We check the site.Then the agents.

We record what is public. Then we test whether an agent can find the information and use it in an answer.

Step 1

What is public?

We check 11 places where a buyer may look for key information. We also save the homepage baseline and views of buyer routes that did not respond. We do not log in or change anything.

See the 11 places we check

Homepage/

Pricingfound address · up to 12 checks

Securityfound address · up to 12 checks

Privacyfound address · up to 12 checks

Termsfound address · up to 12 checks

robots.txt/robots.txt

Sitemap/sitemap.xml

llms.txt/llms.txt

Agent guide/agent.md

OpenAPI/openapi.json

Agent Card/.well-known/agent-card.json

The online check opens public HTTPS pages without cookies or a login. For pricing, security, privacy, and terms, an explicit versioned vocabulary of English and Polish phrases matches link text and address words on the homepage and in a limited part of the sitemap; the check then tries a fixed fallback list. It tries no more than 12 addresses per topic. The other resources keep their exact paths. Each request follows up to three safe redirects and reads up to 192 KiB. A named browser saves two homepage views and up to four non-reachable buyer pages; selection marks a review surface, not a proven defect.

Step 2

Run the same buying task with agents.

First, we define the buyer's question and the facts to look for. Then we compare the agents' full answers.

01
Record the site.
Save the date, addresses, and page responses. This shows what was public before any changes.
02
Define the buyer's question.
Write down the buyer's needs and the facts an agent should find.
03
Test the agents.
Repeat the same task in named AI tools and save the full answers. Treat owner-provided request logs as separate evidence.
04
Publish the result and its limits.
Keep facts separate from conclusions. Missing evidence, failed hypotheses, and corrections stay in the record.

How to read the result

One check cannot prove everything.

The online report shows whether a page responded. Every stronger claim needs a separate test.

What the online report does not prove

It does not show that ChatGPT, Claude, Gemini, a search engine, or a buyer found or used the page.

See the seven evidence states

01
PreparedThe page or file exists and has a clear purpose.
02
ReachableWe confirmed that the page or file opens.
03
ObservedAn external platform requested the page or file.
04
IngestedThe platform stored this version in its data.
05
RetrievedThe platform returned the content for a specific task.
06
UsedThe answer relied on the content.
07
InfluentialThe content changed an answer or buying decision.

Full documentation

Want to check the details?

The full protocol, the Signmos result, and the article are public.

Read the full protocol See the Signmos result Read the article