Six questions, five minutes, and a clear next step for the AI work in your business.
We’ve all experienced AI slop. It’s what happens when a bad writer automates their writing. Of course, I am constantly improving mine :).
Slop is the AI mistake you can see. The costlier kind is quieter: a wrong price in a polite email, sent before anyone read it.
So the useful question isn’t which tool you picked. It’s what needs a human look before it goes out. This check sorts your AI work three ways: what needs a check, what needs practice, and what’s fine as it is.
It’s a formal assessment, just not a comprehensive one. It helps you see where you stand on the decisions you make about AI and how you use it.
You’re not behind: most firms are still figuring this out
The Federal Reserve’s 2026 Report on Employer Firms draws on its 2025 Small Business Credit Survey. It found that 46% of small employer firms use AI. Among the firms that use AI:
- About half are still experimenting
- 44% have partially integrated it into their business processes
- 7% have fully integrated it
So if you’re still figuring it out, you’re in good company. (A lot of company, actually.)
One idea runs the whole check
Review scales with risk and complexity.
Simple, low-risk work needs only a quick glance. Work that could cost money, upset a customer, or is hard to get right needs a careful human check before it goes out.
Answer six questions
Think of one AI use that matters most in your business.
You can repeat the check for other uses later.
For each question, pick the answer that fits most of the time. Score Yes = 2, Sometimes = 1, No = 0, and add up your points as you go. A Yes means more risk or more difficulty.
- If AI got this wrong, could it cost you money, a customer, or your good name?
- Does what AI produces go to a customer, a supplier, or the public without a person reading it first?
- Does it set a price, make a promise, or use private information, like customer details, pay, or bank details?
- Is the task hard to explain, with lots of steps or lots of exceptions?
- Would you or your team struggle to tell a good answer from a bad one?
- Is it unclear when a person must check the work before it’s used?
Add up your score
Your score is out of 12. 8–12 = Check, 4–7 = Train, 0–3 = Leave alone. The higher your score, the more risk and complexity in that work, and the more human review it needs.
One exception: if you answered Yes to question 3, treat it as Check, whatever your score. Prices, promises and private information don’t get a free pass.
Check: 8 to 12 points
A person reviews every result before it goes out. Use this when a mistake could cost money, upset a customer, or be hard to spot.
Do this: Name one person to read and approve this work before it goes out, every time, starting today.
Train: 4 to 7 points
Your team builds the judgment to tell a good result from a poor one. Use this when the work is useful but your team is still learning what good looks like. Training is practice, not a review step. Until your team can tell good from poor, keep a light check in place.
Do this: Pick one common task this week and practice together. Look at a good result and a poor one side by side, so everyone learns what to watch for.
Leave alone: 0 to 3 points
The work is low risk and easy to judge. The quick glance you already give it is enough.
Do this: Keep going as you are. Run this check again if the task, the risk, or the team changes.
A quick sort for everyday work
| Sort | Typical examples |
|---|---|
| Check | Quotes and prices, messages to customers, anything with money, deadlines, promises, or private information. |
| Train | First drafts, summaries, and plans your team is still learning to judge. Useful work that needs better judgment, not more rules. |
| Leave alone | Sorting and tidying lists, rewording routine internal notes, and early ideas that stay inside the business. |
Get the printable sheet
Download the AI Confidence Self-Check (PDF) to use on a call or share as is. It’s practical guidance for your own judgment. It’s a formal assessment, just not a comprehensive one.
Want to talk through your results? Get in touch at adambloom.me.
Sources and claims
| Claim | Source | Status |
|---|---|---|
| 46% of small employer firms use AI | Federal Reserve Banks, 2026 Report on Employer Firms (2025 Small Business Credit Survey, 6,525 employer firms with 1–499 employees, fielded Sept–Nov 2025) | Verified at source, Nov 11, 2026 |
| Of AI users: about half experimenting, 44% partially integrated, 7% fully integrated | Same report | Verified at source, Nov 11, 2026 |
| Six questions, scoring, the three sorts and the Question 3 exception | A.Bloom & Co. AI Confidence Self-Check | Adam’s own assessment |
| “We’ve all experienced AI slop…” | Adam Bloom | Adam’s own words |
| A wrong price in a polite email | — | Illustration |