Start with the claims, not the homepage
The instinct is to open the homepage and read down. That order puts the most-read page first and the most dangerous content somewhere in the middle, which is exactly backwards for a first pass.
Search the whole site for the things that carry consequences. Credentials, licenses, years in business, number of customers, results achieved, reviews and anything phrased as a promise.
Each of those is a statement somebody could act on, or that an answer engine could repeat. Our page on what to keep off covers why these specific items are the ones to hunt for first.
The first pass, in order
This is the sequence we would use on any generated site. Working in this order means that if you run out of time, the things you did not get to are the things that matter least.
- Credentials, licenses and registrations
- Any figure about results, customers or years
- Reviews, testimonials and endorsements
- Prices and anything described as included
- Service area, hours and contact details
- Promises, commitments and anything phrased as an assurance
- Everything else, read normally
Checking factual detail
Open the source for every figure. A number with no source behind it is either your own record, in which case confirm it, or an invention, in which case remove it.
Confirm credentials with the body that issued them rather than from memory. Registration numbers, expiry dates and the states or counties covered are all easy to state slightly wrong.
And check the coverage list against reality. A generated service area often includes towns that sound right and that you do not actually serve, which produces wasted calls at best.
Checking the pages describing your process
These read convincingly and are frequently wrong, because a tool describing how your job runs is producing a reasonable version of how such jobs generally run.
Read them as a customer who is about to rely on them. Does the first visit really work like that. Do you really provide that document. Is that really the order things happen in.
Rewrite rather than edit where the description is wrong in substance. Correcting a plausible invention sentence by sentence usually takes longer than writing the paragraph yourself.
Structural checks that are not your job
Heading order, link targets, alt attributes, page titles, descriptions and duplicate content are all mechanical, and checking them by reading is a poor use of your time.
Run them instead. Our site check reports the structural faults across a whole site in one pass, which leaves you with only the questions a person can answer.
That division is the point. The mechanical half should be automatic so the judgment half remains possible, particularly on a site of more than a handful of pages.
Style, last and quickly
Read for voice at the end, and only on the pages that matter most. The homepage, the main service pages and anything a customer reads before calling.
Cut the openings that establish the subject matters. Cut the sentences a competitor could publish. Our page on AI website copy covers where the effort actually pays.
Then leave the rest alone. Educational pages that are accurate and plain do not need a distinctive voice, and time spent polishing them is time not spent on the pages that earn money.
Signing off
Record who reviewed what and when. On a small site a single file is enough, and it converts a claim that your content is checked into something you can actually show somebody.
Put a review date on the pages that age, and honor it. Our page on content freshness covers the routine, which matters most for prices, hours and coverage.
Then publish. A reviewed site with a few plain pages beats an unreviewed site with fifty polished ones, and the review is the only part of this process that cannot be delegated to anything.