As of 13 August 2026, AI can analyse feedback from new starters.

What to actually do

1. Use a tool built for this

   The route this page recommends

2. Do it yourself

   Second choice

   A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open the survey, interview notes or feedback document and remove names, email addresses, employee numbers and details that could identify a person who is not needed for the analysis.
   2. Gather the exact questions, response dates or period, team or location labels, response count and the onboarding changes you want the analysis to inform.
   3. Paste the context and anonymised feedback into the prompt, keeping each response or answer separated and retaining any labels needed for a fair comparison.
   4. Ask the chatbot to produce themes, evidence, cautious interpretations, suggested improvements and unresolved questions using the supplied prompt.
   5. Compare every reported theme, count and quotation with the original feedback, deleting any claim that cannot be traced to a source response.
   6. Ask an HR or management colleague who understands the onboarding context to challenge the interpretation, check for missing minority views and decide which actions to take.
   7. Share only the checked findings and agreed actions with the people responsible for onboarding, keeping confidential individual concerns in the appropriate HR channel.

   Prompt

   Analyse the anonymised feedback below from new starters about our onboarding process.
   
   Context:
   - Organisation or team: [brief description]
   - Onboarding period covered: [dates or month range]
   - Number of responses: [number]
   - Questions asked: [paste the questions]
   - Purpose of the analysis: [what decision or improvement you are considering]
   
   Feedback:
   [paste anonymised responses]
   
   Produce:
   1. A short summary of the main findings.
   2. The recurring themes, with the number of responses supporting each theme only where the source allows a reliable count.
   3. Separate positive feedback, problems and suggestions.
   4. Representative short quotations, labelled by response or question rather than by a person's name.
   5. Any differences between teams, locations or groups only if the source contains enough evidence, and do not identify individuals.
   6. A list of practical onboarding improvements, linked to the evidence.
   7. Questions or uncertainties that need a human to investigate.
   
   Do not invent facts, quotes, counts or causes. Do not diagnose people or infer protected characteristics, performance, personality or intent. Distinguish clearly between what the feedback says, your cautious interpretation and your recommendations. Flag comments that may require a confidential HR follow-up instead of including them in a general report. Keep the tone neutral and suitable for sharing with an HR or management colleague.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

3. Hand it to a person

   The distant third

   A person who owns the outcome does this end to end, worth it when the failure is dear.