Making the machines read it right.

British Manufacturing Group

Client

British Manufacturing Group

Sector

Manufacturing & Industrial

Discipline

AI Search

Outcome

Read correctly by machines, without the keys

Client name and identifying details withheld for confidentiality.

The problem on the table.

AI systems and search engines read the raw HTML a server sends, before any script has run. That single fact decides where AI readability work has to live, and it rules out most of the places it is convenient to put it.

This group had a further constraint. The websites are built and maintained by an external development partner, so the content management system was not ours to edit. Everything capable of moving the measure sat in a part of the page that reaches the site through somebody else's release process.

The tempting shortcut is to inject the markup client-side through a tag manager, which was available. That helps a rendering crawler and does nothing for the systems reading the raw response, so it would have produced the appearance of progress and none of the substance.

What forensic looked like here.

The work was reorganised around what could actually be deployed. Every change that moves machine readability is authored and validated here, then handed over as a small self-contained block to paste into the server head, pre-tested so that deploying it is a two-minute job for whoever holds the keys.

The first block established identity. The group trades under several brand names on separate domains, and to a machine those looked like unrelated companies. Each brand's block declares its own registered legal name alongside one shared company-register identifier, so the markup states the route from any single site to the company behind all three.

The external profiles each brand claims were verified one by one before being declared, rather than listed on the assumption that they exist, and every block was validated in the structured-data testing tools before it was sent.

Two things were deliberately left out of what we authored. No self-declared rating markup, because the search engines' own policy treats that as self-serving on a company's own pages. No sitewide search declaration, because the feature it once earned has been retired and a declaration that buys nothing is still a declaration to maintain.

• Entity and organisation markup for every brand site, authored and validated here
• One shared company-register identifier across every brand's block
• Verified external profile declarations, each checked before it was claimed
• Copy-paste server-head blocks, pre-validated, sized to deploy in minutes
• A staged twelve-month plan, sequenced around the access that exists

What changed.

Every brand site now carries organisation markup in the raw HTML that AI systems and search engines actually read, rather than injected by a script in a place those systems never look. That single decision is what makes the rest of the work count, and it is the one most commonly got wrong.

Just as important is how the engagement runs. The plan concentrates the work that depends on an external release early, so the later phases rest on content and authority work that can proceed independently. Success is stated as whether the brands are cited in AI answers, which can be measured, rather than as a rank, which cannot.

A handover is not a deployment, so the live pages are re-read rather than trusted. Every block is validated before it leaves the practice, then the deployed page is measured against it, because a block retyped into somebody else's template can differ from the one that was sent in ways that stop it working while still looking right. Finding that while the engagement is running is the point of measuring.

• Server headwhere the work has to live
• Validatedbefore a developer sees it
• One identityacross every brand site
• 12 monthssequenced around access

Other engagements.