Let AI Teach Me · Hard reading

Let AI Keep the Key Facts When It Edits

AI says the edit has no problems — maybe because it deleted the hardest, heaviest paragraph outright.

You rename the file “PM resume · final.pdf” and hit send. The portal says submitted. You close the laptop and go wash a cup.

You edited this version Sunday night. The posting asked for “owned a project from requirements to launch.” The closest thing in four years was last year’s new-user benefits page. The draft ran 6 lines: you joined requirements, coordinated design and engineering, and wrote signup conversion up 30% — without saying where the number came from.

You gave AI the resume and the posting: “edit until nothing is wrong, and every line can survive interview follow-ups.” Two minutes later it returned 8 checks: typos pass, numbers have sources, experience matches the role — last row green too. You glanced again. Two pages became one. The sentences read smoothly. You decided this version could go out.

The next afternoon the recruiter called and asked one thing: “Why isn’t there an example of you shipping a new product?” The call lasted 6 minutes. You opened both files side by side and saw it: those 6 lines on the new-user benefits page were gone. AI did not add the number’s source. It did not fix “participated.” It deleted the experience that would have drawn follow-ups.

The draft you gave AINew-user benefits page: joined requirements, coordinated design and engineering, launched in 7 days; signup conversion up 30%.
The edit AI said had passedProficient with prototyping tools, strong communicator, delivers on schedule.

You stopped because the checklist only looked at sentences still in the edit. Delete “up 30%” and every remaining number has a source. Delete the whole experience and there is no overstated scope to catch. Every item passed. The evidence the recruiter most wanted left with it.

Fewer problems is not the same as a better document. If you only check what stayed, deletion becomes the cheapest path to a pass.

AI can still edit. The check just has to ask one more thing: are the draft’s key facts still there? Have it list the experiences, numbers, and limits that must stay, then fix sentences. At the end, compare item by item. Every deletion has to name the cost. It can repair the unclear line. It cannot quietly walk off with the whole piece of evidence. Run both checks on the same resume — what comes out different?

Try it first · Watch the same resume pass a check
The same resume, edited once under each check
What this check will look at
Play the left setting first. Watch which experience disappears at the end.
Checking deletions?This run will not check what was deleted
8Key draft facts still left
0Lines that can survive follow-ups
0Problems this check found
The numbers count the whole resume. The four cards below each show one case.
Fact A · Project resultThe number’s source is clear
Fact B · Scope ownedNeed to check how far you actually owned it
Fact C · Launch processNeed to confirm no key move was dropped
Fact D · Role evidenceNeed to confirm the closest experience is still there
What the edited resume says now
After play, AI’s edits will show here item by item.
What the matching draft said
Here the edit and the matching draft lines will sit together for a check.
Pick a check, then play the full run.
Why every check passedThe check only reads sentences left in the edit. Once a whole paragraph is gone, nothing is left to trip a flag.
What happens under the other checkPut the draft’s 8 key facts on the checklist too. Deleting 5 of them shows up at once.
Teaching demo: resume lines and numbers are written for the demo, not a real candidate.
Paste these three paragraphs straight to AI

Before edits, have AI list the draft’s facts. After edits, have it walk that list item by item. The three blocks below can go in one send.

Copy this whole block
  1. Lock the facts that must not quietly vanishRead my draft and the target requirements first. List the experiences, numbers, limits, and proof details that must stay. Cite the matching sentence from the draft for each. List only. Do not rewrite yet.
  2. Then fix the shaky wordingEdit the draft against that list. When you find overclaim, missing sources, or unclear lines, point at the problem first, then ask what I still need to add. Do not make the check pass by deleting a whole paragraph.
  3. Finally check, item by item, what was droppedCompare the edit with the draft item by item. Output three columns: the draft fact, where it sits in the edit, and why it changed. If an item is gone, say what evidence I lost — then wait for me to confirm.

The third block’s “what evidence I lost” is the hinge. If you only ask what was deleted, you may hear “removed redundancy.” Ask for the cost, and you see whether that cut took the experience the role cares about most.

After it passes, ask one more thing

When you see “no problems,” ask next:Compared with the draft, which experiences, numbers, conditions, or examples are gone? List them one by one, and say what kind of proof I lost with each.

That question moves the check from the remaining sentences to the gap between draft and edit. What stayed has to survive follow-ups. What should have stayed needs a destination. See both, and a green result means something.

That opening new-user benefits paragraph can be shortened. You can also change 30% to “signup conversion rose after launch; metric definition still to add.” Keep the gap until the source is filled, and the recruiter can still see what you did. Delete the whole block and the sentences are clean — the proof of skill is zero.

This comes from editing and review practice. It is meant to stop a goal from being over-optimized. There is no lab threshold attached.

All green only means what stayed passed. It does not mean what should have stayed is still there.