AI Sunday
EN

Annual review: ask Copilot for an evidence file

4–5 minutes

Reconstructing a year of work from memory for your annual review is long and imprecise. A feedback report published by Duke University shows another approach: let Microsoft Copilot collect the facts from your emails, meetings and Teams notes, then write and judge for yourself.

The essentials in 30 seconds in our Short:

Watch the Short on YouTube

What’s happening

On April 17, 2026, Stephen Toback of the Duke Digital Media Community described how he prepared his annual review with Microsoft Copilot. His prompt asks the assistant to go through his emails, meetings and Teams notes, specifying the period, his mid-year job change and the projects to cover. The form’s questions are then asked one by one, with a “brief but complete” answer.

The author insists: the tool plays the role of archivist and writing aid, it doesn’t do the review for you. He also states that he doesn’t recommend this method as a standard for his department: the decision to use Copilot for performance management rests with management and HR.

Key points to remember

This topic doesn’t rest on figures, but on a four-point method drawn from the feedback report:

  • A precise period: so the tool doesn’t bring up achievements from another year.
  • The job context: a role change, flagged in the prompt, helps sort “before” and “after”.
  • Anchor points: projects and tools named explicitly, to keep out the noise.
  • An intermediate Word draft: the author doesn’t paste the result directly into the official HR tool, first to proofread tone and accuracy, then because edits entered in this web portal were not always saved.

Limits to qualify

This is an individual account, presented by its author as a proof of concept, not a study. Copilot only sees what the logged-in person already has access to, as Microsoft community discussions remind us: decisions made verbally or outside transcribed meetings escape it. These same discussions advise treating the summary as a first draft, because the tool can lack nuance or give too much weight to a single piece of feedback.

Another point noted by Duke: when the author asked Copilot to write a blog post about this experience, the tool returned an error, with no explanation. He worked around the problem with a screenshot handed to another assistant. The lesson is simple: plan a fallback.

What it changes at work

For the employee, the gain is mainly collection: finding dates, deliverables and exchanges without digging through ten mailbox folders. What remains is the work that has value: choosing what matters, placing it in the team’s strategy, and standing behind the analysis in front of your manager.

On the company side, the question is about rules. Institutions like Ithaca College regulate usage: Copilot connected to an institutional account, mandatory review of results, an assistant dedicated to self-evaluations backed by HR documents. Before using it, check your company’s policy on internal data and on AI in HR processes.

Our take

This is a good use of AI: it saves time on memory and archiving, and the human keeps the judgment. The risk is mistaking an evidence file for a complete review: it reflects what left a written trace, not everything that mattered. To be used with proofreading, and in line with your company’s rules.

A prompt to adapt

Here is a generic version, inspired by Duke’s method and to be adapted to your context:

I’m preparing my self-evaluation for my annual performance review. I need your help to build a factual, structured file from my work data.

Go through my emails, my Teams meetings, my OneDrive/SharePoint files, my documents and my conversations from the last 12 months.

I’m not asking you for a grade or a judgment of me as a person. I’m asking for a neutral summary of what my work traces show, so that I can defend my record with evidence.

Give me:

  1. Key achievements: the projects, deliverables and decisions I contributed to, with dates and observable results.
  2. Impact and collaboration: the teams and people I worked with, the situations where I helped, unblocked or led a topic.
  3. Demonstrated skills: the skills that stand out from my work, each illustrated by a concrete example.
  4. Feedback received: the feedback, thanks or criticism present in my messages, quoted as is with their source.
  5. Areas for improvement: points that seem recurring (delays, topics left open, difficulties flagged), worded constructively.
  6. Trend analysis: how my workload and priorities evolved over the period.
  7. Self-evaluation draft: a 300-word text in the first person, ready to adapt, backed by the facts above.
  8. Describe my observable work behaviors: collaboration styles, how I handle problems, types of responsibilities taken, recurring practices and factually documented areas for improvement.

For each point, give your source (document type, date). If the data is insufficient on a point, say so rather than extrapolating. Stay factual and descriptive.

Sources

À découvrir aussi sur AI Sunday

Portrait de Daniel, fondateur d'AI Sunday

Écrit par

Daniel

Professionnel des achats, j'utilise l'IA au quotidien dans mon travail et je raconte ici ce qui marche vraiment, et ce qui coince. En savoir plus