The form opens in the last week of November. Five text boxes, a drop-down for an overall rating, and a deadline that lands in the busiest fortnight of the year. Most people fill it in the night before: twenty minutes, a handful of adjectives, and “meets expectations” in the drop-down because claiming more feels presumptuous. Then they submit it and forget it.

That is a costly way to treat the one document in the review process that you write yourself. The research on self-evaluations is narrower than the advice industry suggests, and in places it cuts against the employee. But it is clear on two points. The rating you propose pulls on the rating you receive. And a claim your manager cannot check is a claim your manager has good reason to discount. A self-review that holds up is built around both facts.

The number you give yourself anchors the number you get

The best evidence comes from an accident. At a multinational financial services firm studied across four review cycles from 2015 to 2018, a software glitch in 2016 meant managers could not see employees’ self-evaluations before rating them. In that cycle, average manager ratings fell and tracked employees’ self-ratings less closely [1]. In the other years, with the self-rating in front of them, managers stayed nearer to it.

A laboratory experiment had pointed the same way decades earlier: 203 participants acting as supervisors rated the same clerical work significantly higher when it arrived with a favorable self-assessment than with an unfavorable one [2]. Psychologists call this anchoring. A number placed in front of a judge before the judgment is made becomes the point the judgment starts from.

Two cautions belong here. The field study covers one firm, and its public summary gives no effect size. It also concerns the self-rating, the number, and does not test whether the written narrative changes anything. Still, the practical reading is hard to avoid. If your process asks you to rate yourself, the modest choice is not a neutral one.

Nor is modesty evenly distributed. In the same study, women of color gave themselves the lowest self-ratings [3]. The lead author, Iris Bohnet, draws the conclusion for employers: where managers have no rating history to go on, it may help not to share self-evaluations with them until they have formed their own view [4]. That she recommends removing the anchor tells you how seriously the researchers take its pull. Until your employer does remove it, the anchor is yours to set.

Why your manager is right to be skeptical

None of this means a high self-rating will simply be accepted. A meta-analysis of more than a hundred samples found that self-ratings and supervisor ratings of job performance correlate at just .22, and that in Western samples people rate themselves about a third of a standard deviation more leniently than their supervisors do [5]. Managers know this from experience even if they have never seen the figure. An unsupported “exceeds” reads as the usual inflation.

Self-promotion does not reliably help either. In a study of 173 supervisor–subordinate pairs, employees who leaned heavily on impression tactics, self-promotion among them, were rated more favorably by their supervisors when they were politically skilled than when they were not [6]. Asserting your own excellence is a high-variance move.

There is also a second anchor competing with yours. When the self-evaluations in the financial-services study were hidden, gender and race gaps in ratings did not close, because managers anchored on the previous year’s ratings instead [3]. Last year’s number is already in your manager’s head. If you are asking to be seen differently from a year ago, you are arguing against a prior, and that takes more than adjectives.

What makes a claim hold: someone can check it

Forty years ago, selection researchers faced the same problem from the other side: how to score what people write about their own achievements without rewarding the best self-publicist. The method they developed, the accomplishment record, is still described in US federal hiring guidance. Candidates write specific accomplishment statements, trained raters score them against benchmarks, and each statement usually has to name someone who can verify it [7]. The design principle transfers directly. A statement is worth more when it is specific enough to be compared with a standard and a named person could confirm it. Not all valuable work fits that test: mentoring, a conflict defused, anything confidential. Say what it was and who saw it, and keep the checkable outcomes at the front.

Records have a second effect worth understanding before you rely on one. In field experiments at two organizations, raters who kept structured diaries of performance recalled more, and their ratings became less elevated and better at distinguishing between people [8]. That study concerned the manager’s record, not the employee’s, and nobody has tested whether an employee-kept record changes ratings. But the direction is instructive. Evidence sharpens a judgment; it does not inflate one. A record makes a rating easier to justify, and that works in both directions.

The person reading your self-review is short of time. Gallup finds that 97% of US managers carry individual-contributor work of their own, at a median of 40% of their time [9], and that 56% of employees formally review their performance goals with their manager once a year or less [10]. For many people the self-review is the only structured account of the year their manager will read before deciding. Which is why capturing outcomes as they happen matters: in late November you can only write down what you can still retrieve.

Write the arc as well as the list

Raters do not simply average a year. In an experiment where participants evaluated 26 weeks of an employee’s results, the trend across the period had a strong effect on the overall rating, separately from the average level [11]. A study of professional basketball contracts found the same pattern in pay: the trend in a player’s performance predicted changes in compensation even after accounting for how the player went on to perform [12]. Athletes with fully measured output are a long way from knowledge workers, so treat that as supporting colour. The point that carries over is that the shape of the year registers alongside its total.

Left unstated, the shape will be inferred from whatever your manager remembers best, which is usually the most recent quarter. So state it. Say where you started the year, what changed, and where you are now: a scope that grew in the spring, a problem inherited in June and closed by October, a capability you did not have in January. A list of achievements tells your manager what happened. An arc tells them what to conclude from it.

What a rating band is worth

In one dataset of merit outcomes for 46,000 US employees in early 2025, people rated “exceeds expectations” received a median raise of 5%, and those rated “meets expectations” a median of 3.5%. Among those rated below expectations, only one in ten received any raise [13]. The data come from a compensation-software vendor and lean toward technology companies, but the scale is useful: one rating band is worth roughly a point and a half of base pay, compounding from then on.

