The site Google saw as 38 characters
A company from the esoteric sector had a modern website offering sessions and selling online. All the content was added only in the browser, so the search engine crawler received a practically empty document. After the work, the raw code of a page grew from about 38 characters to about 6 KB.
Why the company is not named
We publish a client's name only when we have their written consent. We do not have it, so this page has no name, logo, website address or any other detail by which the company could be recognised. What remains is the sector, the diagnosis and the scope of work, because that is the only part that has value for the reader here.
There are no growth charts here either. We explain below why, in this particular project, we would have had no way to document them.
A good-looking site, an empty document
The site was built from components loaded in the browser. A person saw the full page: service descriptions, expert profiles, the shop. The search engine crawler and AI models saw a skeleton with no content, because they fetch the raw code, not the picture from the browser.
- The raw code of a page was about 38 characters. That is what is left when all the content is generated on the user's side.
- There was no sitemap and no robots file, so nothing guided indexing.
- There was no
llms.txtfile, the entry point for AI models. - The login, account, basket and purchase history pages were open to indexing.
- Structured data was missing, so nothing told the search engine who this was and what it sold.
- Search Console was not connected, so there was no measurement of the baseline at all.
The cheapest test you can run on your own site today: open the page source with Ctrl+U. If you cannot see your texts in it, the search engine cannot see them either.
Fixing the technical layer without rewriting the site
The first decision was architectural, and it removed the biggest cost from the project. Instead of rewriting the site for server-side rendering, we made sure that the titles, descriptions, structured data, and the heading, description and links went straight into the raw code. We moved full prerendering of the content to the performance stage, because it was not a condition for indexing.
- An audit and an implementation plan broken into 14 steps, each with an acceptance criterion, an owner and a risk. The whole plan was written before the first line of code.
- Choosing the rendering strategy and deciding what we deliberately do not do at this stage.
- Content available to crawlers, meaning the heading, description and links, in the raw code of 35 pages.
- A robots file, a sitemap with 35 URLs and
llms.txtfor AI models. - Unique titles and descriptions on 42 pages, with the most important word at the start of the title.
- Six types of structured data: service business profile, website, service, the experts as persons, blog posts and breadcrumbs.
- Tidy URLs: the canonical URL on clean paths, and a redirect from the
/indexURL to the root directory. - Login, panel, account, basket, data and purchase history pages excluded from the index.
Everything reached production through three merged pull requests in the client's repository. No downtime and no change to how the site looks.
What could be verified after launch
We verified the points below on the live site in June 2026, in the raw code and in the browser. These are not forecasts or models, just a state that could be opened and counted.
- The robots file, the sitemap and
llms.txtrespond with a 200 code, which means they exist and can be read by search engines and AI models. - Transactional and private pages dropped out of the indexing queue, so they stopped frittering away the crawl budget.
- Content in the raw code confirmed manually, page by page, not only in a tool.
What we deliberately do not show here
There is no traffic chart, no number of keywords in the top ten and no sales growth percentage. The reason is mundane: Google Search Console was waiting for an entry in the domain on the client's side, so when the technical layer was closed there was no measurement from which we could take such figures. Putting them here would be making things up.
We treat the rest of the scope the same way. Pillar content, the blog, performance and authority building were planned out, but they depend on the client's materials and editorial decisions, so we do not claim their effects.
How long it took
| Stage | What happened |
|---|---|
| Audit and plan | Diagnosis of the site, a 14-step plan with acceptance criteria, a complete set of ready configuration files and meta texts. Closed before work on the code began. |
| Technical rollout | June 2026. Three merged pull requests, after which the technical layer was in production and verified. |
| Stage that depends on the client | Access to the domain for measurement, plus materials about the team and the offer. This part cannot be closed on our side, and it is what held up the measurement. |
| Horizon for further work | 90 days for content, performance and authority, set out in the schedule handed over to the client. |
The conclusion from this project that we see most often: the technical work itself takes days, not months. The months go on waiting for access and materials.
What this means if you run a similar business
- Check the raw code before you buy content. If the site is built from components loaded in the browser, no number of articles will help while the crawler sees an empty document.
- Not every site like this has to be rewritten. Rewriting for server-side rendering is a big project. Often it is enough to push the titles, descriptions, structured data and basic content into the raw code.
- Exclude the basket, account and login from the index. These pages will never sell anything to anyone from a search engine, and they eat up the crawler's attention.
- Connect measurement at the very start. Without Search Console you will not prove any effect later, even if it really happens.
- Prepare your materials early. Photos, bios of your people, prices and social media profiles are the only part an agency cannot do for you.
Who this story repeats for
For service businesses and shops that in recent years got a new, eye-catching website from a design studio or a developer, and whose search visibility dropped after the change or never took off. It is the same pattern: a modern front end and no layer that the crawler can read.
Three questions we get about this case study
We publish a client's name only after their written consent. As of the day this page was published we do not have that consent, so we describe the sector and the scope of work without details that would allow the company to be identified.
Because we would have no way to document it. Google Search Console was waiting for a DNS record on the client's side, so when the technical layer was closed there was no measurement from which we could show rankings and traffic.
Not always. In this project it was enough for the titles, descriptions, structured data, and the heading, description and links to reach the raw HTML. We moved full prerendering of the content to the performance stage, because it was not a condition for indexing.
Start with the same thing we started with here
Give us your website address and we will check what Google sees on it and what AI models see. No fee and no obligation. You get the result within five working days.