Reads your content numbers honestly: engagement and conversion rates with the denominator shown, fair comparisons between posts and periods, and what the data does and doesn't prove.
Each one is a whole message. Pick one and it is sent with the skill the moment you are in.
3 files
SKILL.md
Calculate and interpret content performance — engagement, view, click and conversion rates per post and for a whole period, comparisons between posts, formats, accounts and months — showing the formula, the denominator and what is excluded. Use whenever the user gives post metrics (interactions, views, reach, clicks, sales) and asks for a rate, a total, a comparison or what the numbers mean; it carries a helper that computes period totals correctly instead of averaging rates.
Numbers answer a question only when the question, the metric and the denominator are explicit.
Agree on what is being evaluated and which metric represents it. Views, reach, impressions, likes, comments, saves, shares, clicks and sales are different things; never swap one for another. Keep account, channel, period and format attached to every number so you only compare like with like.
Use what the user provided, or read the account with the social tools available in this workspace. Record the account, platform, date read, sample size and coverage. If a read is truncated or partial, say so: a partial feed does not prove that other posts don't exist. A missing value is unknown, never zero.
State every formula in words the user can check: engagement rate = interactions ÷ views × 100. For an aggregate, sum the numerators and the denominators, then divide — never average per-post rates. A zero denominator makes a rate not calculable, not 0%.
For plain rates or a period total, the bundled helper does this exactly (references/rates.md). For anything else, compute it in code; do not force every question into an engagement rate.
Compare on one explicit metric at a time. Separate observations ("carousels averaged 4.1% vs 2.3%"), associations, hypotheses and causal claims. Name the sample size and the limits that affect a recommendation. A rise in likes or views is not a rise in sales or unique people unless data shows it. Ten posts is a hint, not a law.
Give results and recommendations proportional to the evidence, keeping units, denominators and rounding consistent (two decimals for percentages unless asked otherwise).
Adapted from Endless · MIT
Add this skill to a free workspace and use it in chat today.