The layers of a result
An Aiscovery result answers one question for one website: what works for AI crawlers and agents today, what to fix first and what we haven't tested yet. It is built from four layers: the verdict, what we tested, the fix list and the evidence behind each fix.
Today every result comes from the automatic site check. Our checker reads your public pages the way a polite automated visitor would: at most one page per second, following robots.txt, never logging in, submitting forms, adding to cart or paying. It reports reachability and redirects, what your robots.txt declares for each AI company's crawler, AI-facing files such as robots.txt, sitemaps and llms.txt, and page basics such as titles, descriptions, headings and structured data syntax.
A verdict is limited to what we tested. "Not tested yet" means we don't know. It is never counted as a pass or a fail.
Coming later: key task results from named AI agents. When they are released for your plan, an agent tries a task such as "find a product and its price" in Aiscovery's controlled browser, and the result adds a line such as "Key tasks with real AI agents: 2 of 3 pass". A passing site check never replaces a completed task.
Each fix answers four questions:
- What did we see?
- Why does it matter for AI crawlers or agents?
- Who should change it, and how big is the change?
- How will we check it again after you fix it?
Check and task statuses
Checks and key tasks use different statuses because they answer different questions. A check looks at one thing on your public pages. A key task asks whether an AI agent reached a goal.
| Record | Status | Meaning |
|---|---|---|
| Check | Pass | We saw what the check expects. |
| Check | Fail | We saw the problem the check looks for. |
| Check | Couldn't check | We couldn't read the page, usually because a firewall or robots.txt stopped our checker. |
| Check | Not tested yet | We haven't looked at this. We don't know the answer. |
| Key task | Passed | Every literal check for the task passed on the pages the agent visited. Coming later. |
| Key task | Didn't pass | At least one literal check failed, or the agent didn't reach the goal. Coming later. |
| Key task | Waiting for review | The test run finished, and Aiscovery hasn't reviewed it yet, so it doesn't count yet. Coming later. |
A literal check looks for exact text or a link on a page, for example that a price is visible. An AI check can read the same evidence and confirm the result or flag it for review. It can never turn a failed check into a pass. Results are labelled AI-checked or Reviewed by Aiscovery so you can see which kind of review stands behind them.
"Couldn't check" and "Not tested yet" never count as failures. They lower our confidence instead, so an incomplete result can't look more certain than it is.
Failure attribution
Coming later. The same visible failure can have different causes, and we plan to separate them before a fix is assigned:
- Website: the page didn't show what the task needed, or a control didn't work.
- Agent: the page had what the task needed, but the agent didn't use it.
- Policy: the agent product declined the task because of its own rules.
- Platform: our browser, the network or an outside service failed.
What is true today: platform failures are ours. They are shown as "Couldn't check" or a failed test run that is given back to your plan, never as a problem with your website.
Priority
Priority expresses how much a fix matters for what we tested, not how hard it is.
| Priority | Interpretation |
|---|---|
| Critical | AI crawlers or agents can't reach the pages or facts a key task depends on. |
| High | A systematic problem with a primary page or materially wrong information. |
| Medium | Meaningful friction or a problem on a secondary page. |
| Low | A minor gap or an optional improvement, such as a canonical link. |
Aiscovery doesn't publish a single readiness score. The verdict and the fix list carry the result.
Limits and safe interpretation
Results are dated observations from a defined environment. They don't guarantee that every current or future AI model will behave the same way.
Review these details before comparing two results:
- the date of each result and the version of our checks;
- whether the result covers your Live or Staging website;
- which pages we read and which ones we couldn't check;
- for key tasks (coming later), the exact agent and model that ran;
- actions that were intentionally not taken.
AI crawler rules are declared policy: what your robots.txt tells each crawler it may read. They are not a test of what the crawler actually does, and we never pretend to be an AI company's crawler. Our checker and AI agents stop before login, cart, checkout, payment and any form that sends data.