A meaningful human review under GDPR Article 22 needs a reviewer with the authority to change the decision, the competence to judge it, all the relevant information, enough time, and the independence to disagree with the system. A signature on an automated outcome does not count. Each condition is a demand on you.
What makes a review "meaningful" under Article 22?
Article 22 gives people the right not to be subject to decisions based solely on automated processing that have legal or similarly significant effects. Organisations often avoid that restriction by putting a person into the decision. Regulators' guidance has been clear, in substance, that this only works if the human involvement is real: a person who routinely applies the system's output without weighing it does not stop the decision being solely automated.
The same standard applies when a person exercises their right to obtain human intervention after an automated decision. The reviewer must actually reconsider the case, not confirm what the system did.
So "meaningful" is not a legal technicality. It describes how the reviewer works. The wider role is described in GDPR Article 22 and the human reviewer's job; this article is about the standard you hold yourself to inside it.
What does it require of the person doing it?
Five conditions, each a habit you can practise.
Authority. You can change the outcome. If your only options are "approve" or "send back to the same system", you are not reviewing. Know what your authority is, and say so when a case needs more.
Competence. You understand the decision being made: the policy, the criteria, what good and bad evidence looks like. A reviewer of credit decisions needs to understand affordability; a reviewer of claims needs to understand cover.
All relevant information. You look at the evidence, not only the agent's summary of it. If the agent's recommendation rests on a document, open the document. If a source is weak, such as an unverified statement or an emailed attachment, weigh it as weak.
Time. You have enough time per case to do the above. If volume makes that impossible, raising it is part of the job. A review done in four seconds cannot be meaningful, whatever the form says.
Independence from the output. You resist automation bias, the tendency to accept a system's answer because it came from a system. A good habit is to form a view from the evidence before reading the recommendation.
Above all of these sits the written reason. A meaningful review leaves a record of what you looked at and why you decided as you did. A short definition of the concept is in what is meaningful human review.
Who needs reviewers who meet this standard?
Any organisation subject to the GDPR that makes significant decisions about people with automated help: lenders, insurers, collections teams, payment firms, employers and public bodies. Collections is a clear example, because decisions about arrangements and escalation can significantly affect someone in financial difficulty; employers' training there is covered in training reviewers for meaningful human review in collections.
The UK applies the same idea through UK GDPR and the Data Protection Act 2018, so the standard travels across both markets.
How do you show an employer you review meaningfully?
| Requirement | What employers will expect of you | How to show it |
|---|---|---|
| Authority to change the decision | Use it, and escalate where your authority ends | Examples of overrides and escalations, with reasons |
| Competence in the decision area | Explain the criteria and how they apply | A worked case explained in plain words |
| Consideration of all relevant data | Check underlying evidence, not just summaries | Decision notes that cite the evidence used |
| Real reconsideration | Reach your own view before reading the agent's | Review decisions to a standard, including disagreements |
| Engagement with the person's view | Address what the person said when they contest | A response to an appeal that answers their points |
| A record of reasoning | Write a reason that could be shown to the person or a regulator | Anonymised samples of your reasons |
Employers increasingly measure reviewers on accuracy and on the quality of their reasons, not on speed alone. A record showing that you overturn the agent when the evidence requires, and agree when it does not, is stronger evidence than a high approval count.
What might an interviewer ask to test this?
- "How would you know you had become a rubber stamp?" Look at your own override pattern, the time you spend per case, and whether you can explain each decision without the agent's text.
- "The agent's recommendation is based on an income figure the customer gave. Is that enough?" It is declared, not verified. Weigh it accordingly and check whether policy requires verification.
- "What would you do if your team's targets made proper review impossible?" Raise it in writing with the evidence. Meaningful review is a legal requirement, not a preference.
A claims assessor we will call Nadia practised by timing herself on old cases with and without reading the system's recommendation first. She found she disagreed more often when she read the evidence first, and used that in interviews to explain automation bias without jargon.
What does Article 22 not specify?
It does not set a minimum time per case, a caseload, or a qualification for reviewers. It does not say who in the organisation must review. It does not cover decisions without legal or similarly significant effects. And it does not shift the organisation's accountability onto you; the controller remains responsible, which is one reason organisations want named, competent people behind every decision. That argument is developed in every AI agent needs a person with a name, and that person needs a career.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
The Operator track at Pryme Intelligence Academy trains review to a standard: how a governed agent decides, human review, and teaching the agent from corrections. The Certified Pryme Operator practicum asks for two weeks of review decisions, nineteen in twenty correct and no missed red line, signed off by an assessor. That is evidence of meaningful review an employer can check.
Questions people ask
Does a meaningful review always mean disagreeing sometimes?
Not necessarily in every batch, but a reviewer who never disagrees over a long period should ask why. Real review produces disagreements when the evidence calls for them.
Do Swiss rules work the same way?
Switzerland is outside the EU, but its data protection law, in substance, also gives people ways to have certain automated decisions reviewed by a person. See how to become a human-in-the-loop specialist in Switzerland.
Where does this skill lead?
To senior reviewer, quality and agent owner roles. The guide to AI agent jobs and careers maps the routes.