For Claude our site had a single page. By the evening, five.
On the morning of 25 September 2026, for Claude's web search, Itria's site had a single page. In the evening, after asking Brave to recrawl seven, it had five.
Claude's web search also goes through Brave. Anthropic, in the list of its providers, names Brave Search for web search across all products, together with TurboPuffer. What Brave's index knows about a site is therefore part of what Claude can find.
From one page to five, in one day. With the same questions to Claude, itria.io appeared in the search results 4 times out of 8 in the evening, against 2 out of 8 in the morning, and with 5 different pages instead of one. In between, the request to Brave to recrawl seven addresses of the site.
Cited, still zero times. Appearing among the results does not mean being cited in the answer: in both measurements Claude did not cite itria.io. And a single pair of measurements does not prove the cause: we write it down as an observation.
This is a piece of Itria's lab: things measured on our own site, with the real numbers, to understand how search engines and assistants read it. The service that grows out of it is presence on search engines and AI assistants.
Why Brave matters for Claude
When Claude searches the web, it uses external providers. In Anthropic's Trust Center, the list of sub-processors, Brave Search appears with the «Web Search» function across all products, and next to it is TurboPuffer with the same function. A site that Brave's index knows poorly is less likely to appear among the results Claude reads.
«Brave Search. Products: All products. Web Search» (Anthropic, Trust Center, list of sub-processors, read on 25 September 2026)
The method: nine questions to Claude, with web search
The probe asks Claude, with web search on, a fixed set of questions: five on food exports, two about choosing a supplier, one about the brand and a site:itria.io search. For each answer it records whether itria.io appears among the search results and whether it is cited. One round costs between 1.3 and 1.8 dollars in API usage.
We ran it twice on the same day, 25 September 2026: at 11:23 and at 17:44, UTC. Between the two, after the first measurement, we asked Brave to recrawl seven addresses of the site from its submission page: the home page, EX1, the technical sheet template, the export guide, the services, the Dashboard and the English version.
The before and after
On the eight questions common to both measurements, itria.io appeared among the search results in 2 answers in the morning and in 4 in the evening. The different pages of the site seen in the results went from one, the Switzerland page, to five. On the brand question, in the morning no itria.io page came up; in the evening, two.
| Question | Morning, 11:23 | Evening, 17:44 |
|---|---|---|
| Documents to export food to Switzerland | Switzerland page | Switzerland page |
| Who automates foreign buyers' requests with AI | No page | Services |
| What Itria (itria.io) is and what it does | No page | Services, technical sheet template |
| site:itria.io search | Switzerland page | Home, EX1, Switzerland page |
| The other four questions | No page | No page |
| Citations in the answer | 0 | 0 |
How far this figure can be trusted
Not much, and we say so. It is a single pair of measurements, six hours apart. Web searches do not always give the same results to the same question. Between the two measurements two questions changed, which is why we compare only the eight common ones. And Brave might have recrawled the site anyway, without our request: coincidence in time does not prove the cause.
What the figure says, with these caveats, is a direction: three of the seven pages flagged to Brave, the home page, EX1 and the technical sheet template, appear in the evening results, together with the services. Citations remain zero. The next step is to rerun the same probe days apart, with the same questions.
The check, on your site
The simplest check is done on Brave's search: you search site: followed by your domain and count the pages that come up. If there are one or two on a site of dozens, Brave's index knows you poorly. Brave has a public page to request that an address be recrawled, and it is used one page at a time.
Questions and answers
Which search engine does Claude use for web search?
Anthropic, in its Trust Center, lists among its web search providers Brave Search, for all products, and TurboPuffer. What Brave's index knows about a site is therefore part of what Claude can find.
The list was read on 25 September 2026.
Does asking Brave to recrawl your pages help you appear on Claude?
In the itria.io case, on the same eight questions to Claude, the site appeared among the results 4 times in the evening against 2 in the morning, and with 5 different pages instead of one, after asking Brave to recrawl seven addresses.
It is a single pair of measurements and searches are not deterministic: the cause is not proven, and citations in the answer stayed at zero.
Does appearing in Claude's search results mean being cited?
No. Claude reads the search results and then chooses what to cite in the answer. In the two itria.io measurements of 25 September 2026 the site appeared among the results up to 4 times out of 8, and was never cited.
They are two different steps: being found by the index, and giving the best answer to the question.
How do you check how many pages of your site Brave knows?
You search Brave for site: followed by your domain and count the pages that come up. If there are one or two on a site of dozens, Brave's index knows the site poorly.
Brave has a public page to request that an address be recrawled, to be used one page at a time.
How is Itria's probe of Claude's web search built?
It asks Claude, with web search on, a fixed set of questions: five on food exports, two about choosing a supplier, one about the brand and a site:itria.io search. For each answer it records whether itria.io appears among the results and whether it is cited.
One round costs between 1.3 and 1.8 dollars in API usage, and it is repeated with the same questions to compare the measurements.
Notes on sources
- Anthropic, Trust Center, sub-processors: Brave Search and TurboPuffer with the Web Search function. Read on 25 September 2026.
- Brave, page to request that an address be recrawled.
- Itria's measurements: two rounds of the probe on Claude's web search, on 25 September 2026 at 11:23 and 17:44 UTC; recrawl request to Brave for seven addresses after the first round. The raw data is in our working archive.
- A single pair of measurements, with non-deterministic searches and two questions changed between rounds: the figure indicates a direction and does not prove a cause.
Seeing your own site the way search engines and assistants read it.
How many of your pages Brave knows, whether your email address can be read, whether the site answers addresses that do not exist: these are things you measure from outside. Write us a line about what weighs on you: we take the first step, even if we end up not working together.