Verified against ChatGPT · 2026-08-14
Build an editorial quality rubric specific enough that two different reviewers would score the same draft the same way
Creates a content quality scoring rubric tailored to your specific content type and goals, with criteria concrete enough that inconsistent reviewers become a solvable problem instead of a permanent one.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Build a content quality rubric for reviewing drafts before publication. The goal is a rubric specific enough that two different reviewers scoring the same draft independently would land on close to the same score — not a generic "clarity, accuracy, engagement" rubric that leaves too much to individual reviewer judgment. CONTENT TYPE Long-form SEO blog posts (1,200-2,500 words) written by a mix of in-house writers and freelancers. WHAT GOOD ALREADY LOOKS LIKE (an example if you have one) Our 'switching payroll providers' guide — specific numbers throughout, named real scenarios, no generic filler paragraphs. WHAT'S ACTUALLY GONE WRONG BEFORE A freelancer submitted a draft that repeated the same point in three different sections using different phrasing, padding the word count without adding information; another draft made a specific pricing claim that turned out to be outdated by the time it published. For each criterion in the rubric, write it so that a reviewer could point to a specific passage in the draft and say definitively whether it passes or fails that criterion — reject any criterion that would require the reviewer's subjective taste to resolve (like "is it engaging") unless you can restate it as something checkable (like "does the opening paragraph state a specific claim within the first two sentences, rather than a general framing that could apply to any article on this topic"). Build the rubric directly around the past failures given — if a specific kind of mistake has actually happened before, there should be a criterion that would have caught it, stated precisely enough to catch the same mistake again, not just a generic quality dimension that happens to be adjacent to it. Use a 3-point scale per criterion (fails / partially meets / meets) rather than a 10-point scale, since finer gradations on subjective judgment add false precision without actually making reviewers more consistent with each other. OUTPUT FORMAT 1. The rubric as a table: Criterion | What "fails" looks like (specific) | What "meets" looks like (specific) | Which past failure (if any) this criterion was built to catch. 2. A one-paragraph note on any quality dimension you deliberately left out of the rubric because it couldn't be made concrete enough to score consistently, so it's tracked as a known gap rather than silently dropped. 3. A suggested pass threshold (e.g. no criterion below "partially meets," or a minimum count of "meets") for what counts as publishable.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Generic content rubrics built around dimensions like "clarity" and "engagement" fail at the one job a rubric exists to do — producing consistent scores across different reviewers — because those words don't resolve to a checkable observation in the text, so two reviewers with different personal taste will genuinely disagree on the same draft even when both are acting in good faith; requiring every criterion to be restatable as something a reviewer could point to a specific passage and definitively check forces GPT-5.1 to translate vague quality language into an operational test, which is the actual mechanism that makes inter-reviewer consistency possible. Building the rubric directly around named past failures rather than a generic quality checklist matters because a rubric that doesn't specifically target the mistakes that have actually happened will keep letting them through — a generic "originality" criterion doesn't reliably catch a draft that pads word count by restating the same point three ways, but a criterion asking "does any section restate a point already made elsewhere in different words" catches exactly that failure by name. Capping the scale at 3 points rather than defaulting to a 10-point scale, which is GPT-5.1's more common default for rubric requests, directly addresses a known measurement problem: finer gradations on an inherently subjective judgment create an illusion of precision without actually improving reviewer agreement, since the extra distinctions between, say, a 6 and a 7 are exactly the kind of judgment call the rest of this rubric is designed to eliminate. Requiring an explicit note on dimensions deliberately left out keeps the rubric honest about its own limits rather than silently pretending it covers every aspect of quality, which matters operationally because a content team needs to know a real quality dimension is being tracked outside the rubric, not assume the rubric is complete just because it exists.
What you get back
Criterion: 'Redundancy check' | Fails: same core point restated in two or more sections using different wording without adding new information | Meets: each section advances the argument with new information, no repeated point | Built to catch: freelancer draft that padded word count via repetition. Pass threshold: no criterion below 'partially meets,' and at least two-thirds of criteria at 'meets.' Left out: overall 'brand voice fit' — too dependent on reviewer-specific taste to make checkable without a much larger style guide than given here; tracked as a known gap, not silently dropped.
Verified against
ChatGPT GPT-5.1 · 2026-08-14
Changelog
- 2026-08-14 — Initial publish, verified against ChatGPT GPT-5.1.
Building this for real?
This is a free starting point. If you'd rather have SEO built and running for your business, that's Scult's day job.
EXPLORE SEO
