thinkhard.SI

Prompt repository Hiring & people

Summarize workplace investigation interviews neutrally

You are looking into a workplace complaint and need a neutral record of what each person said, kept apart from any conclusion, for the person who will decide.

Registry no.
PPL-009
Collection
Hiring & people
Laid out for
Any chat model
Score
10 out of 10
Blanks to fill
5
Added

Hiring & peopleAny chat

The prompt

Score 10 of 10, Excellent

Use it when You are looking into a workplace complaint and need a neutral record of what each person said, kept apart from any conclusion, for the person who will decide.

  • {{decision_maker}}
  • {{advisor}}
  • {{complaint}}
  • {{interview_notes}}
  • {{policy}}
Full prompt and why it works
Prompt
Summarize the interview notes below from a workplace investigation into a neutral record for the person who will decide what happens next, {{decision_maker}}, who is advised by {{advisor}}. The summary is a working document, not a finding, so it reports what each person said without judging who is right.

Complaint as received:
"""
{{complaint}}
"""

Interview notes, one section per person, each starting with a role label such as "Complainant", "Respondent" or "Witness 1":
"""
{{interview_notes}}
"""

Policy sections that may apply:
"""
{{policy}}
"""

Format, with these headings:
1. Allegations: each allegation from the complaint as a numbered item, in neutral words.
2. Accounts: for each allegation, a table with the columns Person, What they said, Dates or documents mentioned. Use the role labels, not names.
3. Agreed facts: points on which every account agrees.
4. Disputed points: points on which accounts differ, side by side, with no view on which is true.
5. Gaps: evidence mentioned but not yet seen, people mentioned but not yet interviewed, and questions nobody was asked.
6. Policy sections that may be relevant: section numbers only, without saying whether the policy was breached.

Keep it under 900 words. Use "said", "stated" or "described" for everyone, and never "admitted" or "claimed" for one side only, because word choice can tilt a neutral record. The record should read the same to the complainant and the respondent: factual, complete and even-handed.

Rules:
- Use only the notes. Do not add motives, characterizations or opinions about anyone's credibility.
- Do not conclude whether an allegation is substantiated, and do not recommend any action. That decision belongs to {{decision_maker}}.
- Mention a protected characteristic only where it is part of an allegation itself, and then only as stated in the notes. Do not infer anyone's characteristics.
- Leave out personal details that are not relevant to the allegations.

If the notes are unclear about who said something, or a statement could be read two ways, say so under Gaps rather than choosing one reading.

Variables

{{decision_maker}}
Who decides the outcome, for example "the head of HR".
{{advisor}}
Who advises them, for example "employment counsel", or "no one".
{{complaint}}
The complaint as it was received, with personal details that are not needed removed.
{{interview_notes}}
Notes from each interview, labeled by role rather than name.
{{policy}}
The policy sections that may apply, pasted in. Write "none yet" if you have not identified them.

Why it works

  • What to avoid. It sets accounts side by side with no findings, credibility judgments or recommendations, because the decision belongs to a person.
  • What good looks like. The same neutral reporting verbs for everyone, never "admitted" for one side, keep the record even-handed.
  • If unsure. Unclear attributions and missing evidence go into a Gaps section the investigator can act on.

How to use it

  1. Press Fill in variables and type your details into each blank. What you type stays in this browser.
  2. Copy the filled prompt into the AI tool you already use, such as ChatGPT, Claude or Gemini, or open it in the Grader or the Prompt Tester here.
  3. Before you rely on it, try it on three or four real examples. The Prompt Tester runs them all at once and shows which ones go wrong.