Most quality-of-work review comments say nothing. “Maintains a high standard of output.” “Work occasionally falls short of what’s required.” Both sentences could be pasted onto any employee in any role, in any company, in any year. Each one is a rating label with a person’s name attached to it.
That vagueness has a cost, and it lands well outside the page it was written on. Gallup’s research on feedback found that only 21% of U.S. employees strongly agree they have received meaningful feedback in the last week. Meaningful is the operative word. Most people are getting comments. Very few are getting comments with anything in them.
The cost shows up later, in a room. Vague praise is the first thing challenged in a calibration meeting, because a peer manager can dispute “high quality” in four words and you have nothing to answer with. Vague criticism is the first thing challenged in an HR appeal, for the same reason. A comment that can’t be substantiated with evidence is an opinion wearing a form.
This post gives you phrases for the quality-of-work competency specifically, organized by rating tier, for both managers writing reviews and employees writing self-evaluations. More importantly, it gives you the pattern underneath them, so you can write your own instead of hunting for one that almost fits.
What makes a quality-of-work phrase defensible
The most useful model for this is SBI, developed by the Center for Creative Leadership: Situation, Behavior, Impact. Name where it happened, name what the person actually did, name what followed. Every phrase in the banks below is SBI compressed into one written sentence.
The paste test
Before any comment goes into the form, run it through one check. If the phrase could be pasted onto any employee in any role without changing a word, it fails. That’s the whole rule. Binary, and fast enough to run on every line.
It also explains why vague criteria invite bias, a point Harvard Business Review has made about review fairness generally. A sentence with no evidence in it gets filled in by whatever the reader already believed, and what they supply is their bias.
The slot pattern
Every phrase in this post is built on the same three-slot structure:
[observable behavior] on [specific artifact], [quantified impact versus baseline]
That’s it. Three slots. Fill all three before you write the sentence, and the sentence writes itself.
- Observable behavior: what the person did, in verbs a camera could have recorded. “Was thorough” describes an impression. “Included the currency-conversion edge case” describes an action.
- Specific artifact: the named thing it happened on. The Q2 client report. The March invoicing batch. If you can’t name the artifact, you’re rating a memory rather than a result.
- Quantified impact versus baseline: the number, and the number it should be compared to. “Zero errors” means little alone. “Zero errors, against a team average of four per cycle” means something. The baseline is what turns a fact into a rating.
Watch it work on three comments that fail the paste test:
| Vague | Rewritten with the slot pattern |
|---|---|
| ”Work is generally of a high standard." | "Delivered the Q2 client report with zero factual corrections across three review rounds, down from an average of four corrections per cycle." |
| "Needs to be more careful." | "Submitted the March invoicing batch with six line-item errors that finance corrected manually, the third consecutive month with errors in the same file." |
| "Sets a high bar for the team." | "Rebuilt the reporting template to the new data-labeling standard without being asked, and four other analysts have since adopted it.” |
The rewrites run longer for one reason. Each of them carries a fact the reader can go and check.
The four dimensions of “quality of work”
“Quality of work” is four things wearing one label, which is exactly why generic phrases about it are so easy to write and so hard to defend. Break it into four dimensions that don’t overlap, and write to each one separately. That’s the same logic behind competency-based performance reviews: you rate the behaviors you can observe, one at a time.
- Accuracy (is it correct?) Error and defect rate found downstream, by a reviewer, a client, or production.
- Completeness (is it whole?) Scope covered against scope required. Edge cases, caveats, the section everyone skips.
- Consistency (is it correct every time?) Variance over the period. Rework rate. The gap between their best deliverable and their worst.
- Standards adherence (is it to standard?) Meets the agreed spec, format, or process without being reminded.
Attention to detail sits underneath all four of these, which is why it makes a poor fifth row on a rubric. It’s the skill that produces accuracy, completeness, consistency, and standards adherence, so rating it separately just gives you four versions of the same comment. Rate the four outputs. Coach the skill.
Most managers believe their feedback is more specific than it is, and the review form is the worst possible place to discover otherwise. Risely’s free constructive feedback assessment gives you a baseline on how your feedback actually lands in about five minutes.
And there’s a reason to care beyond the paperwork. A phrase is a claim, and every claim you type into the form is one you will have to make out loud, to the person it describes, in a room where they are allowed to disagree with you. That is the standard the slot pattern is built for. Writing the phrase is the easy half.
Manager phrase bank: quality of work, by rating tier
Four tiers here, where most performance review rating scales run five. The fifth usually separates “exceptional” from “strong,” and once every phrase has to name a specific observable behavior, that distinction collapses into an adverb. “Consistently exceeds” and “frequently exceeds” describe the same behavior with different amounts of confidence attached. So the tiers below are: Exceeds, Meets, Inconsistent, Below standard. What separates Exceeds from Meets is the magnitude of the evidence you can put behind it.
