Team Review Practices for AI-Assisted Work
Module 4 · Discernment in Practice

Team Review Practices for AI-Assisted Work

Discernment breaks down when everyone assumes someone else checked. Set up clear review ownership, AI-use notes, second-reader checks, and an error log so your team catches problems before they ship.

Everyone Thought Someone Else Checked

On a team, AI errors rarely slip through because nobody cares. They slip through because of diffused responsibility:

  • The writer assumes the reviewer will catch problems.
  • The reviewer assumes the writer verified the facts, and reads for tone and flow.
  • The manager approves because two people have already looked at it.

Each person did a reasonable job of what they thought their job was. Nobody checked the figures. AI makes this worse because drafts arrive faster and look more finished, so there’s more to review and less obvious reason to dig in.

The fix is not “everyone check everything.” It’s making review explicit.


Practice 1: Name the Checker for Each Claim Type

For any piece of work that matters, decide in advance who verifies what:

CheckWho owns itExample
Facts, figures, citationsUsually the authorEvery statistic traced to its source before review
Fit for audience and purposeReviewerWill this make sense to the client?
Specialist contentSubject expertLegal, medical, financial, or safety claims
Final sign-offOne named personAccountable for what goes out

Writing this down once, for recurring work types like client reports or policy updates, takes ten minutes and prevents the “I thought you checked” gap.

Practice 2: Add an AI-Use Note for Reviewers

Reviewers judge better when they know where AI was involved. A short internal note, separate from any public disclosure, tells them where to look hardest:

AI-use note (internal):
- Sections 2 and 4 drafted with AI from my bullet notes.
- Figures in Table 1: AI-extracted from the Q3 report; I verified
  all 12 against the source.
- Market sizing in Section 3: AI estimate, NOT yet verified.
  Please flag if this needs a source before sending.

This turns review from “read it and see if anything seems wrong” into “check these specific things.”

Practice 3: Second-Reader Checks for High-Stakes Work

For high-stakes output, one person verifies claims independently. The second reader doesn’t just read the author’s version; they check a sample of claims against sources themselves. A fresh reader who didn’t spend an hour with the AI is less anchored by its framing.

Practice 4: Keep an Error Log

When an AI error is caught, or worse, missed, record it briefly:

  • What went wrong (a fabricated citation, a misread column, a wrong assumption)
  • Where it was caught, or where it surfaced
  • What check would have caught it earlier

After a few months, the log shows your team’s real risk areas, which should shape your checklist and training. Keep it blame-free: the goal is to fix the process, not to embarrass whoever used the AI.

  1. 1 Draft + verify

    Author checks their own facts and figures

  2. 2 AI-use note

    Where AI helped, what's unverified

  3. 3 Review

    Fit, flagged items, sample of claims

  4. 4 Expert

    Specialist claims, when needed

  5. 5 Sign-off

    One named, accountable person

⚠️ Common Pitfall: Hiding AI Use to Avoid Scrutiny

Common mistake: not telling reviewers AI was involved, because people worry their work will be judged as less valuable.

This removes the most useful signal reviewers have. Teams that treat AI use as normal and expected, and focus on whether the work is verified, get more honest notes and catch more errors. If your team stigmatizes AI use, raise it: the stigma itself is a quality risk.

🧪 Try It: Design a Review Plan for One Recurring Deliverable
  1. Pick a recurring piece of work your team or class produces where AI could help, such as a monthly report, a newsletter, meeting minutes, or a project proposal.
  2. Ask an AI assistant to help you think it through:
    My team produces [deliverable] every [frequency]. We're starting
    to use AI to help draft it. What kinds of errors is AI likely to
    introduce in this type of document? Group them by type.
    Judge the list. Add anything it missed from your own experience, and cross out anything that doesn’t apply.
  3. Fill in the ownership table from Practice 1 for this deliverable, with real roles.
  4. Write an AI-use note template your team could copy for it.
  5. Decide which error types, if any, need a second-reader check.
🧠 Knowledge Check

1. A report goes out with a wrong statistic. The writer assumed the reviewer would check figures; the reviewer read only for clarity. What's the underlying problem?

2. What makes an internal AI-use note useful for a reviewer?

3. What's the main purpose of a team AI error log?


Up Next

The next lesson pulls the whole course into one reusable discernment checklist you can tailor to your own work.