A good performance review comment is specific enough to survive calibration. It says what the person delivered, gives one or two pieces of evidence another manager could check, names the behaviour that drove it, and ends with a clear next step. Vague praise, personality labels and absolutes (“always”, “never”) do not meet that test.

This guide is for line managers staring at a blank review box, for employees writing a self-evaluation, and for HR putting a manager pack together. The examples are organised by the playbook’s five rating levels — Poor, Basic, Intermediate, Advanced and Exceptional — from the employee performance rating scales page, each paired with the evidence that would justify it, plus examples by area, before-and-after rewrites and self-evaluations you can adapt. Use them with the performance review template or the UK staff appraisal template. If you are the People lead launching the cycle, use the year-end performance review run sheet and template. If you are collecting peer or upward input, use the 360 feedback template first, then adapt the themes into comments like the ones below. When you write the skills section of a review, score against the behaviours in the competency matrix templates rather than inventing a new bar each cycle. Disclosure: Ash Taherian is Product Sales Executive at Revolut People, which sells performance management software. Treat product mentions as coming from an interested party.

Every comment and self-evaluation on this page is illustrative. The names, roles and figures were written for this guide. They are not real employee data.

What a good review comment includes

Write about outcomes and behaviour, not personality. For each point, cover four things:

The four parts of a comment that can be checked in calibration.
PartWhat to write
OutcomeWhat was delivered, or what was missed, in this review period.
EvidenceOne or two checkable facts: a date, a shipped piece of work, a metric, a recorded decision, or feedback you can point to.
BehaviourWhat the person did that led to the outcome — something another manager could observe, not a trait.
Next stepWhat to continue, change or learn in the next period, in concrete terms.

For example: “You shipped the billing migration on the agreed date and flagged two risks early, which let finance plan around them. Next quarter, bring those risk updates into the weekly product review.” The outcome is the migration landing on time. The evidence is the agreed date and the two early risk flags. The behaviour is raising those risks in time for finance to act. The next step is a specific change to how those updates are shared.

The same four parts make a short evidence note on a scorecard. The review template asks for one checkable line per statement; this page shows the fuller comment. If you are writing in appraisal language, use the same four parts on the staff appraisal template.

Examples by rating level

Use the exact labels on the five-level scale: 1 Poor, 2 Basic, 3 Intermediate, 4 Advanced and 5 Exceptional. The short definition under each heading is taken from that page. Grade against the talent bar for the person’s seniority, then write a comment a calibration room could test. Intermediate is a solid rating when it is earned, not a soft fail. The five people below are fictional mid-level operations or product roles, so the difference is the rating and the evidence, not the job title.

Poor

Definition, from the scale: performance is well below the standard for the role — work is late, needs heavy rework or stays within very familiar tasks.

Illustrative comment. “Jordan regularly missed the weekly operations pack deadline — four of the last twelve packs were more than a day late — and two packs needed a full rewrite after finance found errors in the figures. Newer requests were parked. Next period, we agreed a Wednesday draft check with finance and a written standard for the pack, so timing and quality are visible before Friday circulation.”

Evidence that justifies it. The pack calendar shows four late Fridays in twelve weeks; two finance emails asked for a full rewrite; the request log shows two non-standard packs still open, with no draft started.

Why it would hold up in calibration. The dates, rewrite requests and open work can be checked. The comment describes the work, not Jordan’s personality, and ends with an agreed next step.

Basic

Definition, from the scale: dependable on standard work, but needs support and iteration to reach the quality expected.

Illustrative comment. “Priya delivered the standard weekly packs on time in ten of twelve weeks, and the figures were usable. Four packs needed two or three rounds of correction after review. When a one-off request came in from sales, Priya needed close guidance to finish it. Next period, we agreed Priya will send a first draft of any non-standard pack by Wednesday and keep a short error log so the same corrections are not repeated.”

Evidence that justifies it. Ten on-time standard packs; review comments on four packs that needed more than one correction cycle; a sales pack that, on the 1:1 record, took two guided sessions to finish.

Why it would hold up in calibration. It distinguishes reliable standard work from the extra support still needed. Basic is not Poor: the work lands, but not yet at the quality or independence the level requires.

Intermediate

Definition, from the scale: consistently high-standard work, delivered on time, including tasks without a clear template. Intermediate is a full, solid rating against the talent bar for that seniority, not a dumping ground.

Illustrative comment. “Alex shipped the billing migration on the agreed date and flagged two risks early, which let finance plan around them. The runbook Alex wrote is now the default for similar migrations. Work was high standard after one review pass, including a non-standard cutover plan with no existing template. Next quarter, bring those risk updates into the weekly product review so the wider team sees them in time to act.”

