Content review policy
Nothing here reaches search until a person has signed it off
The review a page goes through before it can be indexed: the workflow states, who checks what, the twelve dimensions we score, the higher bar for advice that touches money or health, and the register where a named person records approval.
The sequence
The states a page moves through before anyone can find it
Every page carries one of these states in the content model. The state decides what happens to the page — it is not a note in somebody’s inbox.
Brief
Nothing is drafted without one. The brief names the reader, the intent, the objections the page has to answer and the claims it is forbidden to make. A page whose brief is incomplete fails validation before a word of it is written.
You get: An approved brief, stored with the page
Draft
One person writes the page against that brief. Drafts are written to be argued with, so the first internal question is always which sentence in here could not be defended if a client asked where it came from.
You get: A complete draft, never a skeleton to be filled in later
Subject-matter check
Someone who knows the domain reads it for accuracy. On a technical search page that means checking the behaviour described is still how the platform works. On an industry page it means checking that the buying process described is how that trade actually operates.
You get: Corrections, or a note that a claim needs a source
Search and conversion review
Two separate passes. The search pass checks intent match, cannibalisation against pages we already have, headings, metadata, internal links and whether the structured data matches what is visible. The conversion pass checks the page has an audience, a problem, a difference and one obvious next step.
You get: A page that answers the query and asks for something
Legal or compliance review, where it applies
Triggered by commercial commitments, regulated sectors and anything that could be read as professional advice. Our own guarantee page sits in this state until a lawyer has read it, which is why that is recorded in the content rather than assumed to have happened.
You get: Sign-off, or wording that has to change before it can proceed
Final QA
The whole gate at once: uniqueness measured against every other page on the site, factual checks, accessibility, mobile rendering, performance, broken links, correct status codes, canonicals and schema validity.
You get: A pass, or a list of specific blockers
Human approval
A named person reads the finished page end to end, on a phone and on a laptop, and records approval with their name and the date. Until that record exists the page is live, linked and deliberately withheld from search.
You get: An entry in the approval register
Refresh or retire
Published pages are re-read on a schedule and whenever the ground moves under them. A page that can no longer be defended is either updated or withdrawn — leaving it up untouched is not one of the options.
You get: A current page, or an honest absence
The gate
Why a page is not allowed to approve itself
Approval is not a field an author fills in. In the content files those fields are always empty. The record of who approved what lives in a separate register that a person edits by hand, and the two are merged only when the site is built.
The effect is that publishing becomes a deliberate act. A route can exist, render perfectly and be linked from the navigation while remaining invisible to search engines, because the only thing that makes a page eligible for indexing and for the sitemap is a human name and a date sitting beside its route.
Deleting that line withdraws the page again on the next deploy, without removing the content or breaking the link. That mechanism is what sits behind the ordinary-sounding sentence "we withdraw content we can no longer stand behind".
It is worth naming the cost. This is slower than publishing on merge, and it means the site goes live with fewer indexable pages than it has routes. That is the direction we would rather be wrong in.
The higher bar
Where being wrong could cost someone money, health or a legal position, a qualified person reviews it or it does not publish.
Most of what we write costs money when it is wrong. Some of it costs more than that. Pages touching healthcare, legal, financial and insurance marketing sit next to regulated advice, and a marketing claim in those categories is easily read as guidance about the underlying service.
Two rules follow. Marketing advice is separated from professional advice explicitly, in the page and not in small print. And where a page describes what a regulator, a professional body or a platform policy requires, it cites that body directly rather than paraphrasing a competitor’s summary of it.
If we cannot get competent review for a claim in those categories, the claim is removed. A page that is shorter and correct is not a compromise.
Roles
Who checks what, and what each of them is looking for
The brief owner
Decides the page should exist at all, then defines the reader, the proof it needs and the claims it may not make, before anyone drafts anything. Most bad pages are bad because this step was skipped, not because the writing was weak.
The subject-matter reviewer
Reads for accuracy inside the domain. On a technical page that means confirming the described behaviour is current rather than two platform changes out of date. On an industry page it means confirming the described buying process is real.
The search reviewer
Checks the page answers the query it targets, does not compete with something we already published, and is technically clean: canonical, metadata, heading structure, internal links, and schema that matches visible content.
The conversion reviewer
Checks the page is honest about who it is for and clear about what happens next. A page that reads beautifully and asks for nothing has failed, and so has a page that asks on every screen.
