narubox™ X-Ray Free No signup 20 seconds
When AI gets your business wrong, nothing breaks.
No error page. No bounce in your analytics. No angry email. An assistant answered a question about a business like yours with a price you stopped charging in 2023 — or with a competitor — and the call simply never came.
The part of your site assistants read is the part you cannot see. Open the page in a browser and it looks perfect; the instructions machines follow are somewhere else, and nothing tells you when they go wrong. X-Ray follows every one of them and asks the only question that matters: is any of this still true?
Nothing is stored, nothing is emailed, and no account is created. We fetch the page and the addresses it names — nothing else.
We found ours by accident
For weeks our own pages told every assistant and every social network that our logo lived at an address that returned a web page instead of a picture. The markup was valid the entire time. Google did not mention it. The validators we ran passed it. We noticed because someone happened to glance at a phone.
If it can sit broken on the site of a company whose entire product is this, it is sitting broken on yours. That is not a sales line — it is the reason this tool exists at all.
Why a date decides whether you win
This is the part almost nobody knows, and it is the reason a page can be perfectly accurate and still lose.
An assistant answers from two places at once: what it learned during training, which is months or years old and carries no date at all, and what it fetches in the moment. When those two disagree about your business — and for a small business they usually do — the one that can prove it is more recent wins.
A page with no dateModified cannot make that argument. It sits
next to a three-year-old directory listing with your old phone number on it,
and to the machine choosing between them, the two look exactly the same age.
Google hands the judgement to you: it returns links, you open one, and you see "updated 2019" for yourself. An assistant does not. It reads the sources, writes one sentence, and the date is gone by the time you see the answer. Nobody can ask it how old that was. So the only moment freshness can possibly be judged is before the sentence is written, by a machine, from whatever the page bothered to state.
What X-Ray checks
- Do the addresses resolve? Every URL the markup declares is fetched. A declared address that 404s is a claim the page cannot support.
- Are the images actually images? A
logothat answers with HTML is the exact failure we shipped ourselves — and on the types that publish a format list, an SVG counts as missing even though it is a real image. - Do the FAQ answers exist on the page? Structured data has to match what a reader sees. Claiming an answer that is not in the visible text is the case search engines act on by hand.
- Does it say how current it is? See above. This one is worth more than it looks.
- Is there a share image, and is it real? The same silent failure, one layer out — every link anyone posts previews without a picture.
What X-Ray does not do
It does not tell you whether a fact is true. Nobody can check from outside whether a business really is licensed, or really opens at nine, and any tool claiming otherwise is selling something that does not exist.
What it checks is narrower and more useful than it sounds: whether the page is internally honest. Whether the things it tells machines are things that actually exist, and whether it says how current they are. Almost every wrong AI answer about a small business comes from one of three places — an old page winning, two sources disagreeing, or no source at all. All three are visible from the outside. Lying is not, and is not usually what happened.
Is this safe to run on my site?
It fetches your page and the addresses your page itself names. Nothing else is requested, private networks are refused outright, and every redirect is re-checked before it is followed. Nothing is stored, nothing is emailed, and no account is created.