Approving work you did not watch being made

Reading used to tell you something about the process

An experienced reviewer picks up more from a document than its contents. Rough patches suggest where the writer struggled. A section that is thinner than the others suggests the evidence was thin. Clumsy phrasing around a claim suggests uncertainty about the claim. None of this is explicit and all of it guides where attention goes.

Those signals came from the fact that writing is effortful and the effort leaves marks. Where a draft has been smoothed, the marks are gone, and every section reads as though it received the same care.

So a reviewer working from the text alone is now working with less information than they think, and the deficit is invisible, because the document reads better than the ones they used to review.

Uniform surface, uneven substance

The specific risk is a document where one section rests on work somebody did and another rests on a plausible account of what such a section usually contains. Both read the same. A reviewer distributing attention evenly will spend it in proportion to length rather than in proportion to risk.

The consequence is not that generated material is untrustworthy in general. It is that the reviewer has lost the cues that used to concentrate scrutiny, and nothing has replaced them by default.

This falls hardest on the person signing rather than on the person drafting, which is worth stating plainly, because the sign-off is usually where the accountability actually sits.

It is made worse by the fact that a smooth document is pleasant to review. Approving something that reads well takes less effort than approving something ragged, and the reduced effort registers as the work being better rather than as the review being lighter.

Ask about the process, not about the tool

The question people reach for first is whether a tool was used, and it is close to useless. It is asked defensively, answered defensively, and the answer does not tell you what you need to know, since a heavily assisted document can be thoroughly checked and an entirely hand-written one can be full of unverified assertions.

The productive questions are about verification and sourcing. Which figures were checked against a source, and which source. What in here is your judgement rather than a summary of a common position. Which parts are you least sure of. What did you leave out.

Asked as ordinary professional questions rather than as an inquisition, these are the same questions a good editor has always asked. They just used to be answerable by reading, and now they have to be asked out loud.

The tone of the asking matters more than it should. A question framed as suspicion produces a defensive answer that tells you nothing, whereas the same question framed as an ordinary part of review produces the actual state of the work, including the parts the writer is uneasy about.

Ask for the working, not only the output

For anything consequential, the reviewable artefact should include what supports it: the sources, the calculation, the notes, the record of what was decided. That has always been true of financial and technical work and it is now worth extending to ordinary documents that carry claims.

This shifts the review from an assessment of plausibility to an assessment of support, which is the only kind of review that survives at volume. Plausibility is exactly the property that generated material has in abundance.

It also changes behaviour upstream, which is where most of the benefit lands. A writer who knows the sources will be asked for keeps them as they go, and the keeping is the discipline that prevents the problem rather than detecting it.

Say what your signature means

A reviewer should be clear, at least to themselves, about what approving something asserts. Sometimes it means the content has been verified. More often it means the document is fit to circulate, the judgement seems sound and the obvious risks have been considered. Those are different claims and they are routinely conflated.

Where the honest answer is that you cannot verify the substance in the time available, the useful response is to say so and to name what you did check, rather than to sign as though you had done more. A qualified approval is a real thing and it is more useful to everybody than a confident one that nobody can rely on.

And there is a category where reviewing the output is simply not the right control: work whose correctness depends on domain expertise the reviewer does not have. There the answer is to route it to somebody who has it, or to accept that the sign-off is administrative rather than substantive and to be honest about which one it is.