Twitter Analytics Tools: 0.57% Median ER

On 6 November 2026, original posts from 20 public X accounts had a median engagement rate of 0.57% (363 posts; exact ratio 0.005652997011893061). The rate is (likes + comments + shares) / views. People searching for the best twitter analytics tools are trying to buy a benchmark.

We kept originals published on or after 9 October 2026, at least 6 hours old when pulled, with views above 0. The account list, including the backup order, was locked before any study call. Reposts and replies are outside this median. Saves are outside the numerator. The rate is not clamped. The median is Hyndman-Fan type 7. With 363 rates, h = 182, so 0.57% is the 182nd sorted value, not a blend of two neighbors.

The number above is the benchmark this page measured. Rival IQ and Socialinsider publish X rates that divide by followers. Mentionlytics, Sprout Social, and SocialPilot publish roundups with no account list, no post count, and no shared formula. A view-based median and a follower-based rate are different measurements. This page does not rank 0.57% against them.

What median engagement rate do original posts get?

The 0.57% figure is the middle of 363 post-level rates. It is not the average of 20 account scores. Accounts that returned more eligible posts pull the pooled median toward themselves. The walk returned 560 deduped timeline rows. After the date, age, views, and original-post rules, 363 remained.

Component medians below are medians of each count on its own. Dividing those medians does not produce 0.57%. Likes at the 25th percentile (58.5), views at the 75th (242,455.5), comments at the 75th (129.5), and shares at the 75th (231.5) are type-7 interpolations, not counts a single post had. Means in the last row are rounded to one decimal. Unrounded means: views 433,583.9035812672, likes 4,547.137741046832, comments 167.1046831955923, shares 283.05234159779616, saves 429.88980716253445.

Display percents are rounded. Unrounded engagement-rate ratios: minimum 0.00019833849955263035, 25th percentile 0.0026141180063415535, median 0.005652997011893061, 75th percentile 0.012741240998051514, maximum 0.07328957009797626, mean 0.010686562476808825. None of the 363 headline rates was above 1. The minimum, 0.02%, is a pinned Figma post from 19 October 2026 with 3,609,990 views. Seven pinned originals stayed in the set, and that Figma post sets the floor.

The mean on the same 363 posts is 1.07% (exact ratio 0.010686562476808825). The maximum is 7.33%. The mean is not the headline. The median of the 20 account medians is a separate number, 0.69% (exact ratio 0.006853345474006916). The two central account medians are linear at 0.65% on 12 posts and levelsio at 0.72% on 7 posts.

129 of the 363 posts were under 24 hours old. Median age was 31.0 hours (exact 30.97196611111111). Rates on posts that young are still moving.

Saves were a finite number on all 363 headline posts. Adding them to the numerator moves the median from 0.57% to 0.66% (exact ratio 0.006560364464692483). That sensitivity is not the headline. The formula was locked before the pull, and saves stayed out.

Follower-based rates already in circulation are a different statistic. Maria Mundaden's Rival IQ piece on 8 August 2024 defines engagement rate as engagement divided by followers and puts the all-industry Twitter median at 0.029%. Elena Cucu's Socialinsider benchmarks, published 16 January 2026 and modified 4 November 2026, are also by followers. The year table lists X at 0.15% for 2024 and 0.12% for 2025, and the 2025 prose calls 0.12% an average. The Q3 2026 callout puts the X average at 0.10%, from 198 million posts and 1.16 million brand accounts. The introduction of that same page says 70 million posts. Both sample sizes are on the page. The methodology section defines the X rate as the sum of likes and reposts on a post, divided by that profile's followers, then averaged across posts. Replies are not in that numerator. This page does not treat 0.57% as higher or lower than 0.029%, 0.15%, 0.12%, or 0.10%.

Why is the news median 0.27% while creator posts sit at 2.08%?

These are post-level medians inside each niche, not medians of the account medians. News supplied 156 of the 363 posts, so the pooled 0.57% sits nearer the newsroom rate than the creator rate. Brand is four accounts because Starbucks returned no rows.

The creator accounts are MrBeast, Marques Brownlee (mkbhd), Pieter Levels (levelsio), Alex Hormozi, and Sahil Bloom. Alex Hormozi's account median is 4.19% on 32 posts. That is one account. On their own posts, MrBeast is 0.95% (10 posts), mkbhd is 1.43% (28), levelsio is 0.72% (7), and Sahil Bloom is 1.66% (6). The 2.08% figure pools the creator posts. It is not a law about creators, and it is not the median of those five account medians.

Why do accounts over 10 million followers land at 0.31%?

