SOCIAL

Post performance post-mortem

Use at the end of a month when you have performance numbers and want to know what actually drove them.

post-mortem-what-worked.md

Download .md

You are analyzing social performance data to find what is repeatable.

Data (one row per post, with metrics):
"""
{{POST_DATA}}
"""
Platform: {{PLATFORM}}
Follower count and typical baseline: {{BASELINE}}

Step 1. Normalize. Rank posts by {{PRIMARY_METRIC}} relative to the baseline, not by raw number.

Step 2. For the top 5 and bottom 5, output a table: Post | Metric vs baseline | Format | Hook type | Topic | Length | Had link? | Your read.

Step 3. State up to four patterns you can support from this data. For each: the pattern, the posts that evidence it, and the posts that contradict it. If a pattern rests on fewer than three posts, label it "weak signal, do not act yet".

Step 4. State plainly what this data cannot tell us - timing effects, audience changes, algorithm shifts, sample size.

Rules:
- Do not attribute performance to a cause the data cannot distinguish. Say "cannot separate" where that is true.
- Do not invent benchmarks or industry averages. If you have no figure, say so.
- No motivational summary. End with three specific things to test next month, each phrased as a change to one variable.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{POST_DATA}}
{{PLATFORM}}
{{BASELINE}}
{{PRIMARY_METRIC}}

Getting a better result

Include your flops - a post-mortem with only the winners in it will invent causes.
The "cannot separate" lines save you from rebuilding a strategy on twelve data points.
Feed last month's three tests back in next time so you build a record rather than restarting.

Questions about this prompt

When should I run this rather than reading the dashboard?

At the end of a period, before you write next month's plan from a feeling. The dashboard tells you which posts did well. This asks what is repeatable, which is a harder question, and it is willing to tell you that twelve posts cannot support the conclusion you were hoping for.

What data do I need?

One row per post with metrics, the platform, a baseline for your follower count, and a chosen {{PRIMARY_METRIC}}. Include the flops, as the tips say, because a table with the losers removed invents causes for things that were never caused. Format, hook type and length per row give the pattern work something to bite on.

What comes back, and which part is the useful bit?

Posts normalised against your baseline, a top and bottom five table, up to four patterns with both supporting and contradicting posts, and three tests for next month. The contradicting column and the weak signal labels are the value, because they are what stops one lucky post becoming a strategy.

What is the mistake?

Acting on a pattern the prompt marked weak signal, do not act yet. Fewer than three posts is noise, and rebuilding a plan on it is how teams chase a format that never worked. Read the cannot separate lines too: timing, audience growth and algorithm changes are not distinguishable inside one month of data.