Show AI what good looks like

The move Show AI five judgment calls you've already made (your score + a one-line why); it infers your standard from them. Your examples carry the standard.

The short version: AI doesn't know your bar. Instead of trying to explain every rule behind your judgment, show it five calls you've already made, each with your score and a one-line why. It picks up your standard from the examples. Deciding which examples to feed it stays yours.

You make the same call over and over. Scoring sales leads. Flagging which support tickets are urgent. Sorting customer feedback into themes. Each time, you weigh the same few things in your head and land on an answer.

Why AI gets your call wrong

So you hand a batch to AI and ask it to score them. It's smart, surely it knows a strong lead when it sees one. But what comes back is off. It rates the tire-kicker an 8 and the real buyer a 5. Not because it's dim, but because "a good lead" means something specific to you and your business, and AI was never in the room for the hundred calls you've already made. It has no way to know your bar.

Your examples already carry the standard

You could try to teach it by writing the whole standard out: a rubric, a scoring guide that spells out every criterion. But you've already made the decisions that guide is trying to describe. Show AI five of them, each with your score and a one-line why. Don't spell the standard out. Show it.

Paste five calls you've already decided, then hand over the new batch:

Here are 5 leads I've already scored, with my rating and a one-line why:

- Enterprise, named a budget, asked for a demo -> 9, real intent
- "Just researching for a class project" -> 2, no buying motion
- Director at a target-size firm, vague timeline -> 6, worth a nurture
- Competitor email domain -> 1, skip
- Referral from a current client, wants pricing -> 8, warm

Before you start, ask me any clarifying questions about how I'm scoring these, so you're not guessing at my standard.

Then score this week's batch the same way I would. For each: the score, the one-line why, and flag any you're genuinely unsure about instead of guessing a number.

What comes back

What comes back checks the gap your examples left open, then scores with your logic, not a generic one:

One thing first: none of your five cover a lead who showed interest then went quiet. Do I weight the interest or the silence? I'll lean on the silence unless you tell me otherwise.

Ops lead, downloaded the guide, no reply to two emails -> 5, interest without motion, nurture.
VP, asked about security review -> 8, buying signal, route to sales.
Unsure: freelancer, could be a one-seat buyer or a tire-kicker. Flagged, not scored.

Make it stick

Then make it stick: save the whole thing as a saved prompt or a Project, so next week you paste the new batch and your standard is already loaded.

Don't spell the standard out. Show it.