Each tier gives you the pattern first, then filled examples across all four dimensions. Use the pattern. The examples show you what a filled slot looks like; they are worked demonstrations, and pasting one onto someone whose quarter looked nothing like this fails the test at the top of the post.
Exceeds
Pattern: [behavior above the agreed standard] on [artifact], [impact the standard alone would not have produced].
The evidence has to carry weight the baseline can’t. Prevention, scale, or spillover onto other people’s work.
Filled:
- “Caught a pricing error in the Henderson vendor contract before signature, preventing an estimated $12,000 overcharge across the two-year term.” (Accuracy)
- “Shipped 14 consecutive client deliverables with zero factual corrections raised at legal review, against a team average of two per deliverable.” (Accuracy)
- “Flagged the currency-conversion edge case in the billing spec that neither the requirements doc nor QA had caught, removing an entire class of post-launch refund tickets.” (Completeness)
- “Documented the three failure states in the onboarding flow the original brief left out, so support had answers ready at launch instead of six weeks in.” (Completeness)
- “Held rework on client decks at one revision round or fewer across all four quarters, while the team baseline sat at three.” (Consistency)
- “Delivered the monthly close file to the same standard in every month of the period, including the two months when the team was down two people.” (Consistency)
- “Rebuilt the team reporting template to the new data-labeling standard without being asked, and four other analysts have since adopted it.” (Standards adherence)
- “Applied the updated disclosure format to all client emails within a week of it being published, before the rollout deadline was announced.” (Standards adherence)
Meets
Pattern: [behavior at the agreed standard] on [artifact], [result at or inside the agreed tolerance].
This tier gets written badly more often than any other, because “meets expectations” feels like it needs no evidence. A Meets rating with no artifact in it is indistinguishable from a manager who wasn’t paying attention.
Filled:
- “Produced the weekly sales report with no more than one correction per month, inside the agreed error tolerance.” (Accuracy)
- “Caught and corrected two data errors in the Q1 forecast before it reached leadership.” (Accuracy)
- “Completed every required section of the client proposal template, including the risk assumptions section that’s frequently left blank.” (Completeness)
- “Closed all 18 sprint tickets with reproduction steps attached, as the team standard requires.” (Completeness)
- “Delivered 11 of 12 monthly reports on the agreed format and deadline, with the single exception flagged in advance.” (Consistency)
- “Averaged two revision rounds on the monthly campaign reports across the year, matching the team baseline of two.” (Consistency)
- “Ran the documented QA checklist on every release this period, with no skipped steps found in the quarterly audit.” (Standards adherence)
Inconsistent
Pattern: [behavior at standard] on [artifact A], [behavior below standard] on [artifact B], [the condition that predicts which one you get].
That third slot is what makes this tier useful. Inconsistency is a pattern, and a pattern has a trigger: deadline pressure, unfamiliar work type, client-facing versus internal. Name the trigger and you have written a development plan. Leave it out and you have written a complaint. If delivering the message is what worries you, our guide to giving constructive feedback covers how to say things like this out loud without the conversation collapsing.
Filled:
- “Delivered the Q3 campaign brief with no corrections and the Q4 brief with six factual errors caught by the client. Accuracy tracks with how much time is left before the deadline.” (Accuracy)
- “Errors on internal documents stayed inside tolerance, while client-facing work needed corrections in five of the last eight deliverables.” (Accuracy)
- “Left the assumptions section blank on four of nine business cases this year, and finance returned each one for context before approving.” (Completeness)
- “Submitted the onboarding deck three times before it met the agreed formatting standard, requiring two rounds of manager edits.” (Completeness)
- “Rework on client-facing documents averaged three revision rounds against a team baseline of one, adding roughly six hours of reviewer time a month.” (Consistency)
- “Used the previous reporting template on three of the last five submissions after the new format was rolled out in April.” (Standards adherence)
- “Ran the QA checklist on releases when reminded and skipped it on the two releases shipped under deadline pressure, both of which needed hotfixes.” (Standards adherence)
Below standard
Pattern: [behavior below the agreed standard] on [artifact], [quantified cost], [what the standard was and when it was communicated].
That fourth slot is not optional here. A below-standard rating without a documented standard and a documented date is the single most challengeable sentence a manager can write.
