Coach PMM work through iterative review that retains feedback

01 Set the frame before you read a word of the draft.

Open every review by writing down four things: who the PMM is, what document type this is, which campaign and iteration number, and the prior review's Focus 3. A campaign is one project across many drafts. An iteration is one comparison of version N to version N+1. The frame is not bureaucracy. It is the filter that decides what you even notice in the draft.

This matters because the same artifact reads differently depending on the question you bring to it. The frame you give AI becomes the filter everything gets read through. A launch brief read as "is the narrative right" surfaces different gaps than the same brief read as "is the copy tight." Name the decision this review has to inform before you start. Otherwise you summarize the draft instead of judging it, and a summary is not a coaching answer. Make the implicit explicit

If you use an AI to help score, the frame is doubly load-bearing. Run it on one document with a clear job, not a batch of drafts with a generic "review this" prompt. Batch prompts average across inputs and start filling gaps with plausible-sounding feedback that traces to nothing. Batch-processing interview transcripts with Claude produces fabrications; reliable output requires a dedicated per-transcript skill at three minutes per file

02 Decide which dimensions apply to this document type, then score only those.

The rubric has 17 core dimensions scored on every document and 12 contextual dimensions scored only when the document type calls for them. Scoring a contextual dimension that does not apply is as much an error as skipping one that does.

Typical totals by document type:

The split between strategy dimensions and craft dimensions is the one that drives the coaching. A PMM has to do both: turn messy product capability into a clear narrative, and write the launch notes and battlecards that sales actually uses. PMM must balance strategy and execution, not choose one. The score has to read both halves separately so you know which one to coach.

03 Score both versions. The improvement delta matters as much as the absolute number.

Score version N and version N+1 on the applicable dimensions. The overall score is the sum of scored final dimension scores, divided by the number of dimensions scored, times 10, on a 0 to 100 scale. The number itself is less useful than the movement between drafts.

Hold the scale honestly. A 5 is average, not bad. Most competent PMM work lands 5 to 7. Reserve 8-plus for strong work and 9 to 10 for exceptional. Grade the document, not the person: the same rubric applies regardless of seniority. The most common scoring failure is inflation, where most work scores 8-plus and the signal disappears. Decision quality at scale comes from process, not willpower

Every score cites specific text from the draft. A score without cited evidence is an opinion, and the PMM cannot act on an opinion. This is the same discipline an eval engineer applies to model output: you read the actual trace and name the actual failure, you do not grade against your assumption of what could go wrong. Evals are systematic data analysis on your LLM application, start with error analysis, not tests Evals are data analysis, single judge, binary rubrics, error analysis first

"It really doesn't have to be scary or unapproachable. It really is, at its core, data analytics on your LLM application."

· Hamel Husain & Shreya Shankar, Lenny's Podcast, 2026-04-28 · “Evals are systematic data…”

04 Run the feedback-retention check on every iteration past the first.

For each item in the prior review's Focus 3, mark whether the next draft addressed it: Yes, Partial, or No. Calculate the retention rate. The target is above 60% across five-plus iteration-2-and-later reviews. This is the primary metric of the whole framework, because feedback that is not applied is feedback that was not useful.

When retention drops below 60%, investigate before you adjust the feedback itself. Ask three questions: Is the feedback unclear? Is the framing wrong? Is the PMM overloaded? The fix is usually to the coaching approach, not the content of the note. Coaching that gets ignored is coaching that failed to land, and changing the words without changing the delivery repeats the failure.

This is the verification loop that makes the system learn instead of just logging reviews. A trace of what you said tells you what happened, not whether it worked. Retention is the feedback signal bound to the trace. Verification, not execution, is the irreplaceable human job

05 Select Focus 3 using effort against impact, and bias the choice to the PMM's level.

Three actions per review. Not ten. Pick them on an effort-by-impact basis: prioritize high-impact, low-effort fixes first, and include at most one high-effort item per review. Ten recommendations is overload, and overload is the fastest way to kill retention. Three is the limit because three is what a person can hold and apply in one revision.

Selection rules that override raw scores:

• Do not prioritize craft over strategy if strategy scores below 5. A perfectly worded sentence built on the wrong frame is still wrong.
• Align at least one action with the PMM's active improvement tracks, if any exist.
• Check whether the prior Focus 3 was addressed before repeating an item.
• Consider pattern history across campaigns, not just this one document.
• Bias toward strategy for new PMMs, craft for experienced ones.

The level bias works because what "good" means changes by seniority. A skill framework that defines the expected skillset at junior, mid, and senior levels lets each PMM self-assess and lets the coach target growth where the level demands it. Build custom skill frameworks to develop PMMs at each level. Coach the dimension that moves that PMM up, not the dimension that is easiest to score.

06 Write each Focus-3 item with cited evidence and a rewrite example.

A Focus-3 item without a rewrite example tells the PMM what is wrong but not what good looks like, and "fix this" is not coaching. Use the three-line format on every item:

Tag each item with an effort level so the PMM can plan the revision: Quick Win (15 to 30 minutes), Medium Lift, or Deeper Work (1 to 2 hours).

