes at most 50 original posts and 30 replies per followed author.

Our read: posting five times a day does not give you five times the reach in a follower’s feed. We found no posting-frequency penalty, just diminishing returns inside each feed load. More in author diversity and your third post of the day.

3. The out-of-network discount

Data: posts shown to people who do not follow you are multiplied by 0.75 (a code default; the production value is unclear from the code). The same discount also applies to replies and reposts from accounts the viewer does follow. And a filter removes out-of-network replies and reposts from For You altogether.

Our read: reaching non-followers is harder by design, and replies mostly travel through the conversation view rather than strangers’ For You feeds. See how your posts reach people who don’t follow you.

4. Negative signals

Data: the model predicts how likely each viewer is to take each action, and the weights turn those predictions into a score. The negative weights are large: report -234.0, mute author -58.8, not interested -52.272, block author -31.2 and not dwelled -0.02. A code comment added on 14 August 2026 says the weights multiply predicted probabilities, not counts, so “one report cancels 468 likes” is false. Any post with a net-negative predicted score ranks below every post with a net-positive one.

Our read: the bait-style post that earns likes from fans but irritates everyone else can lose more than it gains, especially out of network. Full detail in negative signals, visibility labels and Under the Hood.

5. Visibility labels and Under the Hood

Data: visibility filtering runs after ranking. It can remove a post (verdicts Drop, Tombstone or NotEvaluated); its Downrank and Avoid verdicts keep the post and do not change its score anywhere in the feed code. For recommendations to non-followers, extra drop-only rules apply, including account and post labels such as do not amplify, spam high recall, malicious URL and several adult-content labels. Those labels hide a post from people who do not follow you, while your followers can still see it.

Data: Twitter’s Under the Hood report at x.com/i/jf/under_the_hood shows your post labels, account labels, what each means and its effect on visibility. Since 18 September 2026 it also shows legal withholding, including which country.

Our read: if your non-follower reach collapsed but your followers still see you, a label is the first thing to rule out, and Under the Hood is where to look.

What the code says about common reach claims (xai-org/x-algorithm, read 21 October 2026)

Why did my Twitter engagement drop suddenly?

Benchmarks decline slowly. A drop that happened in days usually has a specific cause. Work through these in order.

1. Open Under the Hood. Look for post or account labels and their effect on visibility at x.com/i/jf/under_the_hood.

2. Run a search test. Search an exact phrase from a recent post while logged out, or use the free shadowban checker. If your posts are missing from search, read how to test for a Twitter shadowban and fix a search ban.

3. Check what changed on your side. Fewer posts means fewer posts inside the 48-hour window. A run of bait, rage or off-topic posts can raise predicted mutes and not-interested taps.

4. Check your mix. If most of your reach came from non-followers, the 0.75 discount and out-of-network filters make it more fragile than follower reach.

5. Check any connected app. Twitter’s search rules name third-party apps that post similar content across accounts as a cause of search filtering.

If none of those apply, you are probably looking at the gradual trend the benchmarks describe, plus ordinary week-to-week noise.

If you suspect a label is part of the drop, run these five checks before you post.

The five causes behind the gradual decline

The slow, platform-wide decline has at least five causes operating together. Single-cause takes (“AI killed engagement”, “the algorithm got worse”) each point at one real cause and miss the rest. Causes 2 to 5 are our read, not measured data.

1. Ranking changes. The code shows what moved recently: a +15 reply boost for original posts between mutual follows (tested from 10 July 2026, now 15.0), the like weight at 0.66 (it was 0.5 in July), and the discounts above. Each change shifts reach between kinds of posts. The current weights are in the 25 signals and weights that rank your posts.

2. Attention fragmentation. TikTok, Threads, Instagram, LinkedIn and Substack all compete for the same hours, so each platform’s share of attention shrinks. Threads is the clearest case: our Threads vs Twitter comparison sets the two platforms’ user numbers side by side.

3. AI content saturation. Readers have learned to scroll past template hooks and beige bullet middles. The mechanics are in why all AI-written tweets sound the same.

4. Audience shift. The user mix changed after the late-2022 ownership change, so benchmarks compare a different audience against an older baseline.

5. Maturation. People have seen more posts, so the bar for a like keeps rising. This is the hardest cause to measure because it overlaps with the other four.

Which Twitter metrics are falling, and which are not?

The benchmarks measure engagement rate. For the other metrics we only have patterns seen across accounts, so read this list as our read rather than data.

• Likes per post: the metric most often reported as falling.

• Replies: bifurcating. Reply counts on large broadcast accounts look weaker; accounts that reply back and post specific opinions hold up better.

• Reposts: people are choosier about what they amplify to their own followers.

• Bookmarks: people save reference-grade posts. Note the code fact: bookmarks are predicted but carry no weight in the ranking formula, so a bookmark helps the reader, not your reach directly.

