How to Compare Tweet Performance on X

To compare posts fairly, look at reach and response separately, use the median to describe a typical post, and plot individual posts so you can see the spread. Impressions tell you how many times a post was shown. Engagement rate tells you how well it did with the people who saw it. A chart that blends the two, or summarizes a month with an average that one viral post has inflated, will point you at the wrong lesson.

The data for all of this is the post-level export from the account's own analytics, or the same fields from the official X API.

Impressions vs engagement rate

Impressions measure reach. They depend on your follower count, the time of day, whether the post got reposted by a bigger account, and how the feed algorithm treated it. Engagements count the interactions: likes, replies, reposts, link and profile clicks, and other actions X includes in its total. Engagement rate is engagements divided by impressions.

Rates need a floor. A post with 80 impressions and 6 engagements has a 7.5% rate, which beats most of your posts and means almost nothing. Set a minimum (say 500 impressions, or a share of your typical post) and leave anything below out of rate comparisons.

A scatter plot with impressions on the x-axis (often on a log scale) and engagement rate on the y-axis shows both at once. Posts in the upper right reached many people and kept them interested. The upper left is where the small, enthusiastic replies live.

Dot plots instead of bars

For a list of posts, or a comparison between post types, dot plots usually beat bars. A dot marks a position, so the axis doesn't need to start at zero, and you can zoom in on the range where engagement rates actually sit (1% to 4%, say). Sort the rows by value, not by date, unless time is the point.

When comparing groups, show the individual posts as dots and mark the median with a line. Readers get the typical value, the spread, and the sample size in one view. A bar of averages hides all three.

Slope charts for before and after

When the question is "what changed between September and October?" a slope chart is hard to beat. Two vertical axes, one per period, and a line per post type connecting its value in each.

Fig. 1 Median engagement rate per post type in two months. Threads rose past images; the rest barely moved.

Highlight the one or two lines that tell the story and mute the rest. Label the lines directly at both ends, with the value. If more than about seven lines cross in the middle, split the chart or show only the ones that changed meaningfully.

The outlier problem

Post performance is skewed. Most posts land in a narrow band and a handful do much better. Every so often, one post goes viral and does 20 or 30 times your usual reach.

Fig. 2 Twenty-nine ordinary posts and one viral post. The mean more than doubles; the median stays where a typical post sits.

That single post wrecks averages. If October had one viral post and September didn't, "average impressions per post" will report a huge jump, even if every other post did worse. Anyone reading that report will try to repeat October's strategy, when October's real story was one lucky post.

Two habits fix most of this:

Use the median for "typical post" numbers. It is the middle value and barely moves when one post explodes.
Show the outlier separately. Name it, annotate it, give it its own line in the report. It is often the most interesting post of the month, and it deserves analysis on its own terms.

Don't delete outliers quietly. Readers should know the viral post exists and that it was set aside from the typical-post comparison.

Rule of thumb: if one post accounts for more than a quarter of the month's impressions, report that month with the median and the outlier as two separate numbers.

Mean, median and totals

There is a place for each:

Number	Answers	Watch out for
Total impressions	How much reach did we get this month?	Depends on how often you posted
Mean per post	Rarely the right question	One viral post inflates it
Median per post	What does a typical post do?	Ignores the upside entirely
Share of posts above a target	How consistent are we?	Pick the target before you look

A histogram of impressions per post, often on a log scale, shows the shape of the distribution and makes the case for the median better than any explanation.

Small multiples by post type

Text posts, images, video, threads and link posts behave differently. Links tend to get lower engagement rates because people leave to read the page; threads often get more replies. Mixing them in one chart averages away those differences.

Fig. 3 One row per post type, one dot per post, a tick at the median. Threads lead, but only nine posts sit behind that row.

Small multiples give each type its own row or panel, on a shared axis. Keep the axis identical across panels, or comparisons break. Print the number of posts in each panel; a type with nine posts doesn't support the same confidence as one with forty.

Getting the post data

X's analytics has a content section listing your posts with impressions, engagements and the engagement breakdown, and the desktop view can export it to CSV. Access to the full dashboard and its export has generally required a paid Premium subscription, and column names and date ranges change from time to time, so check the export you actually have before building a template around it. For several accounts, or history beyond the export window, use the official X API under its current terms.

Avoid scraping public post counts as a substitute. It breaks X's rules, and scraped data lacks impressions, which most of the comparisons above depend on. The engagement heatmap guide uses the same export to look at posting times.

Questions

Is engagement rate better than impressions for comparing posts?

They measure different things. Impressions show reach, and engagement rate shows how well a post did for the people who saw it. Report both, and be wary of rates on posts with very few impressions.

Why use the median instead of the average?

Post performance is skewed, with most posts close together and a few far above. One viral post can double the average, while the median stays near what a typical post does.

What chart works best for comparing post types?

A dot plot or strip plot of individual posts per type, with a mark at the median. It shows the typical value, the spread and how many posts sit behind each type in one view.