The largest follower band has the lowest post-level median. Bands were fixed before the pull. Each eligible post from an account inherits that account's band. Nothing under 100,000 followers was on the list. Linear, at 115,585 followers, is the smallest account, so the pre-registered under-100,000 band is empty. Do not read 0.57% as the rate on a typical small account.

The 10 million band is nytimes, TechCrunch, wired, WSJ, Netflix, Spotify, MrBeast, and BBCWorld. verge, at 4,765,006 followers, is the news account outside that band. A post from a mega-account has the same weight as a post from Linear, and a newsroom that contributed more eligible posts has more weight in the headline. News is concentrated in the top band, so 173 posts there can land at 0.31% while 132 posts in the middle band land at 1.14%. This pull does not separate follower count from niche.

How much of the engagement sum sits on reply posts?

0.17% of the public likes, comments, and shares in the date window sat on posts these accounts published as replies: 3,245 of 1,862,651 (exact ratio 0.001742140637188609). The cut uses every in-window original and reply with finite likes, comments, and shares. It does not use the 6-hour age rule. Reposts are not in the denominator.

All 43 replies were replies to the account's own username. None were replies to another account, so every reply row in this pull is a self-thread. The 0.17% figure does not measure audience replies under an original post. Those audience replies are already the comment count on the original. On the 363 headline posts, the median comment count is 33 and the median like count is 303.

Repost wrappers are a bad engagement record. On 79 of the 80 in-window reposts, likes were 0, comments were 0, and shares matched the reposted post. Those share counts belong to the earlier post, not to a new reaction on the wrapper. The headline rate and this 0.17% share both leave the 80 reposts out.

Which four accounts reached the 9 October boundary?

Four of the 20 accounts reached 9 October 2026 inside two pages. The other 16 stopped while has_more was still true. A row count from a truncated walk is not a week, and a newsroom's last day is not a posting rate. This page does not publish a panel posts-per-week figure.

The four full windows below are the accounts whose history fit in two pages. They are not the panel. Weeks run from 2026-10-09T00:00:00.000Z to that account's capture time, divided by 7 days. Each value is slightly over 4.007. A rounded divisor of 4 will not reproduce the last column. The rates are printed unrounded.

Nike's full window is two originals. Linear published 13, MrBeast 10, and Alex Hormozi 32. WSJ shows why the other 16 stay unlabeled: 12 eligible posts inside an 8.8 hour slice (exact span 8.845845 hours). Stretching that slice into a weekly rate would invent a week.

How did we measure this, and can you reproduce it?

The pull ran on 6 November 2026, from 2026-11-06T01:17:13.078Z through 2026-11-06T01:18:40.711Z. Every timeline row came from the SocialCrawl API, GET /v1/twitter/user/tweets, with since=2026-10-09, max_pages=2, and max_credits=2. Each call sent Cache-Control: no-cache and no label parameter. Default labels on the payload were ignored. Every study response was HTTP 200, uncached, and not retried. Credits spent were 42. A schema probe on nytimes at max_pages=1 cost 1 credit and is not in the median. The study re-fetched nytimes. Study credits_used then summed to 41, at 1 credit per page walked, across the 20 measured accounts and the empty Starbucks call.

Twenty primary accounts were locked before any study call, chosen because they publish under a known newsroom, brand, product, or creator name, not because of their stats. No account was swapped for its engagement rate. Starbucks came back HTTP 200, credits_used 1, data.empty_reason nothing_new_since, with the walk stopped at since. An empty walk to that boundary does not prove the account never posts. That limit is what a response does not prove. BBCWorld, the first locked backup, was called only for that empty result and counted as news. Brand stayed at four accounts. Backups not called: Airbnb, vercel, garyvee, buffer.

A row is a repost when post.ext.retweeted_post is a non-null object or the text starts with RT @. A row is a reply when it is not a repost and post.ext.in_reply_to_id is set. Every other row is an original, including quote posts. The payload has no quoted-post object. quote_count is how many times that row was quoted, and it is not in the formula. conversation_id was not used to classify a row. Zero in-window originals had a conversation_id different from the post id.

The formula, fixed before the pull, is (likes + comments + shares) / views. The fields are post.engagement.likes, post.engagement.comments, post.engagement.shares, and post.engagement.views. Comments on this endpoint are the public reply count. Shares are the public repost count. A missing, non-finite, or non-positive view count drops the post. A missing or non-finite like, comment, or share count drops it too. An explicit 0 is a real zero. We kept originals on or after 9 October 2026, at least 6 hours old, with views above 0. The 6-hour rule was fixed first. It left 363 posts, so a 1-hour fallback was not used. 74 otherwise eligible originals were under 6 hours and stayed out, 27 of them from WSJ. Zero in-window originals lacked a positive view count. Zero had views with a non-finite like, comment, or share count.

