Michele Meloni
Contact

Write to me directly.

For a collaboration, a question about my work or an initial conversation, use this form. For a detailed estimate, use the dedicated flow on the Services page.

Your message is sent directly to info@michelemeloni.com.

Back to articles
Article · AI & QualityMichele Meloni7 min read

How to evaluate the quality of AI-generated output

A practical checklist for checking correctness, relevance, tone, safety and usability before generated output becomes a deliverable.

Published · 4 September 2026Updated · 4 September 2026

00 · Premise

“It looks right” is not a quality criterion.

An output can be fluent and polished while still containing errors, omissions or choices that do not fit the context. Review works when it starts from criteria defined before generation.

A four-stage review

Correctness, fit, risk and testing in context.

Promptario organises resources and methods around the use of AI. The same approach applies to review: a repeatable sequence is more reliable than a generic judgement of the result.

  1. 01Correctness

    Facts, data, names, sources and function.

  2. 02Fit

    Goal, audience, tone and constraints.

  3. 03Risk

    Privacy, safety, rights and potential harm.

  4. 04Context

    Test in the real channel, format or journey.

Promptario interface with a catalogue of tools and guides for artificial intelligence
Promptario · turn criteria and verification into a repeatable practice.
01

Set criteria before generating

Quality means different things for informative copy, a visual, a prototype or an analysis. Define requirements and thresholds before seeing the output.

Practical actions

  • Separate mandatory requirements from preferences.
  • Define what would make the result unusable.
  • Assign responsibility for final review.
02

Check accuracy and completeness

Verify every element that can be compared with a source, figure or rule. Look for what is missing as well as what is wrong.

Practical actions

  • Check names, figures, quotations and links.
  • Compare the output with supplied material.
  • Mark unstated assumptions and missing information.
03

Evaluate fit with the context

A correct result can still be wrong for the audience, tone, channel or stage of the project. Readability and usefulness matter as much as form.

Practical actions

  • Reread from the recipient’s point of view.
  • Remove generic language and needless jargon.
  • Check inclusion, accessibility and brand consistency.
04

Test and document

Place the output in the page, journey or final file. Record corrections that reveal recurring weaknesses and use them to improve the method.

Practical actions

  • Test intended sizes, states and devices.
  • Keep instructive failures as anti-examples.
  • Update prompts, criteria or process from the results.

Ready to use

Quality gate before delivery

  • Goal and criteria are explicit.
  • Facts and sources have been verified.
  • Omissions and assumptions have been checked.
  • Privacy, rights and safety are respected.
  • The result works in its real context.
  • A person owns the final decision.

Key principle

AI quality belongs to the process, not the first output.

Value comes from clear criteria, comparison with sources, competent review and verification in the context where the result will be used.