Most performance reviews fail before a single word is written. A manager sits down in December to sum up eleven months of someone’s work and can genuinely recall about three weeks of it, mostly the recent part. The result is a document full of adjectives like “strong,” “reliable,” and “needs to be more proactive” that tell the employee almost nothing they can act on. It shows in the data: only 14% of employees strongly agree that the reviews they receive inspire them to improve, according to Gallup, and just 26% think their reviews are accurate. That isn’t a phrasing problem you can fix with a better thesaurus; it’s a measurement problem. The difference between a useless review and a useful one comes down to what you measured, how you phrased it, and whether you wrote from memory or from notes. This guide leads with the biases that distort ratings, then gives you a phrase bank organized by competency and rating level, a section on writing your own self-evaluation, and the feedback models that hold it together.

The five biases that quietly rewrite your ratings

Before any example phrases, it helps to know what you’re fighting. Decades of industrial-organizational psychology have documented predictable ways human raters distort scores. Five show up in nearly every review cycle, and naming them is the first step to writing around them.

Recency bias. People weight recent events far more heavily than earlier ones. If someone shipped a great project in October, that colors a review meant to cover the whole year, while the quiet save they made in February disappears. According to SHRM-referenced rater-error research, recency and central-tendency errors affect a large share of annual appraisals. The fix is not willpower; it’s a record you can look back on.

Halo and horns effect. One strong trait makes everything else look strong (halo), or one weakness drags down the whole picture (horns). A person who presents well in meetings gets rated high on analysis they never actually did; a person who missed one deadline gets marked down on collaboration that was excellent. The University of Tennessee’s rater-error guide and Dartmouth’s list of common rater errors both name this as one of the most common distortions.

Central tendency. Rating everyone in the safe middle of the scale to avoid hard conversations. It feels fair but erases the difference between your strongest and weakest performers, which is exactly the information the review exists to surface.

Leniency (and its opposite, severity). Grading everyone high to avoid conflict, or holding everyone to an impossible bar. Leniency is the more common failure; it makes the scale meaningless, because a “4 out of 5” signals nothing when nobody scores below one.

Idiosyncratic rater effect. This is the deep one that sits underneath the other four. When Deloitte audited its own appraisal data, it concluded that ratings said more about the person doing the rating than the person being rated. The finding traces to Scullen, Mount, and Goff’s 2000 study in the Journal of Applied Psychology, which decomposed the variance in performance ratings across thousands of managers and found that the rater’s own idiosyncratic tendencies accounted for 62% and 53% of the variance in two large datasets — more than the ratee’s actual performance did. Deloitte cited exactly this pattern when it tore down its rating system. You can’t eliminate the effect, but specific, behavior-based language tied to dated examples reduces how much of your own bias leaks into someone else’s evaluation. Vague adjectives are pure rater; a documented behavior is at least partly ratee.

Why annual reviews got reinvented

Frustration with traditional reviews hit a tipping point around 2015, and two companies became the reference cases. In April 2015, Marcus Buckingham and Ashley Goodall published “Reinventing Performance Management” in Harvard Business Review, describing how Deloitte tore down its own system. Two facts drove the redesign. A public survey found that most executives believed their existing performance management drove neither engagement nor high performance, and the firm counted the hours and discovered it was spending close to two million a year on reviews and ratings. Deloitte replaced cascading annual objectives and 360 tools with frequent check-ins and a short set of forward-looking questions a team leader answers about each person, including whether they would always want that person on their team.

Adobe made a similar move earlier. As What Matters documented, Adobe’s then-HR leader Donna Morris scrapped the annual review in 2012 in favor of “Check-in,” a lightweight model built on regular feedback, quarterly goals, and career conversations. Forbes reported that Adobe had been spending roughly 80,000 manager hours a year on the old process, and that voluntary attrition dropped after the switch, partly because the yearly ranking exercise had triggered a predictable spike in departures.

The common thread isn’t “reviews are bad.” It’s that a once-a-year, memory-based, backward-looking rating is the weakest possible format — the one most exposed to every bias above. Even if your organization still requires a formal annual review, you can write yours from an ongoing record instead of a year-end scramble.

Write feedback with SBI, not the sandwich

The single most useful phrasing tool is the Situation-Behavior-Impact model from the Center for Creative Leadership. It structures any piece of feedback into three parts:

• Situation — when and where it happened. (“In the Q3 client kickoff on July 9…”)
• Behavior — the specific, observable action, no interpretation. (“…you paused the agenda to summarize the client’s three concerns back to them…”)
• Impact — the effect on people or outcomes. (“…which surfaced a scope gap before we’d committed to a timeline.”)

SBI works because it removes judgment and mind-reading. “You’re a great communicator” is a label the person can’t act on and might dispute; “when you summarized the client’s concerns back to them, it surfaced a scope gap” is a fact they can repeat on purpose. CCL also offers an extension, SBI-I, that adds a question about intent to explore the gap between what someone meant and how it landed.

