There is usually a week in Q1 when performance reviews land on your desk all at once. You have rough notes from one-to-ones, a few email threads, and a half-clear memory of who delivered what. What you don’t have is time. Each review needs to be specific, honest, and legally sound, and it needs to sound like you wrote it rather than something produced by a committee. Prompt templates for AI tools can help with all of that, provided you use them in the right order.
What are prompt templates for performance reviews?
A prompt template is a structured instruction you give to an AI tool, such as ChatGPT or Claude, that turns your rough notes or voice memo into a draft performance review. It defines the structure the review must follow, constrains what the AI can invent, and tells it to flag any gaps rather than guess. The output sounds like you, built on what you actually told it.
Platforms like Lattice and Rippling now publish prompt libraries specifically for performance reviews, which signals that AI-assisted drafting has moved from experiment to accepted practice in HR workflows. The prompts generally work in two passes: a structuring pass that organises your raw material into the sections your review template requires, and a fairness pass that checks the resulting language for bias or unsupported statements before you sign it off.
The structure is what makes a template different from simply pasting your notes into an AI and asking it to write you a review. A well-designed prompt tells the AI which sections the review should include, what the rating scale means, and what good development feedback looks like in your context. Without those constraints, the output tends to be polished but impersonal.
Why does this matter for your business?
Writing twelve performance reviews in a week is a significant draw on your time. A 2023 Gartner survey found that 38% of HR leaders were already exploring generative AI for performance management, with time saving named as a primary driver. Lattice claims managers can go from hours to minutes using its prompt-based review tools. For an owner-managed business with a small team, that kind of reduction is worth taking seriously.
The consistency argument is equally practical. When every manager across a team uses the same prompt structure and the same template sections, reviews become comparable. That matters if you ever need to defend a pay or performance decision. Reviews written without a shared structure often reflect the personality of the manager more than the performance of the person being reviewed.
There is also the blank-page problem. Many managers find that starting is the hardest part. When you give an AI tool your one-to-one notes and a voice memo of your honest thoughts, and ask it to organise rather than invent, you get a working starting point rather than a cursor blinking at nothing.
Where will you actually use these prompts?
The prompts work inside any AI tool you already have access to: ChatGPT, Claude, or an HR platform with AI features built in. You use them after gathering your material but before writing a single sentence of the review itself. The sequence matters as much as the prompts: collect your evidence first, form your view second, then use AI to structure the draft.
Start by pulling together your source material: the employee’s objectives for the period, your one-to-one notes, any client feedback, and any informal comments you have already shared with them during the year. Keep these in your HR system or a secure document rather than in an AI tool at this stage.
Next, record a voice memo or write unstructured notes where you talk through what went well, what did not, how they performed against each objective, and where you would like them to focus next. Coach Helen Beedon, whose 30-minute performance review framework has been widely shared among managers, recommends dictating your thoughts and using AI to structure them afterwards. Doing it in this order reduces the tendency to soften feedback as you write, because you have not started writing yet.
Then paste your transcript or notes into your structuring prompt. Tell the AI to keep your natural language, follow your review template, use only the facts you have provided, and flag anything missing with a placeholder rather than inventing it. Finish with a second-pass prompt that asks the AI to identify any feedback describing character rather than behaviour, any criticism without a concrete example, and any wording that might read as discriminatory.
When should you use AI for reviews, and when shouldn’t you?
AI works well for structuring and drafting language around feedback you have already formed. It falls short when you have not decided your view yet, when you are hoping it will assign a rating, or when your notes are too thin for it to do anything useful. Rippling, which publishes specific prompt guidance for performance reviews, warns explicitly that vague instructions can lead generative AI to reinforce biased language or assumptions.
There are situations where this workflow adds more overhead than it removes. For a team of three or four where appraisals are essentially conversations and everyone knows everyone, a structured AI drafting process may not be proportionate. Scale matters.
The Amazon recruiting tool case is a useful reference point. Its historical training data was skewed male, so it consistently rated female candidates lower, and the case is now widely cited by regulators as a warning about AI trained on biased historical data. Performance reviews face the same risk if prompts ask for vague personality assessments instead of behaviour-based, evidence-linked descriptions. The prompt design is the control.
There is also a category where AI creates clear risk: pasting identifiable staff information into consumer AI services without a data processing agreement. In 2023, Samsung banned staff from using public AI tools after employees shared confidential corporate information with ChatGPT. The same risk applies when a manager pastes someone’s full employment history into a consumer chatbot.
What else should you know before you start?
Using AI in performance reviews is personal data processing under UK GDPR. You need a lawful basis, staff need to know you are using AI in this way, and the AI cannot be the decision-maker on pay, promotion, or dismissal outcomes. The ICO has published specific guidance on AI in employment practices, and the CIPD describes the appropriate use as assistive rather than autonomous for decisions affecting people’s livelihoods.
For a typical owner-managed business, a lightweight governance approach is proportionate. Update your privacy notice to mention AI tools in HR drafting, and keep a brief record of which tools you use for this purpose. If a review materially influences a pay or promotion decision, carry out a Data Protection Impact Assessment. The NCSC advises using enterprise or private AI services rather than consumer tools for anything involving identifiable staff data, and checking that your contracts with those services include appropriate data protection terms.
The EU AI Act, finalised in 2024, classifies AI systems that assist in employment decisions on promotion, task allocation, or dismissal as high-risk. UK businesses are not directly in scope unless they have EU-based staff or use EU-hosted HR platforms, but the regulatory direction is towards more accountability for AI-influenced decisions, not less.
The practical boundary is a sensible one to work within. The AI drafts, you read and edit, and every decision that affects someone’s pay or position is made by a person who can stand behind it. That boundary is also exactly where the ICO and CIPD are both pointing.