• Impressions: move with ranking changes, posting volume and the 48-hour window, so they are noisy week to week.

• Downstream value: newsletter sign-ups, customers and real conversations. The benchmarks do not measure it, and it is the one that decides whether the decline matters to you.

Where is the decline worst, and where is it absent?

Our read: the compression looks worst for large brand accounts posting templated content, news accounts that rely on broadcast reach, and niches where AI-shaped posts became the default in 2024 and 2025 (marketing, productivity tips, business advice). It looks mildest for creators whose posts are recognisably their own, niche communities with strong conversation, and small accounts.

Small accounts have a code-level reason too. The data: on each feed request, at most one original post from an author with 50,000 followers or fewer, under 2 hours old and with under 200 home views, can be lifted to about 16th place. The benchmark median says little about your account. Your own trend in your own niche does. For the long view, see the 3 fundamentals of Twitter growth.

What the data does not say

• “It is all AI content.” One cause of five. The audience-side story is in can your audience tell you’re using AI.

• “Posting more fixes it.” The author-diversity discount means extra posts in one feed load earn less each, and templated volume trains readers to scroll past. What the frequency studies say is in how often you should post on Twitter in 2026.

• “The platform is dying.” A falling engagement rate is not the same as falling usage. The benchmarks measure rate, not users.

• “The benchmark is my number.” The 0.015% median describes business accounts. Treating it as your target or your verdict is a category error.

Do the usual engagement tactics still work in 2026?

Most “increase engagement” listicles repeat the same tips (add an image, a GIF, one hashtag, ask for the retweet) with triple-digit multipliers attached. The ones we could trace come from Twitter blog posts and third-party studies of roughly 2013 to 2015, a different feed with a different audience. We could not find a current source for the specific percentages, so we no longer repeat them.

The tips are not wrong, just small. Moving a post from 0.015% to 0.02% is a real relative gain on a number that still rounds to almost nothing. The bigger lever is whether the post is worth stopping for: specific, opinionated and recognisably from one person.

What the accounts holding engagement do differently

These are patterns we see across accounts that are holding steady, so treat them as our read.

1. Fewer, better posts inside the window. A few specific posts a week, spread so each gets its 48 hours, rather than bursts the diversity discount trims. Use the best time to post tool to spread them.

2. Replies as a channel. Reply weight is 5.0, rising to 20.0 for original posts between mutual follows. The playbook is in the smart reply guy strategy.

3. Refusing AI tells at draft time. No template hooks, no beige middles, no generic call to action. The checklist is how to avoid the AI tells.

4. Not provoking the negative signals. Bait that earns likes from fans and mutes from everyone else is a bad trade under these weights.

5. Patience. Ninety to 180 days is a realistic arc for growth that compounds. See how to grow on Twitter without buying followers.

The lever that holds up against the engagement-rate decline is reading like a specific person. audenAI, the brain inside VoiceMoat, drafts from your own posts so your content reads like you, not like the AI-shaped register readers have learned to skip.

What should you measure instead of engagement rate?

If the median rate is a fraction of a percent and falling, it is a noisy target. Four measures tell you more.

• Engagement per follower. A 2,000-follower account where 60 people reliably reply is healthier than a 50,000-follower account at the median. Work yours out with the free engagement rate calculator.

• Saves and shares. Bookmarks show a post was worth keeping, and share via copy link is the largest positive weight in the formula (20.0).

• Profile visits and follows from replies. The broadcast metrics miss reply-driven growth entirely.

• Off-platform conversion. Subscribers, trials and DMs that become real conversations.

More on tracking these is in Twitter analytics that matter for creators.

How to read the next benchmark report

1. Find the methodology first and note the engagement-rate definition (per follower or per impression).

2. Check the sample: business accounts, creators, or both, and in which categories.

3. Check the baseline: the same vendor’s prior-year report on the same method is the only fair comparison.

A report that hides its method or compares against an unstated baseline is marketing, not measurement.

The short answer

Twitter engagement is down in 2026 on the benchmarks that measure it: the median brand rate halved from 0.029% to 0.015% between 2024 and 2025. Underneath, the ranking code limits any single post (a 48-hour window, author diversity, a 0.75 out-of-network discount, heavy negative weights, and labels that hide posts from non-followers), and the audience itself has changed. A sudden drop needs a diagnosis (Under the Hood, a search test); a gradual one needs better, fewer posts and real replies.

If you want help with the writing side, VoiceMoat is a Personal Brand OS for LinkedIn and Twitter. audenAI, the brain inside it, learns from your own posts across the 10 signals of voice, drafts in Studio, checks drafts for Style and Stance, and avoids the usual AI tells; any word you add to Your rules is never used. It does not change or game any ranking. audenAI suggests. You decide.