Now the thing to stop doing: the feedback sandwich — burying criticism between two compliments. It feels kind, but it reliably backfires. As reporting on Adam Grant’s work notes, people learn to wait for the “but,” and the praise gets discounted as setup while the real message gets lost between the slices. State the behavior and its impact plainly, then talk about what changes.

A phrase bank by competency and rating level

Here is the part you likely came for — but it’s built to teach, not just to copy. Most phrase lists are flat: a pile of “communication” lines with no sense of when each applies. This one is two-dimensional. Each row is a competency; each column is a point on a common three-band scale (needs improvement / meets / exceeds). That structure forces the distinction the rating scale exists to capture, and it fights central tendency directly: if you can’t tell your “meets” phrasing from your “exceeds” phrasing, you don’t yet have a rating. Use these as starting points, then attach a real, dated example to each — a phrase without evidence is still just an adjective.

Competency	Needs improvement	Meets expectations	Exceeds expectations
Communication	Written updates often omit context, so teammates ask follow-up questions before they can act. Aim to state the decision, the reason, and the next step in each update.	Keeps stakeholders informed with clear, timely updates and asks clarifying questions before starting ambiguous work.	Adapts the message to the audience — tighter for executives, fuller for new hires — and surfaces risks early enough to change the plan.
Quality of work	Deliverables frequently require rework for errors that a self-review would catch. Building in a checking pass before handoff would raise reliability.	Produces accurate, complete work that meets the agreed standard and rarely needs correction.	Sets the quality bar others follow; catches edge cases upstream and leaves work documented well enough for anyone to pick up.
Teamwork	Tends to work in isolation and shares progress late, which leaves collaborators little time to respond.	Collaborates dependably, shares information openly, and follows through on commitments to teammates.	Actively unblocks others, volunteers for the unglamorous shared work, and strengthens how the whole team operates, not just their own output.
Initiative	Waits for explicit direction on tasks that fall within their role and rarely raises problems before being asked.	Takes ownership of assigned work and flags issues without prompting.	Identifies problems no one assigned them, proposes a solution with tradeoffs, and follows through to a result without needing to be managed.
Problem-solving	Escalates blockers without a proposed option, so the manager owns the diagnosis each time. Bringing one attempted fix would move things faster.	Works through routine problems independently and knows when to escalate the genuinely ambiguous ones.	Reframes messy, ill-defined problems into a workable plan and pressure-tests it against edge cases before committing the team.
Leadership	Gives direction but little context, so the team executes without understanding priorities. Delegating the “why,” not just the “what,” would build more autonomy.	Sets clear expectations, gives regular feedback, and supports the team in hitting its goals.	Develops people measurably — others are promoted or take on more under their coaching — and makes decisions that hold up under scrutiny.

Why this phrasing works. Notice what the “needs improvement” column does not do: it never says “poor communicator” or “not a team player.” Every cell there names a behavior and then a next step, because a review the employee can’t act on is just a verdict. Notice, too, that the “exceeds” column always describes impact on other people or the system — unblocking others, setting the bar, developing people — not just doing their own job well. That’s the real line between “meets” and “exceeds,” and phrasing it that way stops you from inflating a strong individual contributor into a leadership rating they didn’t earn. The middle column is deliberately plain: “meets expectations” is not a criticism, and dressing it up in superlatives is how leniency creeps in.

Reword vague praise into specific statements

The fastest way to improve a draft review is to hunt for adjectives and convert each into a behavior with an impact. Vague praise feels good and teaches nothing; the employee can’t repeat “great attitude” on purpose. Here’s the conversion in action.

Vague version	Specific, behavioral version
”Great communicator."	"In the March board deck, you cut 40 slides to 12 and led with the one metric leadership cared about, which kept the meeting to its 30-minute slot."
"Team player."	"When Priya was out for two weeks, you picked up her two active tickets and kept the client updated without being asked, so nothing slipped."
"Needs to be more proactive."	"In the last two projects, blockers surfaced only at the status meeting. Flagging them in the channel the day you hit them would give us time to reroute."
"Strong attention to detail."	"Your QA pass on the billing release caught a rounding error that would have overcharged about 200 accounts."
"Could improve time management."	"Three of the last five deliverables came in after the agreed date. Let’s set interim checkpoints so we can adjust scope before the deadline, not after.”

The pattern is always the same: name the situation, describe the observable behavior, and state what it caused — which gives the person something concrete to keep doing or change.

Self-evaluation examples: writing about your own work

Managers aren’t the only ones who freeze in front of a review form. If your company runs self-evaluations, you face the same recency trap plus a harder one: how to describe your own contributions without overselling (which reads as spin) or underselling (which quietly lowers your rating, because your manager may anchor on your modest framing). The idiosyncratic rater effect cuts here too — the Scullen, Mount, and Goff study measured “self” as one of the raters, and self-ratings carry their own slant. The antidote is the same one managers use: write evidence, not adjectives.

