The standard
What every site we ship has to do
Five brands sell different things, but a site leaving any of them has to clear the same bar. These are the rules, and the evidence that they were actually followed rather than merely written down.
- PageSpeed target 100, mobile
- Font budget 95 KB, three files
- Third-party scripts Zero
- Old URLs decided 475 of 475
Speed
It has to be fast, and the number has to be checked
Every build targets 100 on Google PageSpeed across all four categories, on mobile, measured after launch rather than promised before it. Mobile, because that is the harder test and the one most visitors actually sit inside.
The clearest case is Lockerfella, a locksmith site that went live in April 2026 on a brand-new domain with no backlinks. It measured 100 / 100 / 100 / 100 on mobile from launch day, with a 28ms server response, on the standard Linux hosting plan rather than the expensive tier. It is hand-coded PHP, CSS and vanilla JavaScript, which is most of the explanation.
The Lockerfella case study carries the figures and the date they were taken.
URLs
Every old URL gets an explicit decision
When a site is rebuilt, every URL on the old one is given a disposition before anything ships: kept, redirected, or retired with a 410. Nothing is left to a catch-all rule, and nothing is silently dropped, because a URL that quietly 404s is a decision nobody made.
Brewood Removals had 247 pages, roughly 160 of which were the same page wearing a different suburb's name. 225 were retired outright, 18 redirected and 3 kept, leaving about 25. The case study has the counts.
This site went through the same process: 475 URLs, of which 449 now return 410 Gone, 22 redirect and 4 were kept. A harness asserts all 475 still return exactly what they were assigned, so the rules and the intention cannot drift apart.
That harness found nothing wrong, and it could not have. The 475 came from walking the old
site's sitemaps, and WordPress lists no date archive in a sitemap: not the year, not the
month, not the day. Google surfaced /2025/03/12/ on 28 July 2026, two days after
the rebuild went live, and it returned a 404 nobody had chosen. So did every feed, every paged
index and every querystring permalink. The count was never wrong. The method that produced it
could not see this whole class of URL, and a second harness that probes shapes rather than a
list now covers 49 of them.
They are handled by pattern rather than by count, because a post permalink here is
/YYYY/MM/slug/ and carries no day, so which days had posts is not recoverable
from anything we hold. The rule covers every day in each month that had a post, which claims
slightly more than we can prove and is the honest trade: the alternative is guessing.
The admission
No page generator a client has to feed
Worth saying who built those 160 Brewood pages. We did. The old site ran on a location system we wrote for it, and the arrangement was that the client would give each town genuinely different content. We explained that, showed how it needed doing, and handed the writing over. It never got done.
That is the outcome the arrangement produces nearly every time, and not admitting it is roughly how the industry filled up with doorway pages. A generator whose output is only safe if somebody sits down and writes 160 unique pages is a machine for producing near-duplicates, and the decision to ship one is ours rather than the client's.
So the rule now runs the other way. Area pages are written by hand before launch or they do not exist. The Brewood rebuild carries exactly ten, each grounded in real dated jobs, against roughly 160 on the site it replaced. Nothing goes live waiting for content that has not been written yet.
Colour
Colour is computed, never eyeballed
Every colour pairing on every site is measured against WCAG 2.2 before it ships. Not sampled by eye, not assumed from a palette generator.
This page is the example. The company's brand orange is #FF7F00, taken from its own
icon. Against white it measures 2.53:1, which fails outright, so an earlier version of this
website could never use the company's own colour as text and came out grey as a result. Against
the near-black this site now uses it measures 7.86:1, which is AAA. That single measurement is
why the site has a dark background: inverting the ground is what made the brand colour usable.
The same discipline decides smaller things. Dark text on that orange is 7.68:1, so the headline in the middle of the homepage grid is legal. Four of the five brand colours take white text and the fifth, AI Visibility's amber, takes dark text, because white on it is 3.88:1 and dark is 5.42:1. Those pairings live in the site's data rather than in its stylesheet, so a future brand colour cannot quietly fail.
Fonts
Fonts are served from our own server
No site in the group loads fonts from a third-party font CDN. Two reasons, and the second one matters more. It costs an extra connection and a render-blocking round trip on the critical path. And it hands every visitor's IP address to another company, which for a UK business is a GDPR problem nobody needs.
Self-hosting also forces a budget. This site runs three variable font files totalling 95 KB, all subsetted. A fourth family is a decision made deliberately with the cost understood, not something that accumulates. The budget is a ceiling rather than a target: this site went from 76 KB to 95 KB in July 2026 when we moved it onto the same faces as one of our own client builds, and that was a deliberate 19 KB rather than an accident.
Machines
The site explains itself to machines
Every site ships a full schema graph and a set of AI Discovery Files: machine-readable statements of what the business is, what it sells and who is behind it. This is what AI Visibility exists to specify, and the group implements its own specification.
"Who is behind it" means a named person, not a company name standing in for a human. Every site declares the person who runs the business as a node in its own right, and where that person has public identifiers the node links out to them. Lincolnshire Chauffeur Services is the clearest case in the group: Bart Griemink kept goal for Birmingham City, Peterborough United, Swindon Town and Southend United, so he has a Wikipedia article and a Wikidata entity, and his site points at both. Anyone, human or machine, can follow those and arrive somewhere we cannot edit. That is the entire value of the property, and it is why a Person node with no outbound identifiers is barely worth shipping.
Whether anything actually reads those files is a fair question, and the honest answer is that most
published llms.txt files never get fetched at all. Over 30 days on mcneece.com, an
active site with a complete and valid set, the logs recorded 73 file accesses from 11 distinct AI
crawlers, including OpenAI, Anthropic, Google, Microsoft and Apple. Nine of the ten files were
fetched at least once. One site is not proof of a trend, and
the write-up
says so plainly alongside the studies that found the opposite across hundreds of thousands of
domains.
The limits
What this standard does not cover
Three honest limits, because a standard with no exceptions is a marketing page.
It applies to new builds, not to everything we host. The group has hosted client sites for over twenty years, and plenty of them were built on page builders in earlier years, to earlier standards. Those sites are maintained, patched and kept secure. They are not quietly rebuilt to make a portfolio look consistent, and a healthy site that a client is happy with is not a problem that needs solving.
Not every site hits 100 on every category. A hand-coded brochure site on our own hosting is the easy case. A WooCommerce store with a payment provider, a review widget and a booking system is not, and pretending otherwise would be the kind of claim this page exists to avoid. The target does not move; what changes is how much of it a given stack allows.
Third-party scripts are the client's call. If a business needs a particular chat widget or analytics tool, we will say what it costs in performance and then fit it properly. We do not get to veto it.
Who builds to this
All five brands, on every site they ship
All five names, but not five teams, and on a page about standards that distinction is the whole point. 365i provides the platform every one of these sites runs on, which is a genuinely separate discipline. Press Forge, 365i Web Design and McNeece all build to this standard, and they are three front doors onto one team rather than three studios with three different bars. AI Visibility writes the specification the machine-readable half of this page refers to.
One team is also why this page can exist. Five independent agencies would need a governance process to agree a standard and some way of policing it between them. This is one set of hands with one set of habits, which is a smaller claim than it might look, and a more reliable one.
You can see the results on the work page.