Applicant has 5 real insights uses AI to expand into 10 polished pages
Reviewer has 10 pages uses AI to compress back into 5 bullet points
Net signal transferred: ≈ 0
Two well-designed systems. Neither broken. No longer compatible.

A foundation I work with received 340 applications last cycle.

Their program team estimated that roughly 60% were AI-assisted. Reasonable. AI makes good organizations sound clearer and gives under-resourced ones a fighting chance at a professional application. That's not a problem. That's the technology working.

The issue: their review process was also AI-assisted. Triage tool, synthesis layer, scoring rubric. Two well-designed systems. Neither broken. Just no longer compatible.

Each pass through AI normalizes the output — filters for legibility, strips the friction and texture that actually predict whether an organization can execute. The process was designed for a world where writing quality was a proxy for organizational quality. AI dissolved that proxy. Which means the process needs a different signal layer, not just a better one.

Three upgrades worth considering
Constrain before you expand.
Ask for 200 words before 2,000. What an applicant can say clearly without AI assistance is the actual idea. What requires ten pages to explain probably isn't ready yet.
Design for specificity over narrative.
Not "describe your impact" — try "describe a decision you made that didn't work and what changed because of it." AI is genuinely good at synthesizing patterns. It cannot fabricate lived operational experience without revealing itself.
Introduce a moment the applicant cannot fully prepare for.
A short unrehearsed response, an unexpected question with a 48-hour turnaround, a conversation that goes off-script. The format matters less than the principle: at some point, the prepared narrative has to make contact with something unpredictable. That's where the signal lives.

The goal isn't to catch anyone doing something wrong. AI-assisted applications are a reasonable response to an AI-assisted world. The goal is to upgrade the process to match it — so the organizations that actually have something real are the ones that surface.

Your process was designed for a world where writing quality predicted organizational quality. That world is gone.