AI performance reviews

An AI performance review is a performance review in which the manager uses a generative AI tool, such as ChatGPT, Copilot, Gemini or the AI features now built into HR software, to summarise evidence, draft written feedback or suggest a rating. Used well, AI saves a manager hours of writing. Used as a substitute for the manager's own judgement, it produces reviews that read smoothly, say very little, and cost the trust a review is supposed to build.

Everyone is doing it. Quietly.

How many of your peers are pasting notes into a chatbot the night before review day? More than will admit it. In an August 2026 survey of 1,034 US corporate professionals by Highwire, only 16% of employees said they had been told AI played a role in their most recent review, while another 40% suspected it had been used or were not sure (HCAMag). A July 2026 survey by The Predictive Index found that 44% of managers admit entering employee names or performance details into public AI tools.

So the question for most managers is no longer whether to use AI in performance reviews. It is whether you are using it in a way you would be comfortable explaining to the person sitting across the table.

Where AI earns its place

Start with the good news, because there is plenty of it. A manager with a year of scattered notes, two project summaries and a stack of peer comments faces a writing job, and writing is something these tools do quickly. AI can turn rough notes into a first draft, put the evidence under the right objectives, and suggest a clearer way to say something awkward. It can also act as a second pair of eyes: ask it to flag vague phrases, personality labels or words that read differently for men and women, and it will often catch what you missed.

There is research behind the quality claim too. A field study at a large financial services company, published in the Strategic Management Journal, found that employees who received AI-generated feedback improved their performance more than those who received feedback from their managers. The catch arrived when people were told where the feedback came from. Once employees knew a machine had written it, trust fell and the performance gain went with it.

That finding is the whole argument of this page in one result. The quality of the words matters less than whose words they are.

The calculator in the exam hall

Think of AI in a review the way a maths teacher thinks of a calculator. In the hands of a student who understands the method, it removes the drudgery and gets to the right answer faster. In the hands of a student who does not, it produces a confident wrong answer, neatly laid out, and nobody notices until the marks come back.

A performance review works the same way. The tool can only arrange what you give it. If you know what good performance looks like in the role, have gathered evidence through the year and have decided what you want the employee to do next, AI will help you say it well. If you have not done that work, it will fill the gaps with fluent, plausible text. We call that the borrowed review: polished on the page, empty in the conversation, and easy for the employee to spot.

What goes wrong when managers lean too hard on AI

Employees can tell, and they trust you less

Researchers Anthony Coman and Peter Cardon surveyed 1,100 US professionals for the International Journal of Business Communication. When a supervisor's message used only light AI help, 83% saw the supervisor as sincere. With heavy AI assistance that fell to between 40% and 52% (HR Dive). The researchers concluded that messages involving praise, motivation or personal feedback are better handled with minimal help from technology. A performance review is all three.

The tool was not there all year

AI knows only what you paste in. It did not see the employee cover for a colleague in March or rescue a client in July, so unless your notes record it, the draft leaves it out. This makes the oldest appraisal error, recency bias, worse, because the most recent and best-documented work fills the page. The fix is upstream: keep evidence through the year, which is what the IPM Performance Framework calls the Evidence stage.

Bias goes in, bias comes out polished

If your notes describe one employee as "assertive" and another doing the same thing as "abrasive", the AI will not correct you. It will write both up beautifully. Language models can also add patterns of their own from the text they learned on. Any rating or adjective the tool suggests needs the same test you would apply to your own: what is the evidence, and would I say this about anyone else doing the same work? There is more on where bias enters a rating in inclusive performance management.

Confidential data leaves the building

Performance details, health disclosures, salary and disciplinary history are some of the most sensitive data a company holds. Pasting them into a public chatbot may breach your company's policy, your data protection obligations or both. That 44% figure from The Predictive Index should worry every HR director who reads it. Use only the tools your organisation has approved for personal data, and strip names out when you can.

The regulators have noticed

Under the EU AI Act, AI systems used to evaluate employee performance are classed as high-risk. The Digital Omnibus on AI moved the date those obligations apply to 2 December 2027, but the classification itself is unchanged (DLA Piper). Any company with staff in Europe will need human oversight of AI in performance decisions, and a manager who cannot explain a rating without the tool will struggle to provide it.

Rules for using AI in a performance review

1. Decide the rating before you open the tool. The judgement is yours. AI can help you explain it, not reach it.
2. Feed it evidence, not impressions. Specific examples, results against agreed objectives and dates. Vague notes produce vague reviews.
3. Keep personal data out of public tools. Use your company's approved tool, or remove names and identifying details first.
4. Rewrite the draft in your own voice. If a sentence does not sound like something you would say out loud, change it. The employee knows how you talk.
5. Check every claim against the record. AI can state things that never happened with total confidence. Nothing goes into a review that you cannot point to.
6. Be open about it. A short line such as "I used AI to help organise my notes" costs far less than the employee finding out later.
7. Hold the conversation yourself. The written review is the record. The meeting is where performance actually changes, and no tool can do that part for you. If that meeting is the part you dread, performance review training for managers is where to start.

Why training matters more now, not less

Here is the uncomfortable part. AI does not level managers out. It widens the gap between them. A manager who knows how to set measurable objectives, gather evidence, spot bias and run a difficult conversation gets a better first draft and knows exactly what to fix in it. A manager who never learned those skills gets a confident draft of a weak review, with nothing to tell them it is weak.

That is why formal performance management training is worth more in the age of AI, not less. The tool handles the typing. The skill is still yours to learn.

Questions

Is it OK to use AI to write a performance review?

Yes, as a writing aid, provided you have already decided the rating, you feed it real evidence, you keep personal data out of public tools and you rewrite the result in your own words. Using AI to decide what the review says, rather than how it says it, is where the problems start.

Should I tell my employee I used AI?

Yes. Most employees already suspect it, and research on AI-written workplace messages shows trust drops sharply when people believe their manager leaned heavily on AI. A short, honest line about how you used it protects the relationship far better than silence.

Can AI give an employee a performance rating?

It can suggest one, but the manager must own the decision and be able to explain it from the evidence. In the EU, AI used to evaluate employee performance is classed as high-risk under the AI Act, which requires human oversight.

Is ChatGPT safe for performance reviews?

Not with identifiable employee data unless your organisation has approved it for that use. Check your company's AI and data protection policy. If in doubt, remove names and personal details before pasting anything in.

AI will write a review in thirty seconds. Earning the right to deliver it still takes a year.

Where this is taught

Gathering evidence, avoiding rating errors and running the review conversation are all part of the Certificate in Performance Management, our performance management training for managers, and the review itself is the subject of performance appraisal training for managers. You can also start with the free performance review template or the guide to writing a performance improvement plan. For the wider picture beyond the review itself, see AI in performance management.