Pre-Publish Quality Gate: 12 Rules, 0-100 Scorecard (Claude)
Contents
I stopped trusting my "looks good" gut check on drafts six months ago. Now every post goes out only after Claude scores it above 75 out of 100 — and the verdict comes back in about 40 seconds.
Here's the 12-rule scorecard I run on every draft before it ships. Each rule is binary. The math is fixed. 75 is the line I won't cross without a revision pass.
The 12 Rules
- Headline clarity — H1 promises a specific outcome to a specific reader. No "Guide to X."
- Intro hook — First two sentences open with a result, scene, or contrarian claim. Not a definition.
- Source citation — Every stat, date, or named study is linked or attributed inline.
- Claim substantiation — Every non-obvious claim has a number, example, or named source behind it.
- Length fit — Word count lands within ±20% of the target band for the format.
- Internal links — At least 2 contextual in-body links to other posts (not the sidebar).
- External sources — At least 1 authoritative external source per 500 words.
- Heading hierarchy — One H1, scannable H2s, no skipped levels.
- CTA presence — One clear next action at the end (read, try, download).
- Anti-AI tells — No "in today's fast-paced world," no "delve into," no wall-to-wall 3-bullet lists.
- Specificity — At least 3 concrete numbers, tool names, or named examples in the body.
- Intent match — The post actually answers the H1 question, or it drifted.
Each rule scores 0 (fail) or about 8.3 (pass), summing to 100. I round.
The Prompt
I paste the draft and this prompt into a single Claude turn:
You are a strict pre-publish editor. Score the following draft against these 12 rules, one line per rule, with 0 or 8.3. End with the total. If a rule fails, quote the offending sentence. Be honest — a score above 75 means the post is publishable, not perfect.
[12 rules listed]
Draft: [paste]
That's it. No JSON schema, no chain-of-thought theater, no rubric preamble. The "be honest" line matters more than it sounds — without it, Claude drifts generous.
The 75 Threshold
A 75 isn't a quality bar. It's a publish-or-revise gate. In my last 40 posts:
- 12–18 points lost on one or two rules → rewrite that section, ship
- 19–30 points lost → re-edit, re-score, second Claude pass
- Below 75 → I missed something obvious in the brief. Kill the post and start over
The real win isn't the score. It's that "75" turns a vibe check into a decision. I no longer argue with myself about whether a draft is ready — the number tells me.
The Honest Caveat
Claude catches structural and stylistic problems. It does not catch factual errors, fabricated citations, or a post that misses its audience by a mile. I still do a 90-second human fact-check on stats and a one-line "would I actually send this to a client" check. The scorecard gets a draft to the door. The human decides if it walks through.
A quality gate is not quality. It's just a cheap, consistent reason to send a draft back for one more pass.