Case study, AI readiness and technical SEO
A transport firm whose home page carried eleven words of HTML: 1,123 words after the fix, clicks up 38% in the next thirty days
A removals and transport firm had a single-page React site built by another developer. Google ranked it well, because Googlebot renders JavaScript; every crawler that does not render received eleven words. The words went into the HTML in a day. The month after, the clicks rose. This page says what moved, what did not, and which of two changes deserves the credit, which is: we cannot tell, and we say so.
Written by Daniel Stoychev, who did the work. Measured with the page source, Search Console and a re-check on 19 September 2026. The firm is not named; the business is real and the numbers are its own.
What the crawler received
Two readings of the same home page, seventeen days apart
| Measurement | 2 September 2026 | 19 September 2026 | How it was measured |
|---|---|---|---|
| Words in the raw HTML of the home page | 11 | 1,123 | View the source, strip the tags and scripts, count the words |
| What a non-rendering crawler could read | The title, the navigation labels and a footer line | The services, the towns, the questions and answers | The same source view |
| Google's average position for the main search | 6.8 | 8.2 across the 30 days to 3 October | Search Console, query report |
| Pre-rendered head (title and description per route) | Yes, since 18 August | Yes | The same source view |
The first column is the fault. The page had been given per-route titles and descriptions in August, so a crawler received a correct head and an empty body. Google did not mind, because Googlebot renders the page and read the eight hundred words that appeared once the script ran. ChatGPT, Claude and Perplexity do not render, so for them the firm's site said its name and nothing else.
What was done, in order
Four dates, two of which matter
- 18 August 2026: per-route metadata and a pre-render of the head, so each page carried its own title and description in the HTML. The body was still empty until the script ran.
- 2 September 2026: the eleven-word count, taken while auditing another site. The firm's Google position that day was 6.8 for its main search.
- 3 September 2026: the page content put into the HTML at build time. One commit, one day, no change to what a visitor sees.
- 5 September 2026: two blog posts published. Between them they earned nine impressions and no clicks in the window below, so they are not the cause of anything on this page.
- 19 September 2026: re-measured. 1,123 words in the source of the home page.
- 20 September 2026: a wrong meta description corrected. It had advertised two services the firm does not offer, a fault from the original build.
The last item is the honest complication. A better description changes how many people click a listing without changing where it ranks, and that is exactly the shape of the numbers below.
What the numbers say, and what they do not
Thirty days against the thirty before
| Metric | Previous 30 days | 30 days to 3 Oct | Change |
|---|---|---|---|
| Clicks | 39 | 54 | +38% |
| Impressions | 2,692 | 2,900 | +8% |
| Click-through rate | 1.4% | 1.9% | +29% |
| Average position | 8.2 | 8.2 | No change |
Position did not move. Impressions barely moved. Clicks per impression rose by about a third, which is what a better snippet does and also what richer page content can do when Google rewrites the snippet from it. Both changes sit inside the window, so this page does not claim that putting the words in the HTML raised the clicks. It claims the narrower thing it can show: the words are now in the HTML, and the month after was the firm's best month in the period we hold.
Two more caveats. The previous window straddles the August 2026 spam update, which is a reason not to grade a single page across it. And the site is hosted by the firm's own developer and deploys on a timer, so the dates above are commit dates; the live change followed within twelve hours.
What is certain
The result that does not depend on the click numbers
Any assistant that fetches the page now reads the firm's services and the towns it covers, which it could not do before 3 September. That is a measured state, repeatable by anyone with the view-source command, and it is the outcome the work was for. A mention in an AI answer is not claimed, because we did not measure one. The eleven-word test, and the list of which crawlers render and which do not, is in the guide linked below.
What we did not do: no links were built, no new service pages were added in the window, and nothing was changed on the page a visitor sees. Everything above came from one build step and one corrected tag.
Where this leads
The pages that do this work
Find out what an SEO company would fix first on your site
Send your website address and your sector. Within one working day you get a written audit from Search Console and PageSpeed data, the three competitors above you, and a quote from the public price list. No sales call unless you ask for one.