Across the 20 measured accounts, data.dropped summed to 0 and duplicate ids removed were 0. Four walks returned one fewer row than they fetched, with dropped still 0: stripe 29 fetched and 28 returned, NotionHQ 24 and 23, figma 35 and 34, levelsio 17 and 16. The median uses returned items only. The missing row is not estimated. Class counts on the 560 returned rows, all of them inside the date window: 437 originals, 43 replies, 80 reposts. Views were finite on 559 of 560. The gap is an excluded MKBHD repost, id 2112986725526176075, which is also the only null computed.engagement_rate in the 560.

We did not call /v1/twitter/profile, /v1/twitter/profile/full, or /v1/twitter/tweet. Follower counts are post.ext.author_followers on the timeline row. Author keys observed were avatar_url, display_name, username, and verified. There is no lifetime likes field on these rows.

The computed fields use the same post formula, (likes + comments + shares) / views, rounded to 6 decimal places, with a raw ratio above 1 clamped to 1. On the 363 headline posts the field was present on every row. None had a numerator of 0. All 363 matched the unrounded ratio rounded to 6 decimal places. The largest absolute gap was 4.992174840380992e-7. The headline uses the unrounded ratio. Do not quote the clamped field. Two excluded reposts were clamped to 1: Stripe id 2117028402175594506 at raw ratio 6.535211267605634 (1,562 views, 10,208 shares copied from the reposted post) and MKBHD id 2110337604126204189 at raw ratio 109 (1 view, 109 shares copied from the reposted post). Those are not 100% engagement rates, and they are not in the 363.

Twitter analytics tools, as the roundup pages sell them, advertise link clicks, profile clicks, owner impressions, and an organic-versus-promoted split. This public-post formula can reproduce likes, the public reply count, the public repost count, and the public view count. Those click counts and the organic-versus-promoted split are not in the response. The X metrics reference (dateModified 25 July 2026) marks URL clicks, profile clicks, an engagements total, video quartile playback, and the organic-versus-promoted split as non-public, and only for posts from the last 30 days.

curl "https://www.socialcrawl.dev/v1/twitter/user/tweets?handle=HANDLE&since=2026-10-09&max_pages=2&max_credits=2" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Cache-Control: no-cache"

With a SocialCrawl key, classify each row with the rules above, keep originals on or after 2026-10-09 with views above 0 and age at least 6 hours, and take the type-7 median of (likes + comments + shares) / views. A later rerun will not match 0.57%. Posts under a day old were still collecting views, and the timelines move. The route is X user tweets.

What does an X analytics dashboard leave out?

An X analytics dashboard, on the product pages growth marketers compare, is the owner's screen: impressions, link clicks, profile visits, video playback, and a paid-versus-organic split. This pull is a public timeline. The overlap is likes, the public reply count, the public repost count, and a public view count.

A Twitter analytics dashboard page from Sprout writes engagement rate as interactions per impression, with likes, reposts, replies, and clicks in the numerator. Clicks are not a field we have, so that formula is not rerun here. Hootsuite's analytics page names engagement rate and does not publish an X formula. X does not publish an official engagement-rate formula on the metrics page.

X defines an impression as a post appearing on a screen. The same person counts again if the post appears twice. That page marks impressions public, available with a bearer token. The private-metrics example on the same page also returns non_public_metrics.impression_count, with the same example value as the public field, 5,432. The requirements table still says public_metrics needs only a bearer token and non_public_metrics needs user context on posts the user owns. Both statements are on the page. This study does not pick one, and it does not claim the timeline view count equals either field.

We did not fetch the native dashboard. help.x.com and business.x.com returned HTTP 403 during research, so this section does not describe current dashboard screens, plan gates, or prices.

Is Twitter analytics free?

A search for twitter analytics free is usually someone opening X's own product. The query x analytics free is that same product after the rebrand. twitter analytics, the higher-volume query, is mostly the same screen. This page did not verify a price. help.x.com and business.x.com returned HTTP 403 on 6 November 2026, so a dollar figure or a Premium gate would be a guess.

The likes, public reply counts, public repost counts, and public view counts behind the 0.57% median do not require the account owner's dashboard. They are the public counts on the timeline rows. Owner-only metrics from the last 30 days, including URL clicks and profile clicks, are a separate product from this median.

What does Twitter account analytics include on a public account?

On a public account, Twitter account analytics in this pull means the counts on the timeline, not a login to someone else's dashboard. Each row below is one measured handle. Follower count is post.ext.author_followers on the newest row in that response with a finite value. Inside one response the count moved by at most 1,201 (Netflix, from 29,328,774 to 29,329,975). Bands use the table count. ER posts is how many originals entered the headline median. Median ER is rounded to two decimal places. Nike (exact ratio 0.01428979503007579) and mkbhd (exact ratio 0.014279530455083948) both display as 1.43% and are not the same rate. Span hours in the table are rounded to one decimal.

