Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
Main idea
A reviewer given a draft article, and nothing else, can only produce objections that would hold anywhere: the example could be more concrete, the transition is fast, the thesis deserves more nuance. These remarks are true and without effect, because they don't depend on the text under examination.
The objections that count are those born of a comparison: this passage repeats an earlier text, this example doesn't support the stated thesis, this conclusion goes further than the material allows, this angle was already ruled out and you don't say why you're coming back to it.
None of these remarks can be formulated without access to something beyond the text — the earlier writings, the notes, the decisions made, the goal pursued.
Layer added by "AI Doesn't Replace Judgment" (2026-09-02). The negative test gets sharper: generic advice is recognizable by the fact that it could be given to almost any text — specify the audience, add an example, define the terms, qualify the thesis. This is good advice, and that is precisely what makes it worthless here.
This layer also adds what situated feedback is made of. It isn't born of a general competence in analysis, but of the meeting of five things: an intention, a material, a memory, rules and a real state of the work. Remove one of them and the feedback becomes generic again — feedback that knows the material but ignores the intention judges a text by a goal that isn't its own.
Why it matters
This shifts the question "is this feedback relevant?" to "could this feedback have been formulated without reading my material?". The second criterion is checkable, the first is not.
It is also what separates interchangeable assistance from situated assistance. The difference lies not in the subtlety of the reasoning but in what is accessible at the moment the feedback is produced.
Nuances and limits
Access isn't enough: having a corpus doesn't guarantee that it is referred to. Feedback can cite the material and remain generic.
And feedback that is too situated has its own risk — it pulls the text towards what has already been written, at the expense of what departs from it.
Open questions
- How do you tell feedback that prevents a repetition from feedback that prevents a deliberate return to a subject?