When the fix is a craft fix, name the craft principle so the lesson transfers to the next draft. Conciseness is one the rewrite can demonstrate directly: every word that does not earn its place signals that the writer values their own message over the reader's time. Conciseness is respect, every unnecessary word signals that you value your message more than the reader's time

"Conciseness is respect. Every unnecessary word signals that you value your message more than the reader's time."

· Eddie Shleyner, VeryGoodCopy, 2024-02-01 · “Conciseness is respect, every…”

For copy and content dimensions, the rewrite examples cluster around a few reinforcing fundamentals: reader-centredness, emotion as the persuasion mechanism, voice, clarity in the headline, and subtraction. Copywriting craft is built on five reinforcing fundamentals Coach against those, not against your personal taste.

07 Separate strategy coaching from craft coaching, and never swap them.

The single most common coaching error is fixing the wrong layer: polishing craft when the strategy is broken, or rewriting the strategy when the craft is fine. The rubric split exists to prevent exactly this. Read the strategy score first. If the frame, the audience, or the differentiator is wrong, no amount of word-level editing saves the document.

For copy-level fixes, a few operator tests do reliable diagnostic work and travel well as rewrite rationale:

• The so-what test catches inside-out claims. Most content answers the company's question instead of the reader's, and that is the most-skipped move in content creation. The "So what?" step is the most-skipped move in content creation across B2B and B2C

"Most marketing content answers the company's question ('What do we want to say?') rather than the reader's question ('Why should I care?'). She treats this as the single most skipped step in content creation across B2B and B2C organizations."

· Ann Handley, Ann Handley — Writing GPS and the 'So What?' step, 2026-03-03 · “The "So what?" step…”

• The face test catches manipulation. If the PMM would not say the line to the buyer in person without flinching, it has crossed from candor into manipulation. Would you say it to the reader's face without flinching?, the test that separates copy from manipulation

"The test: would you say this to the reader's face, in person, without flinching? If not, it is manipulation."

· Cole Schafer, Sticky Notes — The Honesty Test, 2024-01-15 · “Would you say it…”

• The headline is where most of the persuasive weight sits, because most readers never get past it. If the draft buried its strongest line in paragraph three, that is a Focus-3 item on its own. Spend 20% of total writing time on the headline alone, it carries 80% of the persuasive weight

When you are coaching voice and resonance, the rewrite should mirror the buyer's own words rather than smooth them into a clean theme. The exact phrase is the hook; the paraphrase is just a description, and descriptions do not convert. AI synthesis smooths over verbatim buyer phrases: mine Reddit and review sites directly, mirror the words, do not paraphrase

"Resonance comes from mirroring, not paraphrase."

· Nicolas Cole, Build and Launch a VOC Landing Page, 2026-05-07 · “AI synthesis smooths over…”

08 Add a "Parked for Later" section so the unaddressed dimensions are visible, not dropped.

List 2 to 3 dimensions you saw but did not prioritize this round. Parking is not the same as ignoring. The PMM sees the full picture, knows what is coming next, and trusts that you read the whole draft. It also protects retention: if you raised ten issues and only tracked three, the other seven leak away. Parked items become Focus-3 candidates in a later iteration once the top three are landing.

09 Maintain pattern history across campaigns and flag persistent gaps.

Track each PMM's recurring issues across campaigns, not just within one document. When the same dimension scores low across three iterations, that is a persistent gap, and a persistent gap is a different conversation than a one-off miss. Flag it explicitly. The pattern history is what lets you tell the difference between a PMM who made a single slip and one who has a structural blind spot the coaching has not reached yet.

10 Decide where AI helps and where a human has to verify.

AI can draft the rewrite examples, summarize the draft, and pre-score dimensions. It cannot decide which three fixes matter most for this PMM at this level on this document. That is the judgment layer, and it does not compress. Building costs collapsed; judgement didn't, the squeeze is on positioning, not production

"Judgment is the part that doesn't compress."

· Kevin Indig, Growth Memo, 2026-05-04 · “Building costs collapsed; judgement…”

Wherever AI touches the review, add a human checkpoint at each stage. Models fill gaps with plausible-sounding output when the context is thin, so a generated score or a generated rewrite has to be verified against the actual draft text before it reaches the PMM. Running AI on multiple inputs simultaneously without structured validation checkpoints produces fabricated output, not analysis errors

"AI needs A LOT of help to do a good job (way more than the average person realizes)."

· Anthony Pierri, AI needs A LOT of help to do a good job, 2026-05-10 · “Running AI on multiple…”

This is also why the coaching role survives the cheap-execution shift. When AI does the launch emails and the slide tweaks, the PMM who only did execution is exposed, and the durable value is the judgment about which angle and which narrative to commit to. AI replaces PMMs whose job was launch emails, recap docs, and slide tweaks The coach is teaching that judgment, one Focus 3 at a time.

"When execution becomes easy, thinking becomes the differentiator. When content is everywhere, clarity becomes the advantage."

· Sachin Jha, LinkedIn, 2026-04-10 · “AI replaces PMMs whose…”