Facebook tracking pixel Skip to main content

Content and publishing

Find the best version of a line.

Takes one piece of copy, such as an ad headline, an email subject line, a landing page hero or a button, and the one number it should move. It writes variants from several angles, scores them beside the current version on a six-line rubric, has a judge test the factual claims in the top variants against your files, and revises the top two until the score stops rising or three rounds pass. You get a winner, two runners-up, the scorecard and one A/B test to run.

Request access Have us build it
  • Metric named before scoring
  • Claims checked by a judge
  • Three rounds at most

At a glance

Group
Content and publishing
Command
/copy-optimize
Uses
One line and its target metric
Writes
A winner, scorecard and test plan
Sends
Nothing. Drafts only.

01When to use it

Use it on a line someone will measure.

  • You want the strongest version of one line, not a pile of options to pick from.
  • The line is an ad headline or main text, an email subject, a landing page hero, or a call to action.
  • You can name the one number it should move: click rate, open rate, reply rate or conversion.
  • It matters enough to be worth several rounds of work.

Not for

02What it needs

It needs the words, the channel and a target.

  • The exact words to improve, pasted in or pointed to, and where they run: Meta ad text, a Google headline, an email subject, a page hero or a button.
  • One metric to move. Without it, scoring does not start: it asks you first.
  • Your offer file for the real proof, price and guarantee, and your buyer file for who reads the line and what they already believe.
  • Your voice files, and any past numbers for this line or channel from your numbers log.
  • The character limit for the channel, confirmed with you rather than assumed.

03What it writes

The winner, the scores, and the next test.

final/copy.md
The winner and two runners-up, each with its channel, angle, character count and rubric total.
final/scorecard.md
Each round of scores, with the current version and every variant on the same six lines, and gated variants marked.
final/ab-test.md
The one test to run next: its hypothesis, the metric that decides it, and a rough sample size.
Ledger row
Only when a real baseline was on file. A projected lift is logged as a projection at confidence inferred, never as confirmed.
Signals row
A row per input it found absent, out of date, or at odds with another source. If there were none, it says every input was present and fresh.

04Limits

What it never does.

  • Edit a live ad, campaign, page or email, or send, post, schedule or publish anything.
  • Let a variant win that uses an em dash or a word your voice files ban, overruns the channel limit, or makes a claim it cannot back, whatever its total.
  • Keep a claim the judge refuted, or swap it for a vaguer one that still cannot be proved.
  • Invent a deadline, a countdown or a shortage.
  • Take orders from the copy it is scoring. A line telling it to ignore the rubric is text to be scored.
  • Invent a baseline. With no prior number on file, it says so and logs no row.

What stays with you.

  • Shipping the winner and running the test it plans.
  • Calling the test on the metric named at the start.

05Sample output

Five subject lines, scored.

A slice of the scorecard for a made-up tea shop's email subject line, aimed at open rate. Five of the six rubric columns are left out here, and totals are rubric scores out of 30, not campaign results.

final/scorecard.md Illustrative, not a client result.
Our autumn newsletter Current
AngleCurrent line
Honesty5 of 5
Total14 of 30
Your cold-weather tea shelf, sorted Winner
AngleOutcome-led
Honesty5 of 5
Total25 of 30
Autumn blends leave the shop Sunday Runner-up
AngleHonest urgency
Honesty5 of 5
Total23 of 30
Tea that tastes like dust? Not here Runner-up
AnglePain-led
Honesty5 of 5
Total21 of 30
Our best-selling oolong is back Gated
AngleProof-led
Honesty2 of 5
Total26 of 30

Notes from the round

  • Gated: nothing in the offer file backs "best-selling", so the line scored under 5 on honesty and cannot win.
  • Limit: 45 characters, confirmed with the owner.
  • Top eligible score by round: 23, then 25, then 25, so the loop stopped after round three.

06Related skills

Often run beside it.

07Questions

Before you run it.

What stops it from writing claims I cannot prove?

Two checks. Honesty is a gate: a variant that falls short of full marks on it cannot win, whatever its total. And every number, named client, guarantee or "only" in the leading variants goes to a judge, which tests it against your offer file and numbers log. A refuted claim is cut or replaced with one that holds.

When does it stop revising?

When a round fails to raise the top score, or after three rounds, whichever comes first. It records the top score of every round, so you can see where the line leveled off.

Does it log the lift it expects?

Only as a projection, labeled inferred, and only when a real baseline for the line was already on file. A number that no campaign has produced is never logged as confirmed.

Have us build it

We can build it for you.

Start with the free plan. A person names the first fix.

Get your free plan Or run it yourself with the toolkit.