Filled:
- “Submitted the March invoicing batch with six line-item errors that finance corrected manually, the third consecutive month with errors in the same file.” (Accuracy)
- “Errors in client deliverables were raised by the client rather than caught internally on four of the last six projects.” (Accuracy)
- “Delivered the vendor comparison with two of the five required vendors missing, after the scope was confirmed in writing twice.” (Completeness)
- “Skipped the pre-publish data checks the role requires on the two dashboards that later reported incorrect revenue figures to leadership.” (Completeness)
- “Rework on the weekly client status decks has run at four or more revision rounds since January, against a documented standard of one.” (Consistency)
- “Has not applied the reporting format agreed in the January team standard on any submission since it was introduced, despite two written reminders.” (Standards adherence)
- “Missed the documented sign-off step on three releases, two of which required a rollback.” (Standards adherence)
Where phrases stop and documentation begins
If you’re writing sustained below-standard comments, be clear about what you’re doing. Language for a performance improvement plan, a formal warning, or a termination file belongs in a different document with stricter requirements: dated incidents, prior notice, the specific standard, the support offered, and the consequence if nothing changes. Write it with HR. The phrases above will feed that document. They will not replace it.
Self-evaluation phrase bank: quality of work
Same specificity bar, different voice. The characteristic failure mode in a self-evaluation is effort language. “I always give 110%.” “I take great pride in my work.” “I hold myself to a high standard.” All three describe how hard someone tried, and effort has never been the thing being assessed. The work is.
Write first-person, lead with the artifact, and let the number do the arguing.
Accuracy
- “I caught and corrected two data errors in the Q1 forecast before it reached the leadership review.”
- “Across 22 client deliverables this year, one was returned with a factual correction, down from five last year.”
- “I built a pre-send verification step for pricing figures after the February quote error, and no pricing error has reached a client since.”
Completeness
- “I added the compliance caveats to the standard proposal template after legal flagged them twice, and they’ve since been included in every proposal the team ships.”
- “I raised the two undocumented edge cases in the payments spec during design review, which meant they were handled before build rather than after launch.”
- “On the annual budget model, I documented every assumption inline so finance could audit the logic without booking a call with me.”
Consistency
- “I reduced my average rework on client deliverables from three revision rounds to one by adding a self-review checklist before submission.”
- “I delivered all 12 monthly reports on format and deadline this year, including during the two months I covered Daniel’s accounts alongside my own.”
- “My deliverables under deadline pressure used to be my weakest work. This year I front-loaded the review step, and the two rushed Q4 projects went out with no corrections.”
Standards adherence
- “I adopted the new data-labeling standard in the first week it was published and updated my four recurring reports to match before the deadline.”
- “I ran the QA checklist on every release I owned this period, including the three that shipped under time pressure.”
- “I rewrote our team’s handover doc to the format the wider function uses, which cut the questions I get during my time off from several a week to none.”
One note on tone. Naming a weakness and the correction you made to it (phrase 9) reads as the strongest line in the bank, because it shows you can see your own work clearly enough to fix it without being told. Every reviewer is looking for exactly that.
Avoid the recency trap
The best-written phrase in this post is still indefensible if the evidence behind it all came from the last three weeks. Recency bias is the tendency to weight the most recent stretch of the review period far more heavily than the months before it, and performance reviews are close to a perfect environment for it. You write the review in the same week you’re living the work. Whatever shipped last, shipped loudest. Lattice’s guidance on mitigating recency bias in reviews makes the same core point: without a record, memory selects for you.
The fix is mechanical, not attitudinal. Before you finalize any tier rating:
- Pull evidence from month one, not just month three. Open your notes, the ticket system, the shared drive, whatever holds the artifacts. Find at least one example from the first third of the period for each dimension you’re rating.
- Check the spread. If every artifact you named in the review happened in the last six weeks, you haven’t rated the period. You’ve rated the quarter’s tail.
- Test the rating against the earliest evidence. If the phrase you wrote would have been false in month one, you’re rating a trend rather than a level, and a trend is a different sentence: “Accuracy improved across the period, from four corrections per cycle in Q1 to zero in Q4.”
Improvement is a defensible, useful thing to say. It just isn’t the same claim as consistency, and reviews collapse the two constantly.
Rehearse it before you have to say it
You now have the pattern, the four dimensions, the tiers, and 41 phrases to adapt. What you don’t have is the conversation, and that is the actual test of every phrase above. A written claim only counts as evidence once it survives someone challenging it to your face. The employee disagrees with the Inconsistent rating and says the Q4 brief was rushed because you moved the deadline. The peer manager in calibration says “that just sounds like they were busy.” Neither of them is reading your sentence. They’re arguing with it. A phrase you cannot defend under that pressure was never defensible on the form, whatever it looked like when you typed it.
Merlin, Risely’s AI coach, will run that conversation with you before you have it for real, push back where a real person would, and show you which of your phrases hold up under pressure and which dissolve.
Rehearse your next performance review with Merlin
Writing review comments on a different competency, like communication, teamwork, or leadership? Our full performance appraisal comments library covers those. And if productivity is the competency you’re stuck on, see the companion guide on running a performance review for productivity.