Evidence that justifies it. The agreed cutover date and the date it ran; two risk notes sent to finance the week before; the new runbook, already used by a second team; a single review pass on the cutover plan.

Why it would hold up in calibration. The evidence covers on-time delivery, quality after one iteration, and work without a template — what Intermediate asks for. The next step stretches how the work is shared, not a claim that this period was Advanced.

Advanced

Definition, from the scale: fast, first-time quality on complex work that has little precedent.

Illustrative comment. “Morgan rebuilt the billing cutover plan for a first-time, multi-entity migration, shipped it a week early, and the first-pass quality meant finance signed off without a second cycle. Morgan also coached two peers through the new checklist so the team is not dependent on one person. Next period, we want Morgan to set the quality bar for the next two migrations, not only deliver this one.”

Evidence that justifies it. The signed plan had no previous multi-entity run to copy; cutover landed a week early; finance signed off on the first formal review; two peer coaching sessions are in the 1:1 notes, and both peers have since run a checklist item on their own.

Why it would hold up in calibration. The work is complex, has little precedent, and is high standard the first time — which is what sits above Intermediate. Coaching is useful context; the grade rests on the new plan, which the room can inspect.

Exceptional

Definition, from the scale: shapes the work itself — sets targets, designs new processes and links tasks to the business vision. It is reserved for people who change how the work is done, not only how quickly they do it.

Illustrative comment. “Lee identified that billing migrations were being planned case by case, designed a new shared process that finance and product now use, and tied it to the quarter’s cash-collection target. The first two migrations under the new process beat their cutover windows by more than a week. Lee also retired two manual steps that had been the source of previous errors.”

Evidence that justifies it. A written process now used by finance and product; two migrations dated against their original windows; a decision note listing the two manual steps removed and the earlier errors they caused; Lee’s write-up linking the process to the quarter’s cash-collection target.

Why it would hold up in calibration. It shows new targets, a new process and a link to a business aim — the Exceptional definition. Without that change to how the work is done, the same speed on one migration would be Advanced.

Examples by area

The same four parts apply when you write about delivery, collaboration, communication, ownership or leading others. These are also illustrative, and fictional.

Illustrative comments by area, written for this page. They are not real employee data.
AreaIllustrative comment and evidence
Delivery“Sam reduced the team’s average first-response time on the priority inbox from two working days to one, and kept that time for the last eight weeks of the period. The weekly inbox report is the evidence, and the next step is to document the triage rules so cover does not slip when Sam is away.”
Collaboration“Taylor set up a Tuesday 20-minute stand-up with finance after two missed hand-offs in March, and there have been no missed hand-offs since. Both teams now use the same status note. Next period, Taylor will add a short written update on weeks when the stand-up cannot go ahead.”
Communication“Chris’s weekly risk note to product and finance is now the source of truth for the migration: it named two risks a week early and gave each an owner and a date. Next quarter, bring that same note into the weekly product review, rather than leaving it in email.”
Ownership“Omar picked up the vendor issue when the usual owner was off, closed it with a written decision in three days, and updated the runbook so the next person does not start from scratch. The decision note and the runbook change are dated and in the shared folder.”
Leading others“Nina coached two new starters through their first cycle of the operations pack. Both delivered an on-time pack in their third week, and the training checklist Nina wrote is what the next starter will use. Next period, we want one of those starters to run the Wednesday draft check, so the coaching is not held only by Nina.”

If you are building a scorecard rather than a paragraph, put these as short evidence notes on the review template, or get behavioural anchors from the review scorecard builder.

Before and after: rewriting vague comments

These eight pairs show the usual failure modes: empty praise, missing evidence, absolutes, personality labels and recency. The rewrites are illustrative, not real reviews.

Vague: “Jordan is a great team player.”

Rewritten: “Jordan covered two colleagues’ priority inboxes during leave in May, and first-response time on those queues stayed at one working day. Next period, we want that cover written into the rota, not arranged ad hoc.”

Why the rewrite holds up: It names the work, a date, a checkable result and a next step. “Team player” cannot be tested in calibration.

Vague: “Priya needs to be more proactive.”

Rewritten: “On the April sales pack, Priya waited for a chased request before starting the draft, which left one day for review. We agreed that any non-standard pack is drafted by Wednesday, and that Priya flags clashes in the Monday 1:1 rather than absorbing them.”

Why the rewrite holds up: “Proactive” is a personality judgement. The rewrite describes one missed behaviour, its effect, and an agreed change.

Vague: “Alex always delivers.”

