What it just did
Four steps, and none of them are clever:
- Fetched your homepage and checked your robots.txt first — if you ask crawlers to stay out of a section, we stayed out.
- Followed your internal links, shallowest first, because /about and /services describe a business better than a 2019 blog post does.
- Stripped the markup down to the readable text.
- Handed that text to a language model with one instruction: answer only from this, and admit it when the answer is not there.
That last rule is the whole thing. A model asked about your delivery radius without being given your delivery radius will invent one, confidently. Grounding is what separates a system you can put in front of customers from one you cannot.
Why the answers are thinner than the real thing
This demo reads about a dozen pages and forgets everything afterwards. A deployed agent knows your whole site, your price list, your stock, your calendar — and it can act: book the appointment, raise the ticket, write the enquiry into your CRM.
The gap between them is not model quality. It is connections.
If the demo came back empty
That happens with sites rendered entirely in JavaScript. The crawler sees an empty shell where a visitor sees a page.
Worth knowing, because search engines and AI assistants see roughly what our crawler saw. If we found nothing readable, that is a discoverability problem you had before today, and it is worth fixing regardless of whether you ever buy an AI agent.
What we did not do
No account, no email gate, no card. We did not store your pages beyond a short cache, and we did not read anything your robots.txt asked us to leave alone.
Questions
- Do you store my website content?
- Only briefly, as a cache, so that trying the same site again does not mean crawling it a second time. It is public content from your own pages, and it is not used for anything else.
- Why did it only read a few pages?
- The demo caps the crawl deliberately — it is meant to prove the idea in thirty seconds, not to index your site. A real deployment reads everything and keeps it up to date.
- It could not read my site. Why?
- Usually one of two reasons: the site is rendered entirely in JavaScript, so a crawler sees an empty page, or the robots.txt asks crawlers not to read it. The first is worth fixing — search engines see much the same thing.
- Do you honour robots.txt?
- Yes. We read it before anything else and skip whatever it disallows. Plenty of tools in this category do not; we publish a security page, so we do.
- Can the real one do more than answer questions?
- That is the point of it. A deployed agent books appointments, qualifies enquiries against your criteria, routes to the right person and writes the result into your CRM. Answering questions is the easy part.
Now you have the number, the useful question is what to do about it. That is one call.
Book a Consultation Call