AI Agent Output Review Burden and Trust Calibration Survey

Thanks for taking a few minutes on this! We're studying how much work it takes to review AI agent outputs before you trust and use them. Your responses are completely confidential and anonymized. Honest answers help us reduce your review burden. About 5-6 minutes.

Which AI agent or tool do you use most often in your work?

• (Replace with Agent A)
• (Replace with Agent B)
• (Replace with Agent C)
• Other

In the last 30 days, how often did you use this agent's output?

• Multiple times a day
• Once a day
• A few times a week
• Once a week or less

When this agent gives you an output, how much do you currently trust it to be correct without checking?

How much of the agent's output do you actually review or verify before using it, on average?

Rate the effort each of the following review activities takes, on a typical task.

How often does each of these happen with this agent's output?

What most often triggers you to do a full manual re-check instead of a quick glance?

• The task is high-stakes (money, legal, customer-facing)
• The output looks unusual or inconsistent with what I expected
• I've been burned by an error from this agent before
• It's a new or unfamiliar type of task
• A colleague or policy requires it
• I always do a full check regardless

Rank these factors by how much they currently increase your review burden, from most to least.

1. Output is hard to verify against a clear source of truth
2. Agent doesn't explain its reasoning or show its work
3. Errors in the past have been high-impact when they happened
4. Task volume is too high to check everything carefully
5. Unclear who is accountable if the output is wrong

Overall, is the current level of review you do on this agent's output too much, too little, or about right?

Reconstruct the most recent specific instance where the respondent's trust in this agent's output was wrong in either direction: a time an error slipped through with too little review, or a time they over-reviewed something that turned out fine. Get concrete details on what the output was, what checking they did or skipped, what happened as a result, and how that changed their review habits afterward. If they say they always fully check everything, probe whether that's sustainable given their task volume and what would let them safely check less.

Last few questions are about you, totally optional.

How long have you been using AI agent tools in your work?

• Less than 3 months
• 3-12 months
• 1-2 years
• More than 2 years
• Prefer not to say

Which best describes your role?

• Individual contributor
• Team lead / manager
• Director or above
• Other
• Prefer not to say

All done — thank you! Your answers feed directly into a report on where AI agent review effort can be safely reduced and where trust needs stronger guardrails.