A simple test for AI work that sounds smarter than it is

The move Before you send the AI-assisted draft, ask what's polished, what's evidenced, and what would change the conclusion. Fix the gap in your own words.

The short version: Most AI drafts make your case sound more confident. The test that tells you whether it's actually sharp takes thirty seconds: ask what would change the conclusion. If nothing in the draft addresses that, it's decoration, not thinking.

You've got an AI-assisted client recommendation sitting in your drafts. Tight structure, confident framing, a summary that lands in two sentences. Before you hit send, one question tells you whether it's actually sharp or just polished.

AI is good at making a case sound better than your raw notes. Tighter phrasing, cleaner structure, a more confident read. That's useful. It's not the same as being right. A polished recommendation can still fall apart the moment someone asks "but what if that assumption is wrong?"

If nothing in the draft would change your mind, the draft isn't thinking. It's decorating.

Run this before you send it

The test takes thirty seconds:

I'm about to send this [memo / recommendation / brief] to [who: their role
and what they care about].

Before I do, answer three things honestly:

1. What's polished? What reads better (cleaner, tighter, more confident)
   but doesn't add evidence or thinking I didn't already have?
2. What's evidenced? What's grounded in specifics I provided or facts
   that can be verified?
3. What's the one thing that, if true, would change this conclusion?
   Does the draft address it?

Don't soften it.

Filled in, it looks like this:

I'm about to send this quarterly strategy recommendation to the VP of
Marketing at Relay Corp. She cares about pipeline conversion and has
pushed back twice on projected numbers.

Before I do, answer three things honestly:

1. What's polished? What reads better (cleaner, tighter, more confident)
   but doesn't add evidence or thinking I didn't already have?
2. What's evidenced? What's grounded in specifics I provided or facts
   that can be verified?
3. What's the one thing that, if true, would change this conclusion?
   Does the draft address it?

Don't soften it.

AI hands back:

1. Polished: The executive summary, the three-pillar structure, and the "we recommend" framing are tighter than your original notes but don't introduce new evidence. The competitive comparison reorganizes your bullets without adding data.

2. Evidenced: The Q2 retention numbers come from the dashboard you pasted. The churn rate and NPS delta are grounded. The "industry average" benchmark of 74% has no source in what you gave me. I pulled that from general training data, not from your inputs.

3. The conclusion-changer: Your ROI timeline assumes Relay's Q3 pipeline converts at the historical rate. If that pipeline is already contracted, not projected, the payback window slides by two quarters and the recommendation reverses. The draft doesn't address that.

Why this helps: You catch the gap between "reads well" and "holds up" before the client catches it for you.

Save this as a Project in ChatGPT or Claude and call it "Final-pass review." Drop in any draft before you send it. The three questions stay the same; only the context changes.

AI flags the gap. Whether to fix it or send it as-is is your call.