The approver
Reads the finished page the way a stranger would, then puts their name against the route. They are the person who has to answer for it afterwards, which is the entire point of the role.
Scoring
The twelve dimensions every page is scored on
Each is scored from zero to five before publication. A priority page has to average four or better, and nothing publishes with a critical dimension below three. The floor is the part that matters, because it stops a strong average carrying a page that is simply wrong.
- Intent fit — does this answer what the person actually searched for
- Buyer specificity — could this page have been written about any other audience
- Originality — does it say anything the pages already ranking do not
- Expertise — does it show operating knowledge, or summarise other people’s
- Clarity — can a busy reader get the answer without decoding it
- Persuasion — does it make an argument, or list features
- Factual accuracy — has every checkable claim actually been checked
- Search coverage — does it answer the question fully, including the awkward parts
- Internal linking — is it connected to the pages this reader would want next
- Visual quality — has anyone thought about this page specifically
- CTA fit — is the next step right for where this reader actually is
- Technical SEO — canonical, metadata, headings, status codes, structured data
Failure states
What stops a page publishing, or takes it back down
The first four are conditions the build treats as blockers and checks on every deploy. The last two are editorial judgements a person makes about a page that is already live.
- No human approval recorded
- The most common blocker, and the intended one. Until someone has added the route to the approval register, the page renders with a noindex directive and is excluded from every sitemap. There is no override and no expiry.
- Fewer than three real sections, or no internal links
- A stub is the failure mode this whole system exists to prevent. A page with too little substance to be useful, or with no onward path for the reader, is held back rather than published thin on the promise of improving it later.
- A figure with no source behind it
- Any unsourced number fails validation outright. The fix is either a link to the primary source or the removal of the number, and in practice it is far more often the second.
- Too close to another page on this site
- Every page body is compared against every other one. High overlap means somebody produced a page by adapting a neighbouring one, which is precisely the thing we refuse to do at scale, and it is caught before publication rather than after.
- A claim we can no longer defend
- Published pages come down for this. A platform changes, a cited source is withdrawn, or we simply learn we were wrong. Withdrawal is a one-line change and takes effect on the next deploy.
- The page stopped being true while nobody was looking
- The quiet failure, and the one most publishers never address. A page that was accurate when it was approved and is now well out of date is not neutral — it is misinformation with our name on it.
Questions
What people ask about this process
Who approves a page for publication?
A named person, recorded with the date against that specific route. Not a team, not a job title, not a process, and never an automated check.
The approval lives in a separate register from the content itself, so a page cannot approve itself. Nothing is eligible for indexing or for the sitemap until that entry exists, which means an unapproved page is live and readable but invisible to search engines.
What happens to a page that fails review?
It keeps its current workflow state and stays out of search. There is no partial publication and no route by which a page that has not cleared the gate goes live to be improved later.
Most failures are unglamorous: a claim needs a source, two pages are competing for the same query, or a section turned out thinner than it looked in outline. The page goes back to whoever owns it with the blockers listed specifically.
How do you decide a published page needs updating?
Three triggers: something in the world changed, something in the underlying data changed, or the page has aged past the point where we would defend it without re-reading it first.
The first is the one that matters. A platform changing how it works can turn an accurate page into a misleading one overnight, and no calendar catches that — somebody who follows the area has to notice and flag it.
What counts as YMYL content for a marketing agency?
Anything where acting on our writing could affect a person’s money, health or legal position. In practice that means our healthcare, legal, financial and insurance pages, plus any page describing what a regulator or professional body requires.
Those pages get subject-matter review before publication, cite the governing body directly, and separate marketing advice from professional advice in the page rather than in a footnote. Where we cannot get competent review for a claim, the claim comes out.
Does an approved page ever get withdrawn?
Yes, and it is deliberately easy to do. Removing the route from the approval register takes the page out of search on the next deploy without deleting it or breaking a single link to it.
We would rather withdraw something quickly and reinstate it after a rewrite than leave a page we can no longer defend collecting traffic while people argue about whether it is bad enough to touch.
Hold this policy to its own test
If a page on this site says something you can disprove, tell us. A review process that has never rejected anything is not a review process, and one that nobody outside the company can test is worth less than it looks.
Last updated · Reviewed by Zubair Afzal