Rewritten: “Alex shipped the billing migration on the agreed date and completed eight of eight weekly status notes on time. Two mid-period requests were re-scoped in the Monday 1:1 so the date did not slip.”

Why the rewrite holds up: “Always” is an absolute the room cannot verify. Named deliverables and a count for this period can be checked, and they leave room for a fair discussion of what was dropped.

Vague: “Watch your tone in meetings. You can come across as abrasive.”

Rewritten: “In the 9 April product review, Morgan interrupted a finance update twice and the risk item was not finished. We agreed Morgan will let the update end, then put comments in the shared note, and we will review that in four weeks.”

Why the rewrite holds up: The rewrite describes a date, a behaviour and an effect, then a next step. “Abrasive” and “tone” are personality labels. They are also the pattern discussed below, and in how to separate personality from performance.

Vague: “Strong quarter — the October launch was excellent.”

Rewritten: “The October launch landed on the agreed date and the first-week error rate stayed within the agreed threshold. Earlier in the period, the July checkpoint slipped by two weeks; Morgan flagged it, reset the date with product, and the October date still held. The grade is for the whole period, not only the launch.”

Why the rewrite holds up: It covers more than the most recent success, which is how you limit recency bias.

Vague: “Lee is not a leader.”

Rewritten: “Lee delivered their own work to the agreed standard, but did not run the agreed weekly coaching slot with the two new starters. Both starters had to come to me for the pack checklist. Next period, Lee will run that slot and we will review the starters’ third-week packs together.”

Why the rewrite holds up: “Not a leader” is a label. The rewrite points to an agreed responsibility that was missed, and to a next step the room can follow in the next cycle.

Vague: “Exceeded expectations this period.”

Rewritten: “Taylor rebuilt the cutover plan for a first-time, multi-entity migration, shipped it a week early, and finance signed off on the first review pass. That matches Advanced on the published scale: fast, first-time quality on complex work with little precedent.”

Why the rewrite holds up: “Exceeded expectations” does not say which expectations, or against which level. A named outcome plus the published definition gives calibration something to test.

Vague: “Quiet in meetings; needs to speak up more.”

Rewritten: “In the weekly product review, Chris did not raise the two billing risks until a private follow-up later that day, which meant finance heard them a day late. We agreed those risks go into the shared note before the meeting, and that Chris owns that agenda item next quarter.”

Why the rewrite holds up: Quietness is a personality judgement. The rewrite describes when information arrived, what that delayed, and a change to how it is shared — which is about the work.

Self-evaluation examples

A self-evaluation is not a defence of your grade. It is your evidence pack: what you delivered, what you would do differently, and what you need from your manager next period. The two examples below are fictional first-person write-ups, not real submissions.

For individual contributors

Illustrative self-evaluation, mid-level operations coordinator (fictional).

“My three most important results this period were: I shipped the billing migration on the agreed 14 September date; I cut the priority-inbox first-response time from two working days to one and held it for the last eight weeks, using the weekly inbox report as the record; and I wrote a runbook that the second operations pod has already used on a later cutover.

“What I would do differently: I flagged two migration risks to finance a week early, but I left them in email, so product only saw them after the weekly review. Next period I will put those updates in the shared note before the meeting.

“What I need from my manager: a nominated reviewer for non-standard packs by Wednesday, and a clear steer in the Monday 1:1 when two projects overlap, rather than leaving me to absorb both.”

For managers

Illustrative self-evaluation, engineering manager (fictional).

“My three most important results this period were: the team shipped the multi-entity billing cutover a week early, with finance sign-off on the first review pass; both new starters delivered an on-time operations pack in their third week, using the checklist I wrote; and we had no missed finance hand-offs after I set up the Tuesday stand-up in March, following two misses earlier in the period.

“What I would do differently: I spent most of October on the cutover and delayed two agreed weekly coaching slots, so one starter had to go to my director for the pack checklist. Next period I will keep those slots through delivery peaks, and ask a starter to run the Wednesday draft check.

“What I need from my manager: a decision, before the next cycle opens, on whether the second migration is in this team’s scope, so I can set a talent bar and a scorecard rather than adding it mid-cycle; and a named deputy for the Tuesday finance stand-up when I am off.”

How to write yours

Pick your three most important results from the period and back each with evidence — a metric, a shipped piece of work, or feedback another person could check. Say what you would do differently, and name one thing you need from your manager to do better next period. Keep to the review period. Three evidenced results will serve you better than twelve unexplained ones.

If your organisation uses a formal self-appraisal form, the same structure still holds. Acas publishes a self-appraisal template among its free appraisal form templates (as checked 13 Nov 2026); pair the form with evidence, not only ratings. For the conversation itself, see what questions a manager should ask in a performance review and the review-prep checklist.

