238 pages and four forms
This case study is about us. Our site grew faster than our ability to measure it: 238 pages and a contact form on four of them. Before we started fixing anything, we recorded the baseline with the measurement method next to every figure, so that we could check later whether any of it made a difference.
Why we are showing our own site
A client's story with their name can be published only after their written consent. We are only now collecting those consents, so instead of making up figures for an anonymous company, we are showing a project in which the client is us. The whole benefit is on the reader's side: we can label every figure with the method we used to count it, and also show the part that came out poorly.
The rule we apply in reports for clients, and stick to here: a number without a stated source is an opinion, not a result.
The site grew, the funnel did not exist
For several months we kept adding content: posts, service pages, topic cards. All so that the site would have the depth that both Google and AI models like. The problem is that the content build-out got ahead of two boring things that everything else rests on: the ability to leave contact details, and measurement.
On 4 August 2026 we counted the baseline. This is what came out.
| Measure | Baseline | How it was counted |
|---|---|---|
| HTML files on the site | 238 | walking through the site's directories, 4 August 2026 |
| Pages with a contact form | 4 of 238 | searching the files for the form marker |
| Pages with tracking code | 78 of 238 | searching for the tag manager code |
| Pages with a consent banner | 168 of 238 | searching for the consent script |
| URLs in the main sitemap | 96 | counting the entries in the sitemap file |
| Pages outside the sitemaps | 112 of 238 | difference between the number of files and the URLs in both sitemaps |
Almost every page that someone could find in a search engine was a dead end. The reader reached the content, nodded and had nowhere to click. Almost half of the site was not even submitted in a sitemap.
The funnel first, more content later
The order was deliberate and the reverse of the one most companies choose. We did not add a single new page until the existing ones had somewhere to lead the reader and until we could see what was happening on them.
- We wrote down the baseline in one document, with a stated source for every figure and a separate list of things we deliberately did not measure. Without it, no improvement could have been proven later.
- We wrote a tool that walks recursively through all pages, nested ones included, and injects the same contact form into one marked place in the code.
- The same for measurement and cookie consent: one tool, one marker, one pass over the whole site.
- One generator for the sitemap and the
llms.txtfile, so that no new page could ever fall out of indexing again. - Duplicate URLs closed with permanent redirects, with the same list of rules in the configuration of both servers, because a mismatch between them is the most common source of silent errors.
- The form connected to our own request-handling system, with a bot trap and a data processing notice on every submission.
The state after the work, counted with the same method
We counted the figures below on 5 August 2026 in the site's source, with exactly the same commands as for the baseline. That is the only way for the comparison to make sense.
- As of 5 August 2026 the site had 278 HTML files, so it grew in the meantime, and yet the share of pages without a form and without measurement fell to isolated cases.
- The rest are pages that should not have a form: the privacy policy, the terms of service, the error page, technical files.
- Duplicate URLs closed with permanent redirects on both sides of the configuration, so the same content stopped appearing under several URLs.
What we still cannot prove
We do not show search traffic or the number of form submissions broken down by source. The service account through which our panel reaches data from Google Search Console and Analytics returns an authorisation error, and the files with the readings are empty. Until it is fixed, any traffic figure would be made up, so there is none here.
This is, in fact, the best illustration of the point of this case study. You can have a correctly built funnel and still be unable to show a result if the measurement does not work. That is why with clients we start with access, not with content.
How long it took
| Stage | When | What happened |
|---|---|---|
| Baseline measurement | 4.08.2026 | Counting the site, the panel data and the infrastructure, and writing down what we deliberately do not measure. |
| Fixing the funnel and measurement | 4-5.08.2026 | Injecting the form, the tracking code and the consent script across the whole site, one sitemap generator, duplicate URLs closed. |
| Final state | 5.08.2026 | Recounting with the same commands and comparing with the baseline document. |
| Open | - | Restoring access to search data, without which there is no way to show the effect on traffic. |
The work that turned four forms into two hundred and twenty-three took two days. Not because we are fast, but because it was done by a tool walking through the files, not by a person clicking through every page.
What this means if you run a similar business
- Count before you start. Three numbers are enough to begin with: how many pages you have, on how many a visitor can leave contact details, and how many of them are in the sitemap. Without them you will not prove to yourself or to an agency that anything has improved.
- A form on every page, not just on the contact page. People coming from search engines and from AI answers land deep inside the site, not on the home page. If there is no way to get in touch there, all the work on content ends as a stranger's visit.
- One tool instead of pasting by hand. Tracking code pasted by hand always goes missing somewhere. A generator that walks through the files does not forget.
- Redirect rules must be in every server configuration you use. A rule added in one place and missed in another only becomes visible when it starts doing damage.
- Sort out access to measurement on day one. It is the most boring item in any engagement, and the only one whose absence can invalidate a whole quarter's report.
Three questions we get about this case study
Because publishing a client's story with their name requires their written consent, and we are only now collecting those consents. On our own site we do not need to ask anyone for consent, and we can show the raw figures together with the method we used to count them.
The service account through which the panel pulls data from Google Search Console and Analytics returns an authorisation error. Until we fix it, we have no right to show traffic figures, so we do not show them.
By counting three things: how many pages you have, on how many of them a visitor can leave their contact details, and how many of them are in the sitemap at all. These three numbers are usually enough to see where the effect of your content work leaks away.
Want the same numbers about your own site?
Give us the address and we will count the same for you: how many pages Google knows, on how many a visitor can leave contact details, and what is missing for AI models to have something to quote you for. Free, within five working days.