Review

Review

The goal isn't a report, it's a decision. Every review ends with exactly one change that gets done differently next week.

Approach

1. Get the numbers: via a connected analytics tool (Buffer, Metricool, or similar), from an export, or from a screenshot. If neither is available, ask for the metrics of the last five to ten videos.
2. Connect each number to its script – pillar, format, hook pattern, length, publish time.
3. Look for patterns, don't judge individual cases.
4. Formulate a hypothesis and derive the change.
5. After confirmation, write it to LEARNINGS.md.

Which Numbers Count

Instagram itself names three signals as decisive: watch time, sends per reach (forwards via direct message), and likes per reach. Two of these are ratios – so always set them in relation to reach, never compare absolute numbers.

Order of evaluation:

1. Retention / watch time – says the most about the video's quality.
2. Forwards per reach – the strongest distribution lever: what gets shared via direct message reaches people outside your own bubble.
3. Saves – utility value. A high number means: someone wants to find this again.
4. Comments – not the count, the content: what's being asked? That's the topic backlog for the coming weeks.
5. New followers per post – measures whether the post convinced, not just entertained.
6. Reach and views – result, not cause. Look at these last.

Never rank by views. Reach and impact can diverge widely; a post with little reach and a high save or follower rate is the better one. So always calculate the rate, not just the absolute number, and show it alongside.

Likes and total follower count are secondary. Don't optimize for them, unless there's a striking break.

Honesty About Significance

• Under five videos per group, no pattern is discernible. Then label it explicitly as an observation, not a finding.
• An upward outlier isn't proof. Check whether it traces back to something one-off – a shared post, an occasion, a trending sound.
• Two things that changed at the same time don't let you draw a conclusion about either one. Say that, instead of settling on one explanation.
• If the data doesn't yield anything, "no reliable pattern this week" is the correct result. Don't invent a finding.

Output Format

# REVIEW · <Period> · <n> Videos

## What Stands Out
<two to four observations, each with the numbers alongside>

## Hypothesis
<one paragraph: what could explain this – phrased as a conjecture>

## Change for Next Week
<exactly one, concrete and verifiable>

## How We'll Measure It
<which number would have to change for the hypothesis to hold>

## From the Comments
<topics and questions that could become a video>

Updating the Learnings

Propose an entry for LEARNINGS.md – dated, with the data basis and the degree of confidence:

## 2026-08-30 · observed across 7 videos
Hooks with a concrete number hold retention longer than hooks with a question
(avg. 61% vs. 44% retention). Not yet confirmed – few cases.

Only write it after confirmation. If a new observation contradicts an older entry, don't delete the old one, add to it and date it – a learning that reverses itself is information in its own right.