We'd rather tell you something true than something impressive
A whole industry selling estimates, and a basic requirement left unsolved
When AI assistants started sending traffic, a whole industry appeared overnight promising to «rank you in ChatGPT». Almost everything on sale was an estimate presented as a measurement: share-of-voice figures worked out from a handful of questions the vendor picked, which shift on their own whenever the model is updated.
Meanwhile, a huge share of websites in Spain and Portugal fail at something far more basic and far cheaper to fix:
A Disallow left over from the development phase. Content that only exists once JavaScript has run. A firewall that hands a challenge to anything it doesn't recognise.
That much can be checked. It isn't an estimate, it doesn't depend on the sample and it doesn't change between runs. It is dull, and it is the prerequisite for everything else: however good your content strategy is, none of it matters if the crawler hits a 403.
This project does only that, and it does it for the Spanish and Portuguese markets, where nearly every tool in the field arrives half-translated or never arrives at all.
AI crawlers cannot get in, or they get in and find no text to read.
Six decisions that live in the code, not on a poster
- Identity
We don't pass ourselves off as anyone else
Our requests identify themselves with their own user agent and a page explaining who we are and how to block us. Your robots.txt is evaluated by reading it and applying its rules; the only time we use another bot's name is a declared test on your entry page, with a user agent that still says who we are.
Our crawler - Rigour
What we cannot check does not count as passed
If your server doesn't return the HTML, we don't invent a verdict about something we haven't seen. That item is declared inconclusive and drops out of the calculation. Rounding it up would be lying with a number.
- Priority
A blocking problem caps the score, it isn't averaged away
If the crawlers can't get in, it makes no difference how good the rest is: the score keeps a ceiling. An arithmetic mean would hide one fatal problem behind fifteen minor wins.
How it's calculated - Scope
We analyse several pages, not one
The home page tends to be the least representative page of a site: static and polished, while the real content lives elsewhere. We walk through pages from different sections and tell you how many of them each problem shows up on.
- Honesty
A failure on our side is never a fault on yours
If the connection drops, if we run out of time or if something breaks at our end, we say so plainly. What we don't do is write it into your report as though it were a problem with your site.
- Respect
We obey your robots.txt while we analyse you
When choosing which pages to visit we respect your own crawl rules, and we cap how many consecutive requests go to the same domain. It would be incoherent to ask you to let bots in and then behave like a rude one.
Four things that keep us out of the most sellable part of the market
We don't predict whether an assistant will cite you
Nobody can do that reliably from the outside. It depends on your brand, your authority and your competition, and none of those three things can be measured with an HTTP request.
We don't sell «share of voice» or estimated mentions
Those numbers depend on the questions you pick, they change between runs and they drift on their own when the model is updated. They can be measured; what they can't be is presented as hard data.
The diagnostic asks nothing of you
Analysing your site requires no sign-up, no email and no accepting anything. Analytics and advertising cookies load only if you accept them, and turning them down costs one click — the same as accepting them.
We don't share or sell your email address
It is used to send you what you asked for and, if you tick the separate box, for content of our own. Nothing else, and you can unsubscribe in one click.
One person, not a company
We are not a team, and saying so costs less than the alternative. What you can check is not who writes this: it is the user-agent we request with, how the score is calculated, and the list of pages in your report.
2026
Nothing goes to third parties until you accept or ask, and the servers are in the European Union
- The diagnostic runs in Paris, and your data does not leave the EU. The text of the pages analysed only leaves it if you ask for «What an assistant understands from your site»: it then goes to Anthropic, which processes it outside the EEA, without your email or your IP address.
- If you turn analytics down, nothing is downloaded from Google or Meta: it isn't loaded and then left unused.
- If you leave us your email, a confirmation reaches you before anything else: without that click, we don't write to you.
- Consent to receive the report and consent to receive content are two separate boxes, and neither comes ticked.