Almost every employer makes that connection. In Mercer’s 2025 survey of more than 500 US employers, only 9% said they do not use performance ratings to determine merit increases, and just over half hold formal performance discussions only once a year [14]. Promotion is tied to ratings less tightly: in an earlier global Mercer study, 43% of companies said they link the two [15]. A strong rating is the entry ticket to a promotion discussion, and that decision is made by a different process.

Employees do not much trust any of it. Only 22% strongly agree that their review process is fair and transparent [10]. That is a reason to put more care into the one input you control, not less.

On letting an assistant write it

Nobody has measured how many employees now draft their self-reviews with an AI assistant, and no study tests how managers receive them. The nearest evidence concerns messages travelling the other way. In a study of 1,100 professionals, 83% of employees judged a supervisor’s message sincere when it was written with little AI assistance, and only 40% to 52% did when the assistance was heavy. The same professionals were lenient about their own use of AI and skeptical of other people’s [16].

Applying that to self-reviews is an inference, but a reasonable one: your reader may judge polished, generic prose more harshly than you would. The useful division of labor is to let an assistant organize, cut and check for gaps, and to supply the specifics yourself. An assistant that has never seen your year can only produce adjectives, and adjectives are the part of a self-review a manager has learned to discount.

A self-review that holds up

Put together, the evidence suggests a document with six parts. None of them takes long if the material exists.

  • Set the anchor deliberately. Choose the rating you can defend, and write one sentence beside it that says why. Do not default downward out of politeness.
  • Lead with three to five outcomes. For each: what changed, a number, when, and who can confirm it. Where your level has written expectations, say which one the outcome meets.
  • State the arc. Two or three sentences on how your scope and results moved across the year.
  • Address the weak point first. Name one thing that went less well and what you changed. It costs little, and it makes the rest more credible.
  • Write for the person who has to repeat it. Your manager may need to defend your rating to peers. Give them sentences they can use without editing.
  • Close with what you want next. The scope, the level, or the problem you are asking to be given.

Two practical notes. Review cycles are shortening in some organizations: in one vendor survey of HR professionals, a third now run reviews quarterly and 12% monthly [17]. A shorter cycle makes each self-review smaller and makes a running record more valuable. And many processes include no self-assessment at all. In a 2021 survey of 344 employers, only about half did [18]. If yours is one of them, write the same one-page account anyway and send it to your manager before ratings are due. The anchor works whether or not there is a form for it.

Persona Map is built to hold the material this depends on: a private workspace where outcomes and feedback are logged as they happen, with an AI advisor that works from that record. But the principle does not need a tool. A self-review is the one place in the process where you decide what your manager starts from. Treat it as evidence, and write it for the person who will have to stand behind your rating in a room you are not in.

Sources

  1. Harvard Kennedy School. “Self-Ratings and Bias in Performance Reviews.” Summary of research by Iris Bohnet, Oliver Hauser & Ariella Kristal, 2025. Link
  2. Shore, Adams & Tashchian. Laboratory study of subordinate self-assessments and supervisor performance ratings, 1998. Link
  3. Iris Bohnet, Oliver Hauser & Ariella Kristal. “Can Gender and Race Dynamics in Performance Appraisals Be Disrupted? The Case of Social Influence.” Journal of Economic Behavior & Organization, 2025. Link
  4. Harvard Gazette. “Why Self-Appraisals May Not Be Best Way to Judge Job Performance.” Interview with Iris Bohnet, December 10, 2025. Link
  5. Heike Heidemeier & Klaus Moser. Meta-analysis of self–other agreement in job performance ratings. Journal of Applied Psychology 94(2), 2009. Link
  6. Kenneth Harris, K. Michele Kacmar, Suzanne Zivnuska & Jason Shaw. “The Impact of Political Skill on Impression Management Effectiveness.” Journal of Applied Psychology 92(1), 2007. Link
  7. US Office of Personnel Management. “Accomplishment Records.” Assessment and selection guidance. Link
  8. Angelo DeNisi & Lawrence Peters. Field experiments on structured diaries and recall in performance appraisal. Journal of Applied Psychology 81(6), 1996. Link
  9. Gallup. Analysis of span of control and team size, 2025 data. January 2026. Link
  10. Gallup. Ben Wigert & Heather Barrett. Survey of US employees and chief human resources officers on performance management. May 6, 2024. Link
  11. Jochen Reb & Russell Cropanzano. “Evaluating Dynamic Performance: The Influence of Salient Gestalt Characteristics on Performance Ratings.” Journal of Applied Psychology 92(2), 2007. Link
  12. Christopher Barnes, Jochen Reb & Dionysius Ang. Study of performance trend and compensation among National Basketball Association players. Journal of Applied Psychology, 2012. Link
  13. Pave. “How Big of a Raise Can You Expect in 2026?” October 27, 2025. Link
  14. Mercer. “Are Employers Shifting How They Think About Employee Performance?” QuickPulse survey, July 22, 2025. Link
  15. Mercer. “Performance Transformation in the Future of Work.” Global Performance Management Study, 2019. Link
  16. University of Florida News. Report on research by Peter Cardon & Anthony Coman, International Journal of Business Communication. August 6, 2025. Link
  17. Lattice. “State of People Strategy 2027.” 2026. Link
  18. HR Dive. “The Annual Performance Review Isn’t Dead Yet.” Report on an XpertHR survey, August 26, 2021. Link