Comments to avoid

Three patterns cause the most trouble in calibration, and they often arrive in the same sentence.

Personality, not performance. Labels such as “abrasive”, “watch your tone”, “not a leader” or “great attitude” describe the person, not the work. In Kieran Snyder’s study of 248 performance reviews from 180 people at 28 technology companies, published in Fortune on 26 August 2014 (as checked 13 Nov 2026), negative personality criticism appeared in 71 of 94 critical reviews received by women and in 2 of 83 received by men. That is a writing problem and a fairness problem. Rewrite the comment as a date, a behaviour, an effect and a next step, as in the before-and-after pairs above. The fuller treatment, including how to coach managers off this language, is in how to separate personality from performance.

Absolutes. “Always”, “never”, “everyone” and “no one” are almost never accurate for a whole period, and they are easy to knock down in the room. Replace them with a count, a date range or a named piece of work.

Recency. If every example is from the last month, the comment is a snapshot, not a review. Go back through the period — goals, 1:1 notes, shipped work — and include at least one earlier example, especially if the recent picture is unusually strong or weak. The performance review bias guide covers recency, halo and horn, and leniency; the free bias checker will flag some of these patterns in a draft.

Using AI to draft comments

If you use software to help draft a review, treat the output as a starting point, not the review itself. In Revolut People, the documented AI assistance for performance feedback can generate a draft from recorded 1:1 meetings — which may include strengths, areas of development and suggested wording based on that 1:1 history — and it can also check feedback you write yourself. As you type, a quality indicator rates the draft Poor, Basic or Great against six checks, including whether it is constructive, evidence-based and actionable, and whether it is around 50 to 100 words. You can keep the draft and edit it, or clear the field and write from scratch. The product documentation, as checked 13 Nov 2026, says AI suggestions may be inaccurate or incomplete and should always be reviewed and revised before submitting, so they reflect your own assessment. It does not say the AI writes the review for you.

The same four parts still apply: outcome, evidence, behaviour and next step. Add the examples the draft is missing, strip any personality language, and make sure the comment matches the rating you are actually putting forward. For the documented feature list, see the Playbook page on Revolut People AI and MCP and Revolut People’s AI assistance for performance feedback (checked 13 Nov 2026).

Related guides

360 feedback template: questions by rater · competency matrix templates by role and level · Year-end performance review: HR run sheet and template · Employee performance rating scales explained · performance review template with worked examples · staff appraisal template (UK) · performance calibration · performance management framework · what questions a manager should ask in a performance review · how to separate personality from performance · performance review bias checker · Revolut People AI and MCP · free tools · People Performance Playbook home

FAQ

What should a manager write in a performance review?

Write about outcomes and behaviour, not personality. For each point, say what the person delivered, give one or two pieces of evidence, name the behaviour that drove it, and end with a clear next step. For example: “You shipped the billing migration on the agreed date and flagged two risks early, which let finance plan around them. Next quarter, bring those risk updates into the weekly product review.”

What are good performance review examples?

Good examples are specific, tied to a published rating level, and paired with evidence another manager could check. This page has illustrative comments for Poor, Basic, Intermediate, Advanced and Exceptional, plus rewrites of vague comments and self-evaluation examples. They were written for this guide and are not real employee data.

How do I write a self-evaluation?

Pick your three most important results from the period and back each with evidence (a metric, a shipped piece of work, feedback). Say what you would do differently, and name one thing you need from your manager to do better next period.

What should I avoid writing?

Avoid personality labels (“abrasive”, “watch your tone”, “great attitude”), absolutes (“always”, “never”), and comments that only cover the last few weeks of the period. Rewrite them as a date, a behaviour, an effect and a next step, and spread evidence across the whole review period.

Should the comment match the rating?

Yes. A comment that reads as Intermediate should not sit under an Advanced grade, and Exceptional needs evidence that the person changed how the work is done, not only that they were fast. Use the published definitions on the rating scales page, and take the draft grade and the evidence to calibration.

Are these examples from real reviews?

No. The people, roles, dates and figures were written for this page as fictional, illustrative examples. The rating labels and definitions are from the playbook’s five-level scale. The study figures in the “Comments to avoid” section are from Kieran Snyder’s Fortune article of 26 August 2014, as checked 13 Nov 2026.

Can I use AI to draft performance review comments?

Yes, as a starting point. In Revolut People, the documented AI assistance can draft feedback from recorded 1:1s and can check feedback you write, but the product docs (as checked 13 Nov 2026) say suggestions may be inaccurate or incomplete and should always be reviewed before you submit. The docs do not say the AI writes the review.