Key takeaways
- Supply relevant source material and say what decision the output must support.
- Separate missing information, proposed assignments and confirmed facts.
- Check commitments and figures after a rewrite, even when the prose reads smoothly.
A chatbot can help an employee prepare a meeting brief, compare documents or revise a draft. Start with material you are allowed to share, describe who will use the output and specify what they need to do next. Reviewing three supplier proposals requires different instructions from preparing a short explanation for a customer.
Choose a task you can check against its source material. NIST’s 2024 guidance on generative AI describes how these systems can present incorrect content confidently, including invented supporting references. A document supplied to a chatbot gives a reviewer something concrete to compare with its answer.
Compare proposals without filling in missing terms
Give the chatbot the proposals and name the criteria that matter: price, contract length, support hours, installation requirements and cancellation terms. Ask it to identify where each answer appears. Tell it to mark missing information as missing, rather than completing the comparison from general knowledge of a supplier.
Read the cited passages before selecting a supplier. A price may cover a different number of users, an installation fee may appear in an appendix, or a cancellation clause may apply only after a minimum term. Turn unresolved points into questions for the supplier and keep them separate from confirmed terms.
Ask for a diagnosis before a rewrite
For a proposal or memo, request a review of specific weaknesses. Ask which claims lack support, which terms need defining and where a reader would need another fact to make a decision. Supplying the intended audience and a length limit gives the model useful constraints.
Then choose which changes to make. If you request a rewrite, tell it to preserve quoted language, figures and commitments exactly. Compare those items afterward. A changed delivery date or an added promise can matter more than a smoother sentence.
An example instruction: Review this proposal for a finance director deciding whether to fund a trial. Identify costs we have omitted, assumptions we have not supported and terms that need clarification. Explain each concern before suggesting revised wording. Do not add claims about results or customers.
Use an explanation to prepare better questions
When reading unfamiliar technical material, provide a passage and ask for its terms to be explained in ordinary language. Follow up on the specific step you do not understand. For a calculation, ask the system to name every input and unit, then reproduce the arithmetic in a spreadsheet or calculator.
Keep the distinction between a general explanation and advice about your situation. A chatbot explaining a contract term has not established how that term applies under a particular agreement or jurisdiction. Bring the relevant text and remaining questions to the person responsible for reviewing the contract.
Turn notes into a draft action list
With an authorized transcript or set of notes, ask for decisions, proposed actions, named owners and deadlines. Require it to distinguish an agreed assignment from a suggestion. If nobody accepted a task or supplied a date, leave that field unassigned.
Send the draft to meeting participants through your normal review process. They can correct missing context and accept responsibilities before the list becomes a project schedule. Keep links to the relevant discussion where your records policy allows them.
Rehearse a conversation with stated assumptions
A chatbot can also play a possible customer during preparation for a sales call. Describe the offer, budget constraints and objections already raised. Ask it to challenge one issue at a time, then review which of your answers contained evidence and which relied on promises.
Treat the exchange as practice. The simulated customer has no independent knowledge of a real buyer’s priorities. Write down unanswered questions to ask on the actual call, such as who approves a purchase, which requirements are essential and what would prevent a trial.
For the next document you need to prepare, save the instructions, the source material and your corrected output together in an approved workspace. On the next attempt, check whether those corrections were applied and how much review the new draft still required.
