Why proofreading is the wrong model
Drafts produced by a capable tool are usually well formed. The sentences work, the structure is reasonable, and nothing looks wrong. That fluency is precisely what makes the review dangerous if it is done by reading.
The failures are not stylistic. They are a confidently stated figure with no source, a credential described slightly incorrectly, a process that is not how your business actually works, or a claim that sounds standard and is not defensible.
None of those announce themselves. A page can read beautifully and contain three statements you would have to retract. The review has to be a check against something external, not a judgment made while reading.
What only a person can settle
Whether alt text describes the actual photograph on the page. Whether a testimonial is real and was given permission. Whether a figure traces to a source that still says what it used to say.
Whether the credential named is held today, by this business, in the states where the page claims it. Whether the process described is the process you follow. Whether the price is the price you charge.
These are the items in the never-automatic tier of our own remediation policy, and the reasoning is the same here. A machine correcting any of them would be inventing a fact on your behalf.
The review checklist
This is what we run on our own pages. Worked through in order it takes roughly ten minutes on a page of moderate length, and almost all of that is the first three items rather than the reading.
- Every number opened at its source and confirmed current
- Every claim about the business checked against how it works
- Every credential checked with the body that issued it
- Every image checked against its alt text
- Every link followed once to confirm it goes where it says
- Every promise removed unless it can be evidenced
- One read for tone, last rather than first
Who should do the review
Somebody who would notice if it were wrong. That usually means the person who does the work rather than the person who manages the website, because the errors that matter are errors of trade knowledge.
That person does not need to write. Reviewing is faster than drafting and uses different skills, which is the honest case for using tools at all in a small business without a writer.
Name them on the page. Our page on author and about pages covers why a named reviewer with a real background is worth more than a byline nobody can check.
Review at volume
Reviewing one page carefully is easy. Reviewing three hundred is a process problem, and the usual failure is that review quality decays quietly somewhere around page forty while the pace stays the same.
Separate the mechanical checks from the judgment checks. Headings, links, lengths, missing alt text and heading order can be checked by a tool. Our site check reports those across a whole site.
That leaves the human with only the questions a human can answer, which is the only version of this that stays honest at scale. It is also the design principle behind our own publishing gate.
Recording that the review happened
Keep a record of who reviewed what and when. On a small site a single file is enough. It costs nothing, and it is the difference between saying your content is checked and being able to show it.
Put a visible review date on the page as well. It helps readers judge how current the material is, and it is one of the signals our page on trust signals describes as genuinely checkable.
Then actually honor it. A review date that has not moved in two years on a page about a changing subject is worse than no date, because it states confidently that somebody looked recently.
When to reject a draft entirely
When it is fluent about something it cannot know. Drafts that describe your process, your history or your results in convincing detail are inventing, and editing them is slower than writing the section yourself.
When the subject is regulated. Anything touching a licensed profession, a medical claim or a financial product needs to be written by somebody accountable for it, then checked by somebody who knows the rules.
And when the draft is about a topic where being slightly wrong is expensive. Our page on what to keep off the list sets out where we draw that line and why.