The rules are on the screen while you write, not in a report afterwards.
58 checks, evaluated live against the draft. 26 for classic search, 16 for being cited by an assistant, 11 for being extracted into an AI Overview. Every one of them is a rule, not an opinion.
An audit after publication is a list of things you will not go back and fix.
The usual shape is that somebody writes a piece, publishes it, and a tool later produces a report of what was wrong with it. That report arrives after the moment when fixing anything was cheap, and it competes for attention with the next piece, which is already late.
The second problem is that most of those reports are graded by a model, so the same article scores 74 on Tuesday and 81 on Thursday. A score that moves when the text did not is a score people learn to ignore, and once ignored it is worse than absent because it consumed the space where a real standard could have been.
A standard only works if it is in front of the writer and it is the same every time.
One writing surface, three standards, no ambiguity about any of them.
- 1
Write, or start from the draft
Docket produces a full first draft from the brief with headings, links, an FAQ and imagery. It is a normal editor from there and the words are yours.
- 2
The checks run as you type
58 rules, grouped and scored separately, each naming what is missing and where rather than reporting a grade.
- 3
Readability is graded too
With the sentences that hurt it marked. Dense prose costs you with a person and with an extraction model, for the same reason.
- 4
Fix with tools, and keep the old version
Rewrite a section, tighten a meta description, change tone. Every rewrite snapshots what was there first, so nothing is one click from lost.
A standard a team can actually hold each other to.
Every line here describes something the software actually does. If one of them turns out not to be true of your setup, that is a bug and we want to hear about it.
The same article always scores identically.
Every check is deterministic. No model grades the output, so two editors comparing a draft are comparing the draft rather than two model runs.
The three standards are scored separately, not averaged.
A piece can be excellent for classic search and useless to an assistant. Rolling 58 checks into one number would hide exactly the gap worth knowing about.
Each failing check says where the problem is.
Not 'improve internal linking' but which paragraph has none, and which of your own pages would be the sensible target.
Metadata is graded and then actually published.
The meta title and description the checklist grades are pushed to WordPress with the post. They used to be graded and then dropped on the way out, which made three of the checks decorative.
Every rewrite is reversible.
Version history snapshots the previous wording before any AI edit, capped at twenty per article, so an experiment costs nothing.
What the checklist is not
- It is not a quality judgement. It measures structure, not whether the piece is worth reading, and a boring article can score well. Nothing can automate the second thing.
- It is advisory. It will not block a publish, by design.
- It is not a fact checker. It does not verify a claim in the text against the world, and a confident sentence with a wrong number scores exactly as well as a right one.
- It is not an originality check. It does not compare the draft against the rest of the web for overlap.
Answers, Not Hedging.
The parts either side of this one.
Twenty minutes, your website on the screen.
Not a slide deck. We run the checks against your actual domain before the call and spend the time on what came back. If it turns out you do not need this, we will say so.