The Anti-AI score, explained
Every post in Postbrander gets a live score from 0 to 100 against the same rules that govern generation. The score updates as you type. The point is to make the rules visible so you can see what slipped through.
How to read the score
The badge in the post editor shows a number from 0 to 100. Higher is better. Anything 90+ is clean. Between 75 and 90 is acceptable but has at least one tell. Under 75 and the post reads as AI to anyone who notices these things.
- 90-100 Excellent: ready to publish
- 75-89 Good: minor tells, worth a pass
- 50-74 Has AI tells: needs editing or auto-fix
- 0-49 Reads as AI: rewrite the flagged sentences
What the rules check
Three categories, each worth different point deductions:
- Banned words (4 points each): 90+ words statistically over-represented in AI output. Examples: delve, leverage, harness, intricate, crucial, pivotal, synergy, holistic, foster, garner, seamless, supercharge.
- Banned phrases (6 points each): 30+ dead phrases and transitions. Examples: "In today's...", "Furthermore", "Let's dive in", "Game-changer".
- Banned sentence patterns (10 points each): negative parallelisms ("This isn't X. This is Y."), copulative avoidance ("serves as" instead of "is"), participle-phrase puffery ("highlighting its importance"). These are the strongest AI tells, hence the heaviest weighting.
See the complete rule set on the public anti-AI rules page.
Using auto-fix
When the score drops below 100, expanding the Anti-AI panel shows a list of violations. The Auto-fix button sends only the flagged sentences back to the AI with the rules attached, and gets a rewritten version of just those sentences. The rest of the post stays exactly as you wrote it.
Auto-fix counts as one AI generation against your monthly quota. It runs on Claude Sonnet, takes 3-5 seconds, and is idempotent (clicking it on a clean post returns "already clean, nothing to do" without spending the credit).
When to ignore a violation
The rules are a guide, not a religion. If a banned word is genuinely the right word for what you're saying, use it. The score is there to make you think about whether the word is load-bearing or a tic. Most of the time it's a tic.
One exception worth taking seriously: negative parallelisms. The "This isn't X. This is Y." pattern is so over-represented in AI output that readers register it as machine-written before they finish the sentence. Even when it feels punchy. Especially then.