The practical takeawayTurn the draft into a short list of claims. Check each against its source and send unresolved questions to the person who can answer them.
- List the claims
- Check the evidence
- Resolve the gaps
Imagine an AI assistant drafts an announcement for a nonprofit's spring fundraiser. The tone sounds right, the agenda is clear, and the last sentence offers a discount for tables of eight. The brief mentions tables of eight. It says nothing about a discount. Someone who reviews only for tone may approve an offer the organization never made. A short training exercise can teach a team how to catch that exact mistake.
Read once as the person receiving it
Give participants the draft as it would appear in an email or on a landing page. Ask what a reader could reasonably do after seeing it. Someone might buy a ticket, travel to the venue, wait for a promised callback, or budget for a discount. Underline the details that would influence that decision.
Set the AI tool aside for this pass. Reading in the final format helps people see what their reader sees. An internally flagged placeholder may look like finished copy once it reaches a page. A footnote about limited availability may disappear below a mobile button. The goal is to identify the commitments the team is about to make.
Separate sourced facts from plausible additions
Work through each underlined statement with the original event brief open. A simple table on paper is enough. Give each line a claim, a source, and a decision. “Doors open at six” can point to a specific line on the venue schedule. “Free parking” needs its own evidence even if the previous event offered it.
A second answer from the same tool gives you another piece of generated text to review. For an event date, use the organizer's approved calendar. For a published product specification, check the manufacturer's current material. For a policy, use the version the organization has approved. If two authoritative documents disagree, send the discrepancy to the owner and record the answer.
Verify the meaning behind a number
Give the group a fictional example with twelve available tables, eight seats at each table, and two tables reserved for staff. Ask them to check a draft that advertises ninety-six tickets for sale. The multiplication is accurate for total seating. The number available for purchase is eighty under the stated assumptions. Both the arithmetic and the label deserve a check.
Repeat the exercise with a dealership example. “Appointments are available for forty vehicles” means something different from “forty appointments were completed.” Ask which date range, location, and status the source covers. Recalculate simple figures outside the AI tool and give someone familiar with the operation a chance to spot a misleading interpretation.
Practice a useful response to uncertainty
Have everyone write a one-sentence note for the event organizer about the unsupported discount. A good note names the exact claim, explains where the team looked, and asks a question the organizer can answer quickly. Once the answer arrives, the editor can publish the approved wording or remove the offer entirely.
This exercise works best when the facilitator rewards people for catching an uncertainty before publication. A team under pressure to return a perfectly finished answer will sometimes hide the unresolved point. Show the difference the note makes to the colleague who has to approve the email and the guest who may rely on it.
Set a review rule the team will use
Before the workshop ends, decide which outputs require a second person. A customer-facing commitment about price, permissions, availability, or a deadline usually deserves an accountable reviewer. A private outline for a meeting may need only the author to verify the details. Use examples from your own organization so people can identify the boundary during a busy day.
Save the practice draft with its corrections and the reasons behind them. A week later, bring another short example and let someone else lead the review. Look for whether participants go directly to the source, distinguish a calculation from its business meaning, and escalate an unresolved claim. Those are observable skills you can teach again.



