ehteshambuildagents bl-review
The only honest evidence about what works for an account is that account. Every rule in every LinkedIn guide, including the ones in this pack, is a prior. The user's own last thirty posts are the data.
Input
Whichever the user has:
- The analytics export (LinkedIn: Analytics, then Content, then Export). CSV.
- A screenshot per post with impressions, reactions, comments, reposts.
- Or just the posts and their reaction counts, which is enough for a first pass.
Also read ~/.claude/bylinekit/log.md if it exists, since it records which
opener and theme each post used. Without it you are guessing at the cause.
Rank by rate, never by impressions
Impressions measure how many people the feed showed it to, which is mostly a function of when it was posted and who saw it first. It is the least informative number on the page and the one everyone leads with.
Rank by:
- Engagement rate = (reactions + comments + reposts) / impressions. This is the number that tells you whether the post was good.
- Comment ratio = comments / reactions. High means it started something. This predicts reach on the next post better than anything else.
- Reach multiple = impressions / the user's median impressions. Corrects for an account that is growing.
Show the table sorted by engagement rate, with the raw impressions in a column so the user can see the two orderings disagree. They almost always do, and that disagreement is the most useful thing in the audit.
Then find the pattern, carefully
Group the top five and the bottom five and look for what separates them:
- which opener pattern from
openers.json - length
- first-person story against advice
- whether it carried a number
- whether it asked something real
- day and time, but treat this as the weakest signal, not the strongest
State how confident you are. Thirty posts is a small sample and most of the apparent differences are noise. Say "three of your top five opened with a cost or a number, which is suggestive but it is five posts" rather than "posts with numbers get 40% more engagement". Overclaiming from thin data is how people end up writing worse posts with more conviction.
Say what to stop
The useful half of an audit is subtraction. Name:
- the theme that reliably underperforms for this account
- the opener pattern that is not landing for this audience, whatever the guides say
- the habit visible across the bottom five
If a rule from this pack is not working for this account, say that plainly. The account's own evidence beats the pack's priors, every time.
Write it down
Append the findings to ~/.claude/bylinekit/log.md so /bl-week can use them
next Monday. An audit that is not written down gets re-derived from scratch in
six weeks.
Deliver
- The ranked table, engagement rate first.
- Three things that are working, with the confidence level on each.
- Two things to stop.
- One experiment for next week, phrased so the result will be readable.