The proof is in the editing
Free to use

The EV Rubric

The full marking scheme we use to score AI-written content out of 100. Use it to check your own content, or send us a page for a free score. See it applied to a real piece in the scored example.

Version 1 · October 2026 · © TilyD Ltd

What the rubric measures

The rubric scores a finished piece of AI-written or AI-assisted content for whether it is fit to publish. It does not try to detect whether AI wrote the text. It measures quality, accuracy and risk, the things a client pays an editor to guarantee.

Unit of review: one piece (article, page, email, product description). Pieces over 3,000 words are scored in sections of up to 3,000 words and the lowest section score is reported.

Scoring method

  1. Score each of the 7 dimensions from 0 to 4 using the descriptors below.
  2. Points for a dimension = weight × (score ÷ 4).
  3. Add the points for a total out of 100.
  4. Check the critical fails. Any critical fail caps the result at "Fail", whatever the total.
  5. Record every issue found on the score sheet with its location, so the score can be checked and the fixes made.

Result bands

TotalResultCertificate
90–100Verified, excellentYes
80–89VerifiedYes
65–79Needs editingNo, re-score after edits
0–64FailNo

Critical fails (automatic Fail)

The 7 dimensions

1. Accuracy and sourcing (weight 20)

Are the claims true, current and supported?

ScoreDescriptor
4Every checkable claim is correct. Statistics, dates and names are sourced or easily verified. No outdated information.
3One minor inaccuracy or unsourced claim that does not change the meaning.
2Two or three minor inaccuracies, or one claim that misleads on a secondary point.
1Several inaccuracies, or one that misleads on a main point.
0Accuracy cannot be relied on.

Method: list every checkable claim. Verify at least the 5 most important, or all if fewer than 5.

2. Fit to brief and audience (weight 10)

Does the piece do the job it was commissioned for?

ScoreDescriptor
4Meets every point in the brief. Right length, reading level, keywords and call to action for the stated audience.
3Meets the brief with one small gap.
2Misses one important point or pitches to the wrong level in places.
1Misses several points, or the audience is clearly wrong.
0Does not answer the brief.

If no brief exists, score against the stated or obvious purpose and note "no brief".

3. Clarity and structure (weight 15)

Can the reader follow it and find what they need?

ScoreDescriptor
4Logical order, clear headings, one idea per paragraph, strong opening and close. No passage needs a second read.
3One or two passages are unclear or out of order.
2Structure works but several paragraphs wander or bury the point.
1Hard to follow; reader must work to extract the main points.
0No usable structure.

4. Language correctness (weight 15)

Grammar, spelling, punctuation, usage and consistency (capitals, numbers, hyphenation, UK/US spelling).

Score by errors per 1,000 words:

ScoreErrors per 1,000 words
40–1
32–3
24–6
17–10
0More than 10

Count each error once even if repeated, but count each inconsistency (for example "e-mail" and "email" in one piece) as one error.

5. Style and voice (weight 10)

Does it sound like the client and follow their house style?

ScoreDescriptor
4Consistent tone that matches the client's voice and style guide. Varied sentences.
3Tone slips once or twice.
2Generic voice; could belong to any company.
1Tone is wrong for the brand or audience in several places.
0Voice works against the brand.

6. AI writing patterns (weight 20)

The habits that make AI text read as unedited and cost trust. Count instances of:

ScoreInstances per 1,000 words
40–2, and the piece reads as written by a person who knows the subject
33–5
26–9
110–15
0More than 15

7. Risk and compliance (weight 10)

Could publishing it cause legal, regulatory or reputational harm?

ScoreDescriptor
4No risk found. Claims are qualified where needed. Required disclosures (AI use, affiliate links, sponsorship) are present.
3One low-risk wording issue.
2A missing disclosure or an over-strong claim that needs softening.
1Several risk issues.
0Serious risk. Also triggers a critical fail if it meets the critical-fail list.

EditorVerified does not give legal advice. Check disclosure rules that apply to you with your own adviser.

Score sheet (one per piece)

FieldEntry
Client / piece title
Word count
Brief supplied?Yes / No
Reviewer (named editor)
Date
1. Accuracy (×20)score __ / 4 → __ pts
2. Brief fit (×10)score __ / 4 → __ pts
3. Clarity (×15)score __ / 4 → __ pts
4. Correctness (×15)errors __ per 1,000 → score __ → __ pts
5. Style (×10)score __ / 4 → __ pts
6. AI patterns (×20)instances __ per 1,000 → score __ → __ pts
7. Risk (×10)score __ / 4 → __ pts
Critical failsNone / list
Total__ / 100
ResultVerified excellent / Verified / Needs editing / Fail
Issue loglocation, dimension, issue, suggested fix