Use the Review Assistant to review AI suggestions, check flagged feedback, and ask Chat for help while writing a self or manager review. When complete, your answer reflects the edits or feedback you chose to apply.
This article applies to organizations enrolled in the AI Assisted Reviews beta.
Note: Your organization may have renamed this feature. If you don’t see “Best-Self Review®” in your navigation, check with your admin for the name used in your account.
Note: To ensure the highest level of security, 15Five is continually investing in our overall information security program, resources, and expertise. To review 15Five's security documents, and how we handle your company data, see our Trust Website.
BEFORE YOU BEGIN
- AI-assisted reviews must be enabled for the review cycle.
- You must be writing a self review or manager review.
- Built-in writing guidance to help structure responses must be enabled.
- Complete the Reflect phase before using the Review Assistant.
Actionability
Does the person walk away knowing what to do differently? Vague directions like "improve communication" evaluate low. Specific, forward-looking language tied to real situations evaluates high.
Specificity
Are claims backed by real examples — numbers, dates, actual situations? "Great communicator" evaluates low. "Led the checkout redesign across three teams, resulting in an 8% conversion improvement" evaluates high.
Balance
Is the review honest about both what is working and what is not? Effective reviews include enough recognition to be motivating and enough growth feedback to be useful. Reviews that lean too far in either direction are flagged.
Tone consistency
Does the language match the rating? A high rating with lukewarm prose sends a mixed signal. A low rating with glowing language is equally confusing. Mismatches between how someone is rated and how they are described are flagged.
Suggestions are question-aware — a strengths question is not penalized for missing forward-looking language, and a growth question is not penalized for critical feedback.
Patterns across your answers
As you move through questions, a Patterns across your answers section builds up. It catches issues that span multiple answers:
- Recency bias — references that only cover the last few weeks
- Sentiment mismatch — inconsistent tone across questions
- Redundancy — the same point repeated in multiple answers
A single instance in one answer might be incidental. When the same pattern appears across two or more answers, it escalates in severity.
Bias detection
Bias is tracked in its own row, separate from writing quality. Bias affects how a person's career gets documented — the stakes are higher than tone or structure.
The Review Assistant flags these bias patterns:
| Bias type | What it detects |
|---|---|
| Gendered language | Communal-trait adjectives disproportionately applied to women ("collaborative," "supportive") and agentic ones to men ("strategic," "technical"). Flags what is being said about a person, not pronoun use. |
| Personality-trait assessments | Character judgments ("is lazy," "born leader") that interpret motivation rather than describe behavior. |
| Affinity bias | "Culture fit" language that measures similarity to the reviewer rather than job performance. |
| Attribution bias | Framing wins as luck ("got lucky") instead of describing what the person actually did. |
| Effort over outcome | Praising effort ("tries hard," "works hard") instead of results. |
| Halo and horn effects | Overgeneralizing from an impression rather than citing evidence. "As always" and "never changes" are flags. |
| Leniency inflation | Stacking superlatives ("outstanding, exceptional, flawless") beyond what evidence supports. |
| Severity deflation | Stacking deficit labels ("mediocre, subpar, disappointing") beyond what evidence supports. |
| Recency bias | Reviews that only reflect the last few weeks instead of the full review period. |
Bias is detected across the whole review, not just within individual answers. When the same pattern appears across two or more answers, it escalates — a single instance might be incidental, but a pattern suggests an unconscious frame.
Chat assistant
The Chat tab is a thinking partner available at any point during writing. It is scoped to the question and answer you are actively working on — each question gets its own conversation.
Use the Chat to:
- Draft from sources — generate a fresh answer grounded in your Moments and context
- Refine existing prose — "make this more concise," "add more specifics," "soften the tone"
- Ask for feedback — ask how something reads without triggering a rewrite
Quick prompts
Quick-select prompts are available for common tasks:
- Draft this from my sources
- Make it more specific
- Check for bias
- Soften the tone
- What's missing?
The current draft answer is always visible at the top of the Chat tab. When the Chat updates the draft, you see both the revised answer and an explanation of what changed.
Every claim the Chat produces traces back to a specific Moment or source item. Citations link directly to the source.
Note: The "Built-in writing guidance to help structure responses" feature toggle at the cycle level controls the Suggestions, Improve writing actions, and Chat assistant. If an Admin disables this feature for a cycle, the Review Assistant panel is not available.
IF SOMETHING GOES WRONG
| Issue | Check | Fix |
|---|---|---|
| Review Assistant options are missing | Confirm Built-in writing guidance to help structure responses is enabled. | Ask a Review admin to enable it for the cycle. |
| Suggestions do not match your intent | Review the explanation attached to the suggestion. | Ignore the suggestion or edit the answer manually. |
| Chat references the wrong answer | Confirm the correct review question is selected. | Select the correct question and send the prompt again. |
| Chat output needs more evidence | Check the Moments below the answer. | Add specific evidence to the answer manually. |