The move that fixes almost every weak self-review is to replace claims about yourself with statements about what happened. “I’m a strong collaborator” is a self-assessment your manager takes on faith; “I ran the weekly cross-team sync and unblocked the design handoff twice when it stalled” is a fact they can verify and credit. Lead with the behavior and the result; let the adjective be their conclusion, not your assertion.

Weak self-review line	Stronger, evidence-based version
”I contributed a lot to the team this year."	"I rebuilt the onboarding checklist, which cut new-hire ramp time from three weeks to ten days based on our tracking."
"I’m good under pressure."	"During the October outage I owned the customer comms, sending four status updates over six hours, which cut inbound support tickets roughly in half versus the prior incident."
"I always hit my deadlines."	"I delivered 11 of 12 sprint commitments on time; the one that slipped, I flagged a week early and renegotiated scope with the PM."
"I could have done better on documentation."	"I shipped features faster than I documented them, which created two support escalations. Next quarter I’m writing the runbook as part of the definition of done.”

That last row matters as much as the wins. A credible self-evaluation names a real gap and what you’ll do about it — that reads as self-awareness, not weakness, and beats a flawless self-portrait your manager privately discounts. The reliable structure: three to five accomplishments written as behavior-plus-impact, one honest development area with a next step, and numbers wherever you have them. Round figures (“about 200 accounts”) are fine; vague ones (“significantly,” “a lot”) are not.

Anchor the review to goals, not personality

Reviews get sharper when they’re tied to goals the person actually agreed to, which is why SMART goals and the review cycle belong together. If someone’s goal was “reduce average support response time from 8 hours to 4 by Q3,” the review writes itself: you either hit the number or you didn’t. Vague goals produce vague reviews; measurable goals take the idiosyncratic rater effect out of at least part of the score, because a number isn’t subject to whether the rater likes you.

This is the honest limit of any phrase bank, including this one. Example phrases are scaffolding; the value comes from the specific, dated instance you attach to each — and those only exist if someone wrote them down when they happened.

How Carly keeps the record so you’re not writing from memory

Recency bias wins when managers reconstruct a year from memory, which is exactly what happens the night before reviews are due. Carly is an AI executive assistant that works across your email, calendar, tasks, and CRM, and it can schedule the review cycle and your recurring one-on-ones, then keep a running thread of notes and prior meeting summaries for each person on your team. When it’s time to write, you open a record that spans the whole period — the project in February, the client save in June, the missed deadline in September — instead of just the last three weeks. Carly organizes the inputs and surfaces the examples; you still write the review and make the judgment calls. It starts at $35/month, and you can set it up here.

FAQ

How do I write a performance review that isn’t generic? Attach a specific, dated example to every rating. Instead of “strong communicator,” write the situation, the observable behavior, and its impact — the SBI structure from the Center for Creative Leadership. If you can’t name a real instance behind a phrase, it’s an impression, not an evaluation, and you should either find the evidence or cut the line.

Should I still use the feedback sandwich? No. Burying criticism between two compliments trains people to discount the praise and wait for the “but,” and the real message often gets lost. As Adam Grant’s work suggests, it’s better to state the behavior and its impact directly, then discuss what changes. Direct is kinder than disguised.

What are the most common biases in performance reviews? Recency bias (over-weighting recent events), the halo and horns effect (one trait coloring everything), central tendency (rating everyone in the safe middle), and leniency (rating everyone high). Underneath them sits the idiosyncratic rater effect — Scullen, Mount, and Goff found the rater’s own tendencies accounted for over half the variance in ratings, more than the employee’s actual performance. Behavior-based language and a running record of examples reduce all of them.

How do I write a self-evaluation without over- or under-selling? Write evidence, not adjectives. Replace “I’m a strong collaborator” with what happened — “I ran the weekly cross-team sync and unblocked the design handoff twice.” Lead with three to five accomplishments as behavior-plus-impact with real numbers, name one honest development area with a concrete next step, and let your manager draw the adjective. Overselling reads as spin; underselling quietly lowers your score because managers anchor on your framing.

Why did companies like Adobe and Deloitte drop annual reviews? Both found the annual, memory-based review expensive and ineffective. Adobe was spending roughly 80,000 manager hours a year and saw attrition spike after ranking season, so it switched to lightweight “Check-ins.” Deloitte counted close to two million hours a year and replaced ratings with frequent conversations and a few forward-looking questions. The lesson isn’t to abandon reviews but to make feedback frequent, specific, and documented.

How often should performance conversations happen? More often than once a year. The reinvention documented by Buckingham and Goodall and by Adobe moved to frequent check-ins — often monthly or after each project — because feedback given close to the event is more accurate than a year-end summary. Scheduling recurring one-on-ones is the simplest way to make that happen.

Related: how to run one-on-one meetings, SMART goals, 30-60-90 day plan, executive time management secrets, meeting agenda.

Related Articles

Guides

How to Run One-on-One Meetings That Actually Matter

Jul 3, 2026

Guides

How to Run an All-Hands Meeting Worth People's Time

Jul 3, 2026

Guides

How to Delegate Tasks Effectively (Without Losing Control)

Jul 3, 2026