How We Built a Removals Site ChatGPT Rates 99/100 (And the Three Things That Won Weren't SEO)

In April we wrote about a removals firm in Brewood, Staffordshire, whose website had quietly stopped working. A year of AI-generated "moving tips" posts had buried it: 220 of its 272 pages refused entry to Google's index. We promised a rebuild. It is now live at brewoodremovals.co.uk, and we did something we have never done with a client site before. We asked OpenAI's ChatGPT to score the finished build against the websites it considers the best in the UK removals trade. It put a one-van Brewood firm on 99 out of 100, ahead of Pickfords, White & Company and Bishop's Move. This is the build teardown: the three decisions that actually earned those marks, and why none of them are the things SEO blogs obsess over.

The 5.0-rated firm Google ignored

Brewood Removals is run by Sean Hamilton, who answers every quote, viewing, call and email himself, day or night. The moves are done by an insured crew led by James, who has more than fifteen years on the vans. The firm has a perfect 5.0 rating from 74 Google reviews. By any human measure it is a good business that does good work.

The old website did not reflect any of that. It was WordPress with a page builder, and at some point the content strategy became "publish more". Around 160 near-identical /removals/{suburb}/ doorway pages. Sixty-nine thin blog posts with titles like "The No 1 Best Man and Van". 247 URLs in total, and a grand total of zero page-one rankings for any "removals company {town}" search that mattered. We documented the full diagnosis in the companion piece on what Google actually rewards now. Great reviews, invisible site. The two facts sat side by side for three years.

So the rebuild was not a redesign. We dropped the lot. Every blog post gone, the doorway sprawl gone, 247 URLs collapsed to about 25 real pages, with explicit 410 Gone responses telling Google the junk was deliberately retired rather than moved. Our hosting sister company 365i told the deletion side of that story from their angle in their own case study on the rebuild. This post is about what we put back, and which parts of it the machines rewarded.

We let ChatGPT mark our homework: the AI visibility test

When the build was finished we ran an experiment. We asked ChatGPT to rate the new Brewood Removals site against the websites it considered the strongest in the UK removals sector, scoring six dimensions: SEO, E-E-A-T, non-commodity content, advice content, useful tools, and an overall mark. We did not tell it which site was ours. This is really a test of generative engine optimisation (GEO), the discipline of getting recommended by ChatGPT and the other AI assistants rather than just ranked on Google. Here is what came back.

A one-van firm in a Staffordshire village, with no backlinks worth the name and a domain that was four days old when we tested it, scored higher than a brand most of the country could name. The honest caveat first, because it matters: this is the model's assessment of on-page quality, not a Google ranking. Nobody should read "99 on ChatGPT" as "number one on Google". But it is a clean read on whether a site is well-built, well-evidenced and well-structured, judged by exactly the kind of AI assistant more and more people now ask "who should I get to move my house?". And the gap was not in the obvious SEO column. It was in non-commodity content, advice content and useful tools, the three columns where the national brands scored their lowest. Those three are the ones we engineered for, and they are what the rest of this post is about.

The performance side was not a trade-off, either. The same build hits a perfect score on Google's own tooling.

Win one: an editorial standards page AI can read

The first decision that moved the needle is the one most agencies would skip as a legal afterthought. We built Brewood Removals a public editorial standards page, and we treat it as a load-bearing part of the site, not boilerplate.

It answers, in plain English, the questions a sceptical reader and a cautious AI both ask. Who writes this site? Where do the facts come from? How are pages produced and reviewed? How are dates and update logs handled? What happens when something is wrong? It even has a section on privacy in the stories we tell, because the site uses real customer moves and we are explicit about how those are anonymised. Sean is named as the person accountable for all of it.

Why does a removals firm need this? Because Google's Search Quality Rater Guidelines push raters to look for exactly this transparency, and because AI assistants reason about whether a source is trustworthy before they cite it. An editorial standards page is the cheapest, clearest way to say "a real, named human stands behind every word here". Every advice guide on the site carries a visible byline, a published date and an "Editorially reviewed by Sean Hamilton" line, and the author box links straight to the standards page. The claim and the accountability are never more than a click apart. We do the same thing on this site: you will find our own editorial standards linked in the footer of every page, for the same reason.

Win two: a free moving-planner tool that earns its keep three ways

The second decision was to build one genuinely useful free tool rather than ten more pages. The Brewood moving planner turns a move date and a few details about your situation into a personalised, tickable, week-by-week countdown.

The detail is where it gets good. You can print it as a tidy job sheet to stick on the fridge. You can email a copy to yourself. You can export your plan to a file and import it back later, so it survives a new phone or a cleared browser. And it holds your data in your browser's local storage, never on a server, so it is completely private. It works with JavaScript switched off. There are no prices anywhere on it, on purpose, because a planner that pushes quotes stops being a tool and starts being a funnel.

One tool, three jobs. It is a real lead magnet, because the optional "email me my plan" step puts an enquiry straight in Sean's inbox. It is the site's internal-linking hub, because real planner stories from past customers surface on the planner, on every area page, and on the homepage, weaving the whole site together. And it is non-commodity proof in its own right, because no rival's "10 tips for moving house" listicle does anything for the reader the way a working, private, printable planner does. ChatGPT gave the site 99 on useful tools. This is most of why.

Win three: one data layer feeding the page, schema and AI files

The third decision is the unglamorous engineering one, and it is the one I would argue matters most over time. Every fact on the Brewood site lives in one place: a structured data layer. The visible page, the JSON-LD schema, and the AI Discovery Files all render from that single source.

