AUTEURIUM.

AI script feedback: what it can and can't do

Updated July 2026. Written by the team behind an AI development tool, which is exactly why the failure modes below get named instead of hidden. Our product appears once, clearly labeled.

Writers are already using AI for feedback, mostly quietly. Some of it genuinely helps. Some of it is worse than nothing. The difference is knowable in advance, so here is the honest map: what is documented to go wrong, what AI actually reads well, what it cannot judge, and the questions that separate a serious tool from a flattery mill.

The documented failure modes

Documented cases: No Film School's tests of paid coverage services (one, two); WGA member guidance on AI hazards (wga.org).

What AI reads genuinely well

What AI cannot do, and no honest tool claims

How to vet any AI feedback tool in four questions

Full disclosure: this one is ours

Auteurium is our answer to those four questions. The testing is public, numbers, sample size, and limits included: how our readers are tested. Zero notes on a clean scene is a valid read, by design. Your pages never train any model, with the contract clause cited: the stance page, signed. And it reads, it never writes your script: notes arrive in the margin of your editor, and nothing touches your page unless you explicitly accept it.

What it is not: coverage, scores, or access. If AI feedback interests you but you buy nothing from us, the four questions above still work on everyone else.

Auteurium · The post-Coverfly map · What coverage costs in 2026 · Nicholl 2026: how to submit · Where we stand on AI