Seven Copy Critics
An adversarial pre-publication review that runs marketing copy through seven hostile reader personas, plus an eighth critic that checks whether a person's voice survived, each attacking the draft at three failure layers: 24 checks. The author of a piece of copy is the worst person to review it: they know the context, they are already bought in, and they cannot see what a cold reader sees. This skill simulates seven cold readers having seven bad days, and one who has read too many ghost-written posts.
Paths below are relative to this skill's directory; resolve them to absolute paths before running anything.
Configuration
This repository's agent configuration should have been provided to you (the
## Agent skills block pointing at docs/agents/). If it is missing, use the
defaults below and suggest the user run the setup-aclarify-skills skill.
- Conventions folder (
docs/agents/workspace.md): house rules. Default: none. If one is configured and holds a voice, tone, brand-voice or writing-standards document, read it before the rebuild and obey it; where it is stricter or more specific than this skill, or contradicts one of the review constraints, it wins. With none, the rebuild follows the review constraints, which are part of the method: each states the reader failure it prevents. The skill adds no voice, tone or brand of its own.
When to use
- Before sending any cold outbound email or sequence
- Before publishing a landing page, ad, or newsletter
- Before scheduling any organic social post that matters (especially a personal post under someone's name)
- Before publishing a blog post
- When copy "feels right" but is not converting
- After
humanizehas cleared the AI tells from a draft, on the rewritten copy. A draft that still reads templated goes there first: that skill owns the tells, this one owns whether the copy converts
When not to use
- To draft the piece: a piece carrying an argument is built by
draft-argued-piece, and conversion copy bydraft-conversion-copy, then reviewed here - To inventory AI tells:
humanizeowns the tell taxonomy and measures it, and runs before this skill - On copy that is already converting: do not fix what is not broken
- For pre-launch strategic review of an entire campaign brief: use
campaign-briefinstead - For visual or design review of marketing assets: use
brand-reviewinstead - For one-off internal messages with no audience consequence
Where a skill named here is not installed, say so and carry on without it.
Procedure
Step 1: gather context
Before running the critics, collect:
- The draft copy (full text, exactly as it would be sent)
- The target audience (specific persona, role, situation, not "marketers" or "operators")
- The channel and format (cold email, social post, landing page hero, blog intro, etc.)
- The desired action (reply, click, book a call, share, subscribe)
If any of these are missing, ask the user for them before proceeding, using your structured-question tool if you have one; otherwise list the missing points numbered in plain text and stop until they reply. A review without audience context produces generic feedback.
Step 2: run all eight critics across all three failure layers
The full matrix is in references/critic-matrix.md. Load it now and work through it: eight critics, each applied at the hook, the body, and the close.
Run every check. A critic that finds nothing is a finding in itself, and skipping one because the copy "obviously passes" is how the failure this skill exists to catch survives review.
Step 3: compile the failure report
Categorize every failure surfaced in Step 2 by severity. Use this exact format:
## FAILURE REPORT
### CRITICAL: [n] failures
(blocks the desired action; copy cannot ship as-is)
- [Critic #] [Layer]: [one-line description]
- ...
### HIGH PRIORITY: [n] failures
(meaningfully weakens the copy; should be fixed before publishing)
- ...
### LOW: [n] failures
(minor polish; address if cheap, otherwise note and move on)
- ...
Severity rules:
- CRITICAL: Any First Impression failure on Critic 1 (Time-Crushed) or Critic 4 (AI-Allergic). Any Action layer failure on the dominant target persona. Any Trust failure on Critic 3 (Burned Skeptic) when the copy makes a quantitative claim.
- HIGH PRIORITY: Any failure on a layer that directly opposes the desired action (e.g., Visual Reader failures on a story-driven post).
- LOW: Cosmetic patterns, minor word choice, second-order trust dings.
Step 4: rebuild
Rewrite the draft addressing every CRITICAL and HIGH PRIORITY failure. The rebuild standard is tighter, not bigger. Sharper, not different.
If the rebuild reads like a different person wrote it, the review failed. The rebuild must:
- Preserve the author's voice and core message
- Fix root causes, not just symptoms
- Not introduce new failures across other critics (re-check mentally)
- Be the same length or shorter than the original
- Comply with the review constraints below and with the conventions folder's house rules, where one is configured; where the two conflict, the folder wins
- Invent nothing: no number, name, result or quotation the draft or the user did not supply. Where a fix needs a specific the draft lacks, mark the gap and ask the user for it
Output the full rebuilt copy in a fenced block, ready to copy-paste.
Step 5: validation pass
Before declaring done, mentally walk all eight critics through the rebuilt version one more time. For each critic, write a single line:
## VALIDATION PASS
1. Time-Crushed: PASS / FAIL: [one-line reason]
2. Self-Conscious: PASS / FAIL: [one-line reason]
3. Burned Skeptic: PASS / FAIL: [one-line reason]
4. AI-Allergic: PASS / FAIL: [one-line reason]
5. Lurker: PASS / FAIL: [one-line reason]
6. Visual Reader: PASS / FAIL: [one-line reason]
7. Inspiration Seeker: PASS / FAIL: [one-line reason]
8. Operator-Voice Detector: PASS / FAIL: [one-line reason]
If any critic still fails, return to Step 4 and rebuild again. Do not ship copy with any remaining CRITICAL failure. A maximum of 2 rebuild rounds: if the third pass still fails, the underlying message itself is wrong and the copy needs to be rethought from the audience or offer angle, not rewritten.
Step 6: changelog
End the review with a "What Changed" section explaining each meaningful edit and which critic's failure it addressed:
## WHAT CHANGED
- [Original phrase] → [new phrase]: fixes Critic [#] [Layer]
- ...
This is non-negotiable. The author needs to see the fixes mapped to root causes so the same patterns do not reappear in the next draft.
Output template
In references/output-template.md. Load it before writing the report.
Review constraints
Five constraints the rebuild is checked against, each because breaking it is a failure the critics above keep finding in cold readers. They are review method, not house style. Where the conventions folder (see Configuration) states a rule that contradicts one, the folder's rule wins: drop that constraint for this review and say so in the report.
- No recapping the recipient's own business in follow-up emails. Reference shared ideas or next steps instead. They know their business better than the sender, so a recap spends their attention on nothing new.
- No parroting call notes. Focus on what is next, not what was said. Repeating the call proves only that someone listened; the reader needs the next step.
- Dream outcome leads, mechanism explains. Lead with what the customer gets (more bookings, less churn), not how the product works. A cold reader decides whether to keep reading on the outcome; the mechanism matters only once they want it.
- Literal-first imagery. If the copy references visuals, they must depict specific moments from the post content, not abstract metaphors. An abstract image gives the Visual Reader nothing to recognise, so it stops no one.
- Author conviction required. If the draft goes out under a person's own name, the rebuilt copy must contain at least one sentence expressing a belief, opinion, or conviction that person holds. Take it from the draft or ask them for it; never supply one. A bylined post that takes no position reads as generated, which is exactly what the Operator-Voice Detector catches.
A rebuild that breaks a constraint still in force fails the validation pass, however the critics score it.
Key principle
If the rebuilt version reads like a different person wrote it, the review failed. Tighter, not bigger. Sharper, not different.
The goal is not a "better" version of the copy in the abstract. The goal is the same author's same message, stripped of the failure modes a cold reader would catch. Voice preservation is a hard constraint, not a nice-to-have.