Conversion Copy Editor
A line-level copy audit that quotes your text and hands back the replacement.
Audits existing copy line by line against its goal and audience, quoting each problem line and supplying a written replacement ranked by expected impact, with unsupported claims collected as a separate risk list.
Ready-to-use prompt
The prompt
Copy it as-is, then swap the bracketed placeholders for your own details before running it.
Role: You are a conversion copy editor auditing existing copy.
Context:
- The full current copy: {{copy}}
- Page or asset goal and primary CTA: {{goal}}
- Audience and awareness stage: {{audience}}
- Known performance data if any: {{performance}}
- Proof available but not currently used: {{unused_proof}}
- Constraints, brand or legal: {{constraints}}
Task: Return a findings table. Every row must contain:
1. The exact current line, quoted
2. Problem type: CLARITY, WEAK CLAIM, MISSING PROOF, FRICTION, REDUNDANCY or UNSUPPORTED
3. Why it costs conversion for this audience
4. The replacement line, written in full
5. Expected impact: high, medium or low
Then return:
- The top five fixes by impact, as a change list
- Every UNSUPPORTED claim, collected separately as a risk list
- Sections that should be cut entirely, with the reason
- Unused proof, matched to the exact line it should support
- A tightened version of the highest-impact section, applying only the top fixes
Rules:
- No finding without a quoted line and a written replacement, never advise in the abstract.
- Do not rewrite the whole asset, edit it.
- Preserve every constrained element and list what you preserved.
- Where the copy is already working, say so and leave it.
- Treat impact labels as hypotheses unless supplied performance evidence supports them.
- Flag replacements that depend on unconfirmed product facts; do not present them as publishable.
- Identify section and occurrence when a quoted line repeats.
Finish with: the one change to ship first, and how to tell within two weeks whether it worked.Estimated results
Editor's note
Why this prompt matters
Ask a model to improve your copy and you get advice: tighten the intro, add social proof, make the CTA clearer. All true, none actionable, and none of it tells you which line to change. This prompt bans abstraction. Every finding must quote your existing text and supply the replacement in full, which makes the output something you can apply in an afternoon rather than a document you agree with and never use. It also collects unsupported claims separately, because those are a liability issue before they are a conversion issue.
Anatomy
Prompt engineering breakdown
Role
Context
Goal
Constraints
Output format
What you'll get
Expected output
Example: a scheduling tool’s demo page
This fictional audit concerns a scheduling tool for independent consultants who currently arrange appointments by email. The goal is a demo view, not registration. Supplied product facts confirm calendar connection and client-selected booking slots. Available proof includes an approved booking-flow recording. Constraints require preserving “View demo” and “Availability varies by calendar provider.” No performance data was supplied.
Ranked findings
Impact labels are editorial hypotheses, not measured lift estimates. Removing the guarantee takes priority because it creates risk regardless of performance.
| Current line | Problem type | Why it costs conversion | Exact replacement | Impact | |---|---|---|---|---| | “Guaranteed to double your bookings.” | UNSUPPORTED | Promises an outcome the supplied evidence cannot establish. | “Let clients choose from your available appointment times.” | High | | “Unlock your scheduling potential.” | CLARITY | Consultants cannot tell what the tool actually does. | “Let clients book a time without arranging it over email.” | High | | “Start your transformation.” | FRICTION | Makes a demo sound like a larger commitment. | “View demo” | High | | “Booking is simple.” | MISSING PROOF | Asks visitors to trust an adjective instead of showing the workflow. | “Watch a client choose an available time in the booking demo.” | Medium | | “A powerful solution for professionals.” | WEAK CLAIM | Gives this audience no concrete reason to investigate. | “Connect your calendar and share your available booking times.” | Medium |
Top five fixes by impact
- Remove the unsupported bookings guarantee immediately.
- Replace the vague headline with the email-coordination benefit.
- Align the secondary CTA with the actual demo destination.
- Place the approved recording beside the booking-flow claim.
- Replace the generic descriptor with the confirmed setup steps.
UNSUPPORTED risk list
“Guaranteed to double your bookings.” No supplied evidence supports this guarantee. The recording demonstrates functionality, not booking growth; it cannot substantiate the original claim.
Sections to cut entirely
None from this excerpt. These lines need replacement, not wholesale section removal. Do not infer that unseen sections are redundant.
Unused proof placement
Place the approved recording directly beneath “Watch a client choose an available time in the booking demo.” Confirm it depicts the current interface before publication.
Tightened highest-impact section
Let clients book a time without arranging it over email.
Let clients choose from your available appointment times.
View demo
Availability varies by calendar provider.
Preserved: the required CTA and provider limitation. This revision applies only the high-impact fixes.
Ship first and evaluate
Remove the guarantee first. Over two weeks, track demo starts per eligible page visit against a comparable baseline, checking traffic-source changes. Low volume means an inconclusive conversion result—not permission to restore an unsupported promise.
Under the hood
Why this prompt works
Quote-and-replace is the constraint that makes an audit usable. It also acts as a quality filter: a model that cannot point at a specific line usually does not have a specific problem in mind, and the finding disappears rather than becoming a vague recommendation you feel obliged to act on. Limiting the change list to five keeps attribution possible — shipping thirty edits at once guarantees you learn nothing about which one moved the number, which is how pages get rewritten repeatedly without improving.
Model fit
Best AI models for this prompt
ChatGPT
Best default. Fast across long pages and holds the findings table without drifting into prose.
Claude
Strongest on the unsupported-claim list and the most willing to recommend cuts.
Gemini
Useful when you paste a higher-converting competitor page beside your own.
When to use
- Before redesigning a trafficked page, isolate copy hypotheses without claiming layout is innocent.
- When inheriting copy with unclear evidence ownership.
- Before publishing a revised offer page.
- After funnel data identifies a specific CTA drop-off.
When not to use
- For low-traffic pages when you need measured lift, not editorial diagnosis.
- When pricing or offer fit requires customer research.
- When the asset needs structural rebuilding.
- For legal approval or claim verification; use qualified reviewers and original evidence.
Get more from it
Pro tips
- 1
Reject findings without an exact source quote and paste-ready replacement; request only corrected rows.
- 2
Ship up to three related edits, not thirty; log placements and release dates.
- 3
Send UNSUPPORTED claims to the evidence owner or legal reviewer before reuse.
- 4
Supply unused proof with approval status and the specific claim it supports.
- 5
Identify locked CTA labels, disclaimers and navigation text in {{constraints}}.
- 6
Include surrounding headings so local edits do not duplicate adjacent benefits.
- 7
For a second pass, supply accepted edits and rejected assumptions; request remaining conflicts, not another wholesale audit.
Don't ship this
Common mistakes
✗ Accepting abstract advice like improve clarity.
Fix — Enforce quote-and-replace on every row.
✗ Rewriting the whole page at once.
Fix — Apply the top fixes only so results stay attributable.
✗ Editing copy when price or offer is the real problem.
Fix — Check the non-copy factors before another rewrite.
People also ask
Frequently asked questions
Q.How is this different from asking a model to improve my copy?
Every finding must quote your existing line and supply the replacement in full. That single rule removes the vague advice and leaves you with edits you can apply directly rather than a document you agree with.
Q.How many changes should I ship at once?
Three to five. Shipping thirty edits means you cannot attribute any change in performance, and on moderate traffic you will never separate the effects afterwards.
Q.Why is there a separate unsupported-claim list?
Because those lines are a liability issue before they are a conversion issue. Claims you cannot evidence create disputes, refunds and in some categories regulatory attention, so they get reviewed first.