Verified against Claude · 2026-07-29
Force Claude to critique and score its own draft before you ever see it
A four-pass drafting prompt — draft, numerically scored critique, a dedicated blind-spot check, and a re-scored revision — so an improved draft has to earn its improvement against a rubric, not just read like a nicer version of the same thing.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Do this in explicit passes — do not skip ahead to a single polished answer. TASK A 150-word product description for a stainless-steel insulated water bottle, for an ecommerce listing. AUDIENCE FOR THIS PIECE Online shoppers comparison-browsing several similar bottles in the same price range, skimming on mobile. Calibrate every critique-pass judgment against what would actually land for this specific audience, not a generic "is this well written" standard. PASS 1 — DRAFT Write a first draft at normal effort. This does not need to be your best possible attempt; a competent first try is exactly what's needed here, since the point of the next pass is to find what's actually wrong with it, and an already-optimized draft leaves less for the critique to find honestly. PASS 2 — CRITIQUE Critique the draft as a skeptical editor reviewing someone else's work, not as the author defending a choice already made. Score it against each item below on a scale of 1 to 5, with one sentence of justification per score — a bare number with no reasoning is not a critique: 1. Leads with a concrete benefit, not a generic feature list. 2. Contains at least one specific, checkable claim (capacity, insulation duration, material) rather than vague superlatives. 3. Would a skeptical shopper find this credible, or does it read like every other listing on the page? 4. Scannable in under 10 seconds on a phone screen. If a reference example of strong work exists — The listing for a competitor bottle that leads with "Keeps drinks cold for 24 hours, hot for 12" as its first line — that's the bar for a specific, checkable opening claim. — use it to calibrate what a 5/5 actually looks like, not an abstract ideal. Name the single weakest scoring item explicitly, even if every other score is high. Do not write a critique that is secretly praise with one soft caveat attached — if the draft genuinely has a 2 somewhere, say 2 and say why, not "mostly strong, with minor room for polish." PASS 3 — BLIND SPOT CHECK Before revising, ask: is there a real problem with this draft that the rubric above doesn't actually cover? A rubric is written in advance and can miss something obvious once you're looking at the actual draft. If you find one, name it as an extra item, scored the same way; if you genuinely don't find one, say so rather than inventing a minor issue just to seem thorough. PASS 4 — REVISION Rewrite the draft to address every item scored below 4 out of 5, plus anything from the blind-spot check. If a low score from Pass 2 does not get addressed in the revision, say so explicitly and explain why — a criticism you agree with but choose not to act on is a legitimate outcome, but only if it's stated, not silently dropped. SHOW ALL FOUR PASSES Show the draft, the scored critique, the blind-spot check, and the revision, in that order — not just the final revision. I need to see the critique actually earned the changes made, not take the improvement on faith. FINAL CHECK Re-score the revision against the same rubric. If any item is still below 4, say which one and why a further pass didn't fully resolve it, rather than presenting a still-flawed revision as if the process closed every gap.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
A single-pass 'write this well' prompt gives Claude no way to catch its own weak choices, because the first plausible draft is also the only draft — there is nothing to compare it against. Requiring a numeric score per rubric item, with one sentence of justification, is sharper than open-ended feedback because a number forces a real judgment call rather than a hedge phrase: "mostly strong, minor room for polish" can describe both a 4/5 and a 2/5 equally well, while a required score with justification removes that escape hatch and makes the weakest point of the draft visible rather than smoothed over. The blind-spot check as a pass separate from the rubric matters because a rubric is authored before seeing the actual draft, so it can miss the one thing genuinely wrong that nobody anticipated when writing the checklist; a dedicated pass asking specifically what the rubric does not cover catches the class of error a fixed checklist structurally cannot, by design, catch on its own. Re-scoring the revision against the identical rubric at the end closes the loop by checking that the revision pass actually fixed what it claimed to fix, rather than trusting the model's own narrative claim of having addressed the feedback — this is the mechanism that catches a revision that changed wording without changing the substance a low score was actually about. The reviewer-mode framing, explicitly told to critique as a skeptical editor reviewing someone else's work rather than as the author defending a choice just made, targets a real asymmetry: a model, like a person, tends to defend a decision it just made unless deliberately reframed into an adversarial stance, which is a genuinely different cognitive mode from continuing the same train of thought that produced the draft in the first place. Calibrating every judgment to a stated audience, rather than a generic quality bar, matters because 'good writing' is not a fixed target — a description that scores well for one audience can score poorly for another, and naming the audience up front keeps every score across all four passes anchored to the same standard instead of quietly shifting between passes.
What you get back
Draft: a generic 150-word description leading with material and dimensions. Critique: capacity claim scores 2/5 ("leads with dimensions, which every competing listing also does — no specific, checkable claim like insulation duration"). Blind-spot check: none found beyond the rubric. Revision: restructured to open with "keeps drinks cold for 24 hours," "premium quality" replaced with the specific double-wall vacuum insulation spec. Final re-score: all items at 4/5 or above.
Verified against
Claude Sonnet 4.6 · 2026-07-29
Changelog
- 2026-07-29 — Initial publish, verified against Claude Sonnet 4.6.
Building this for real?
This is a free starting point. If you'd rather have what Scult builds built and running for your business, that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
