In September I published a version of this post that ran Google's agent-friendly checklist against my site and put a tick next to every item. It was true, but it was a list of ticks, and you had to take my word for it. That's not what anyone pays £150 for.
So on 4 November 2026 I ran the full audit on my own site, the way I'd run it on yours: every page measured, 20 Lighthouse runs, the design taken apart, and the whole thing compared with the four web designers that rank first for "web design Dover". The full report is published, and it's what you'd get.
What the audit covers
Seven areas, each scored out of 100 by published rules, the same way for my site and for every competitor:
| Area | What's measured |
|---|---|
| AI and agent readiness | Words in the raw HTML, what GPTBot is actually sent, robots.txt, llms.txt, structured data, landmarks, tap targets |
| Search visibility | Positions for local searches, titles, descriptions, headings, canonicals, sitemap, broken links |
| Speed | Lighthouse on a simulated phone and on desktop, page weight, what loads from other servers |
| Accessibility | Contrast of every text element, names and labels, tap targets, alt text |
| Design | Colour palette and how much of the page each colour takes, type sizes in use, line length, an expert review |
| Content and conversion | Contact in the first screen, phone, form, prices, reviews, reading ease |
| Trust and security | Security headers, business details in structured data, ratings |
How it scored
Mine came out highest of the five: 84, against 81 for the best competitor and an average of 68. That sounds like a good result, and it's the wrong thing to take from it. The scores say the site is well made. The search snapshot says nobody can find it.
The problem the code can't fix
I searched "web design Dover", "website designer Dover Kent small business" and a longer variant, set to Dover. All four competitors appear in every one, in positions 1 to 9. rgleeson.com isn't in the top 20 of any of them.
It gets worse with the brand search. My Google Business Profile trades as RG Web Design. Search "RG Web Design Dover" and result 1 is rgwebdesign.co.uk, a different web design business in Sussex. My site never uses the name at all. To Google and to an AI assistant, my listing's name points at someone else.
Both are critical findings, and neither shows up in a technical checklist. They're why the audit includes a search snapshot and a competitor benchmark, not just a crawl.
Where the site is strong
The AI and agent readiness checks are close to perfect, which is what the September post was about, now measured properly:
- All 32 pages carry their text in the HTML: 23,450 words a crawler can read without running a script.
- 160 of 160 fetches as GPTBot, ClaudeBot, PerplexityBot and others got exactly what a browser gets.
- 1,456 controls checked at two screen widths, none without an accessible name.
The competitor benchmark turned up something I didn't expect. Two of the four competitors allow AI crawlers in their robots.txt, then refuse them at the server: one answers GPTBot with a 429, the other drops the connection. Their owners almost certainly don't know. To ChatGPT, those sites don't exist.
Speed: fine on desktop, slow on a phone
On desktop every page scores 93 to 99. On a simulated phone on slow 4G, the main content takes 3.5 to 5.4 seconds to appear. Google's line for "good" is 2.5. Competitor C manages 2.0.
The pages aren't heavy, so the report digs into why. Before anything paints, the browser has to reach five servers: mine, two for one font and two for the other. Google Analytics is the largest file on most pages. And the hero fades in word by word, so the main content doesn't count as shown until a script has run. Self-hosting the fonts and showing the first screen without the animation is half a day's work.
The design review
This is the part most audits skip, and it's where I spend the most time, because it's what a visitor notices first. The report extracts the palette from the rendered page and measures how much of the page each colour covers, counts the type sizes in use, measures body text and line length, and compares all of it with the competitors.
| Measure | rgleeson.com | Best of the four competitors |
|---|---|---|
| Typefaces | 2, plus a monospace in one illustration | 1 |
| Type sizes on one page | 14 | 8 |
| Body text | 16px, 1.6 line height | 19px, 1.6 |
| Characters per line | 47 | 49 |
| Text failing contrast | 1 of 141 | 0 of 55 |
| Expert review, out of 20 | 15 | 17 |
Competitor C has the strongest page of the five: huge, confident type and "5.0 on Google · 65 reviews" in the first screen. Mine has no proof in the first screen at all. Its one testimonial is halfway down. The report's design recommendations are small: more of the text in ink rather than grey, five stray type sizes folded into the scale, a line of proof under the headline, and no animation on the first screen.
The findings
Twelve, graded by three questions: does it stop someone doing what they came for, does it affect many people, will it keep happening? Three yeses is critical.
| Severity | Finding |
|---|---|
| Critical | Not in the top 20 for "web design Dover" |
| Critical | The Google listing's name is missing from the site, and another business owns it |
| Serious | Main content takes 3.5 to 5.4 seconds on a phone |
| Serious | No rating or review count on Web Services |
| Serious | An AI agent can't book a call |
| Serious | 15 descriptions and 7 titles too long for search results |
| Medium | Six more: structured data on half the pages, image weight, heading order, six small tap targets, no Content Security Policy, a double redirect |
Seven of the twelve are under an hour each. The report sorts them into this week and this month.
What it doesn't cover
What ChatGPT, Perplexity and Claude actually say when asked about a business needs real accounts, so I'm running those questions separately and will add the answers here, whatever they say. Search Console and Google Business Profile numbers need the owner's login: a client's report includes them, this one can't.
What you'd get
The sample report is the deliverable: a scored summary, your site against your local competitors (named, with screenshots), search, AI readiness, speed, accessibility, a design review, every finding with its fix, and an action plan, as a web page and a PDF, with the raw measurements for your developer. The audit is £150, credited in full if you go on to a build with me.