Reached since means the 9 October boundary was inside the two-page cap. Truncated means the cap stopped the walk while more pages remained. A truncated count is not a posting rate.

WSJ is 0.10% on 12 posts inside an 8.8 hour slice. The other newsroom medians run from 0.24% to 0.34%: TechCrunch 0.24%, BBCWorld 0.27%, nytimes 0.29%, wired 0.32%, verge 0.34%. Alex Hormozi is 4.19% on 32 posts with the window fully covered. High-volume newsrooms in this table are about a day of posts (WSJ 8.8 hours, verge 24.8 hours, nytimes and TechCrunch 31.0 hours), not a month. Starbucks is absent. The empty result does not, by itself, prove that account never posts.

Where does this 0.57% stop being a benchmark?

The sample is 20 famous public accounts, picked before the pull: newsrooms, brands, product accounts, and creators. It is not a random draw of growth-marketer accounts, and the smallest account is Linear at 115,585 followers. Two pages, newest first, mix two shapes. Nike, linear, MrBeast, and AlexHormozi reached 9 October. The other 16 are a recent slice. There is no panel posts-per-week number. Rival IQ's 3.31 median Twitter posts per week, from the 8 August 2024 piece, is that report's industry median. It is not a count from these 20 timelines.

129 of the 363 headline posts were under 24 hours old, and the median age is 31.0 hours, so views and rates can still change. Leaving out 74 posts under 6 hours, including 27 from WSJ, does not make the remaining rates final. The denominator is the public view count on the timeline endpoint. X's impression is a screen appearance, with repeats counted, and this pull does not claim the two numbers match. Seven pinned originals stayed in. They are a small share of the 363, and one of them is the minimum rate.

The 0.17% reply share is the share of public likes, comments, and shares that sat on self-thread posts these accounts published. It does not measure audience replies under original posts, and it is not another account's private analytics. Follower counts are snapshots on the timeline rows, not a separate profile call. A mega-account with a low rate weighs the same per post as a smaller account, and more in the headline when it contributed more eligible posts. Quoting a clamped computed.engagement_rate of 1 on the two excluded reposts would report a 100% rate from copied share counts. Starbucks is missing because the walk returned nothing after the since boundary.

Frequently asked questions

Are Twitter analytics tools free?

Public like, reply, repost, and view counts in this study do not require the account owner's dashboard. The median those counts produced on 6 November 2026 was 0.57% across 363 original posts from 20 public accounts. This page does not name a price, so it does not answer whether Twitter analytics tools are free. help.x.com and business.x.com returned HTTP 403, so native X Analytics pricing was not verified, and no third-party price was collected either.

Is there a Twitter analytics checker for public posts?

You can recompute the public rate. It is not a login checker and not another account's private dashboard. Call our data API at GET /v1/twitter/user/tweets on https://www.socialcrawl.dev, with the handle, since=2026-10-09, max_pages=2, and max_credits=2. Keep originals on or after 9 October 2026 that are at least 6 hours old and have views above 0, then take the type-7 median of (likes + comments + shares) / views. On 6 November 2026 that median was 0.57% on 363 posts. A later day will differ, because 129 of those posts were under 24 hours old.

Can you see Twitter analytics for another account?

Native analytics for another account stays with that account's owner. This study measured public posts only. The 0.57% median is 363 original posts from 20 public accounts, using likes, public replies, public reposts, and public views. It does not open someone else's dashboard. URL clicks, profile clicks, the owner engagements total, video quartile playback, and an organic-versus-promoted split stay non-public on X's metrics reference, and only for posts from the last 30 days.

Is X analytics the same thing as Twitter analytics?

X and Twitter are the same network after the rebrand, not two products to rank. Native analytics and this median are still different measurements. The median is 0.57% on 363 original posts pulled 6 November 2026, from (likes + comments + shares) / views. X's metrics page does not publish an official engagement-rate formula.

Rerun X user tweets (GET /v1/twitter/user/tweets, since=2026-10-09, max_pages=2, max_credits=2) and take the type-7 median of (likes + comments + shares) / views on originals from 9 October 2026 that are at least 6 hours old with views above 0. The median engagement rate on TikTok, Instagram, and YouTube uses (likes + comments) / views with no shares, on 1,269 posts from those three platforms, so 0.57% does not sit next to 5.52%, 2.66%, or 2.34%. If a rerun of these handles moves the median, send the new date and the new post count to socialcrawl@ridio.io.