How to analyse your X posts and find what works

Start with a decision your analytics can answer

An analytics session is useful when it changes a decision. “How am I doing?” is too broad to guide a review. “Should I share more detailed progress updates or more finished-product demos?” gives you two identifiable groups, a comparison, and an action you can take next week. Write that question before you open a dashboard.

Choose the outcome that fits your work. A designer looking for relevant collaborators may care about thoughtful replies and portfolio visits. A founder announcing a product may care about qualified visits and signups. A creator testing an explanation may start with reach and then inspect whether readers understood it. These goals overlap, but a post that attracts a large audience does not automatically satisfy all of them.

Keep an outcome ladder: distribution, response, and business result. Impressions belong to distribution. Replies and clicks help describe response. Signups or enquiries belong to the business result and usually require your own website measurement. Do not collapse the ladder into a single score. You need to know where a post helped and where the path stopped.

Build a comparison set you can trust

Start with a recent, reasonably coherent period. If you changed your product, audience, or posting style, record that boundary instead of blending everything into one lifetime average. A spreadsheet is enough for the first review. Include post text or a link, publication time, collection time, format, topic, intent, and the metrics you actually have.

Post age matters. A post collected two hours after publication has had a different opportunity to accumulate responses than one collected after several days. Pick a repeatable observation window for future reviews. For existing history, separate very recent posts or clearly mark the unequal ages. Missing metrics should stay missing; turning an unavailable click count into zero makes the data look more complete than it is.

Your retained history may also be incomplete. An API, export, or analytics product can contain only part of your account. Count how many posts are present and which dates they cover. Keep replies, standalone posts, and repost-related activity distinguishable where possible. The point is not a perfect dataset before you begin. It is a comparison whose limitations you can explain.

• Record the date range and the number of retained posts.
• Check when metrics were collected and how old each post was.
• Mark missing fields and unusually large distribution events.
• Separate post types that serve substantially different purposes.

Calculate your typical result before ranking winners

A median describes the middle of a sorted set. It is useful when one exceptional post would pull an average far above the experience of most posts. Calculate both if you want to understand the shape of your results, but use a clearly named baseline when you compare groups. “Twice my usual reach” should refer to an actual calculation.

Consider this invented example: seven posts received 180, 210, 220, 240, 260, 280, and 4,000 impressions. Their median is 240; their average is 770. Comparing your next ordinary post with 770 would make most normal results look disappointing. The viral post is valuable, but it answers a different question: what happened in that particular case?

A baseline ratio is simple: group median divided by comparison median. If progress updates have a median of 480 and comparable other posts have a median of 240, the ratio is 2.0. Show the two medians and group sizes alongside the ratio. A ratio without its denominator or sample count is an impressive-looking number with too little context.

Label topics, formats, and intentions separately

Topic describes what the post is about: onboarding, pricing, performance, illustration, or a customer problem. Format describes how it appears: text, image, video, or thread. Intent describes what it tries to do: explain a decision, share progress, announce availability, ask for feedback, or teach a method. Separating these dimensions stops “screenshots work” from becoming your entire strategy.

For example, a screenshot of a loading-speed improvement and a screenshot of a new logo share a format but may interest different people for different reasons. A product demo can combine video with a launch announcement or with an explanation of a workflow. Inspect the combination before crediting the medium alone.

Use a small vocabulary that you can apply consistently. If every post gets a unique topic, you cannot compare topics. If every post gets a dozen labels, almost every group overlaps. Label the main topic and intent first, allow a second label when it changes interpretation, and write a one-line definition for ambiguous labels. You can revise the taxonomy when real examples reveal its weaknesses.

Read the evidence behind the pattern

Once a group looks promising, open its posts. Ask what readers actually saw. Were the stronger updates more specific? Did they show a visible before and after? Did a larger account share one of them? Were they all published during a launch? Metrics narrow your attention; reading explains which creative choices deserve another attempt.

Look at unsuccessful examples in the same group. They are often more informative than the winner. If a detailed screenshot performed well but three vague progress notes did not, “post more progress” is an incomplete recommendation. A better hypothesis is that showing a concrete improvement with a clear user benefit may be worth repeating.

Follower movement can add context, but a rise near publication is not proof that one post caused it. Other posts, replies, external mentions, and delayed visits may contribute. Describe the relationship accurately: follower growth coincided with this period. Claiming precise attribution requires evidence that a simple timeline does not provide.

Turn the review into a small content experiment

Choose one change that the evidence supports. Keep the next test close enough to past examples that you can learn from it, but avoid duplicating the same post. If concrete progress updates look promising, prepare several updates about different real improvements, each showing the change and explaining its consequence.

Write the expectation before publishing: “These updates may beat the median reach of comparable recent posts.” Define what you will observe and when. Also decide what would make you hesitate, such as strong reach with no relevant replies or visits. This keeps a flattering result from moving the goalposts after the fact.

At the next review, compare the new examples with the stated baseline and read them again. A small series is directional evidence, not a causal study. If the result is mixed, refine the question. If the posts help the right readers even with modest reach, that may be a useful outcome for your goal.

Review this post dataset for the decision: should I share more concrete progress updates?
State date coverage, missing metrics, post ages, and the baseline. Compare medians and sample sizes. Cite the post IDs behind each observation. Separate observed associations from explanations you cannot verify. Finish with one experiment and what would weaken the hypothesis.

Make the review a repeatable habit

Save the question, the group definitions, the baseline, and the next test in the same place. The following review should begin with the previous decision rather than with a fresh hunt for the biggest number. A short record helps you learn whether you are developing a repeatable practice or just reacting to an exceptional week.

X-tra is built around comparing your posting patterns with your own history. Its current public offer is early access. You can use the method in this guide today with a spreadsheet; the reason to use a dedicated analytics workspace is to reduce the repeated work of finding and comparing evidence, while keeping the interpretation in your hands.