This sounds like a tidiness preference. It is actually a trust mechanism. On most sites, the visible content, the schema, and whatever the business told an AI tool last year drift apart over time. A price changes on the page but not in the structured data. The schema claims a service the firm dropped. The result is a site that contradicts itself, and contradiction is precisely what a cautious ranking system or AI assistant reads as a reason not to trust you. When the page, the schema and the machine-readable files cannot disagree by construction, they reinforce each other instead.

It also let us scale the non-commodity content honestly. Sean handed us his real work diary, and we ingested 131 dated jobs into the data layer: 40 across the towns, 40 across the services, and 51 customer stories from people who used the planner. Crucially, those stories are visible copy attributed to a family surname only. They are never dressed up as review schema or star ratings. The star ratings on the site come solely from the 74 verified Google reviews, rendered verbatim, never curated, with the single four-star review left in. That guardrail is the difference between using real customer stories at scale and committing review fraud, and we built three automated checks to enforce it.

"Focus on making unique, non-commodity content that visitors from Search and your own readers will find helpful and satisfying."

John Mueller, Google Search Advocate (21 May 2025)

When I first read that line, the word that jumped out was "unique". Not "more", not "optimised", not "polished". Unique. A data layer fed by one man's actual diary is about as unique as web content gets, because nobody else has that diary. The engineering does not create the uniqueness, the owner does. What the engineering does is make sure the uniqueness reaches the page, the schema and the AI files without leaking away in the gap between them.

Why a village firm outscored Pickfords

Here is the part that surprises people. The national brands did not lose on polish. Pickfords has a slick site and decades of authority. They lost on the columns that reward specificity, and they lost them because of how they have to write.

A national brand writes for everyone, so it cannot name anyone. Its "removals in Wolverhampton" page has to read the same as its "removals in Wolverhampton, the other one" page, because a template fills both. It cannot tell you about the tight access on a particular road, or the grandfather clock that was the heaviest lift of the month, or what a two-bed move in that town actually costs, because head office does not know and the template has no room for it.

Brewood Removals can write all of it, because Sean did the work and James's crew lifted the boxes. That human texture is the single thing a national brand's content cannot copy without changing how it actually operates. It is also the thing the model rated highest. Lily Ray, VP of SEO Strategy and Research at Amsive, put the wider pattern plainly this spring:

"Google recently reminded everyone that 'non-commodity content' is what wins in search, meaning content with a unique perspective, real E-E-A-T, and genuine lived experience."

Lily Ray, Amsive (30 April 2026)

Genuine lived experience is exactly what a small firm has in abundance and a national chain has to manufacture. We poured the build budget into capturing it: exactly ten hand-written area pages, a deliberate cap rather than 160 doorway clones, and a service page for every job the firm actually does, each one carrying its own real job cards. Here is the live build, if you want to inspect it.

The advice cluster is the same idea at length. Fourteen guides, each one passing a full E-E-A-T audit, each grounded in how the firm actually works rather than scraped from the search results. Where the removal costs guide really lands is that it publishes actual 2026 price ranges, openly, while most of the trade hides everything behind a quote form.

The AI visibility layer: AI Discovery Files and the directory

The last layer is the one built specifically for the machines, and it is where my colleague Mark McNeece's work feeds directly into the build. Brewood Removals ships the full suite of nine AI Discovery Files at its root: llms.txt, llms.html, ai.txt, ai.json, identity.json, brand.txt, faq-ai.txt, developer-ai.txt and robots-ai.txt. They follow the ai-visibility.org.uk specification, and because they render from the same data layer as the visible site, they cannot contradict it.

The payoff is verifiable. Brewood Removals holds a Platinum 10/10 listing in the AI Visibility Directory, verified on 13 June 2026, with full conformance and all the Discovery Files found. Mark, who created the AI Visibility Definition standard, frames the whole exercise in one line that has shaped how we build:

"AI Visibility is not about being found. It is about being understood."

Mark McNeece, AI Visibility Explained (April 2026)

That distinction is the whole game. The old Brewood site was findable, in the sense that Google had crawled 272 URLs of it. It was not understood, because nothing on it told a machine who the business was, what it did well, or why to trust it. The rebuild flips that. If you want to check your own site the way we check ours, both tools are free: the AI Visibility Checker grades how readable your site is to AI, and the 365i AI Bot Checker tells you which AI crawlers your server is actually letting in. We run both on every site we ship.

What this proves about ranking in ChatGPT, and what it does not

I want to be straight about the limits, because a case study that only sells is a case study you should not trust.

A 99 on ChatGPT is not a Google ranking. It is a strong read on build quality, and build quality is necessary, but Google also weighs links, reviews and local prominence that a four-day-old domain has not banked yet. We saw exactly this with the same owner's locksmith business: when we built Lockerfella to the same standard, it hit number one on ChatGPT and Gemini fast, while the bare head terms on classic Google took longer because incumbents had years of local signals. Two months on, that locksmith is page one on Google for its city term and recommended by ChatGPT in its home towns, which is the kind of follow-up this removals story will get too; the full data is in our case study on whether ChatGPT recommends local businesses. The honest version of this removals story has its own sequel, and we do not have it yet.

The site went live on 12 June 2026. Search Console has barely started gathering data. The 410s on the old doorway pages need weeks to process, and the new pages need time to be crawled, understood and ranked. We are publishing the build now because the build is finished and verifiable today; the ranking and lead figures come in a follow-up once the data is real. You can hold us to that.

And content pruning is not universal advice. It worked here because the deleted pages were genuinely worthless, the kind Google was already refusing to index. If your blog earns clicks, do not take a hatchet to it on the strength of someone else's case study. The pattern that travels is not "delete things". It is the one this whole post has been about: be specific, be accountable, build one source of truth, and give the machines something real to understand. That is what the recent core updates keep rewarding, and it is the opposite of what got the old site buried.

Frequently Asked Questions