When someone asks ChatGPT, Claude or Perplexity where to eat or drink tonight, the assistant reads your site the way a crawler does: raw HTML, no JavaScript, no pictures. Enter your address and see which guest questions your site can answer, and what it leaves open.
We fetch your home page once, as GPTBot would, plus one menu page if we find one. Nothing is stored beyond a one-hour cache of the result. Only public websites can be checked. The checker accepts twenty checks a minute per visitor and five a minute per website, so it runs when you press the button and never on its own.
The same fetch, regrouped around what a guest actually types into an assistant. Each line shows what the scan found on your pages, or the reason it found nothing.
Answering these makes your venue easier for an assistant to quote accurately. It is not a promise that any assistant will recommend it: each one decides that for itself.
Each one is pass, warn or fail, and they add up to the score above.
An assistant is only useful to a guest if it can answer the question in front of it. These are the five that decide whether your venue survives the conversation, and what has to be in your HTML for each one.
Needs a weekly schedule an agent can read, in page text or in structured data. A photo of a sign on the door, or hours drawn by a script after load, is not one.
Needs the street address as text. A map pin is a link an agent reading HTML cannot open, and the check does not look for a nearest-station line, so add that yourself if guests need it.
Needs a phone number in the text or a tel: link. This is what an assistant falls back to when there is no way to book, so it should not be missing as well.
Needs items and prices as HTML text. The check reads one menu page if it finds one. It does not read dietary labels, so it can tell you whether a menu is readable at all, not whether a vegetarian question would land.
Needs a reservation path an agent can execute on the guest's behalf. This is about the path, not about tonight: nothing in this check, or in any check, can tell an assistant whether a table is free.
Ten technical checks, scored out of ten. The first two decide whether an agent can read anything at all, which is a separate problem from whether the answers are there.
robots.txt and firewall rules. Many Wix and Cloudflare defaults quietly block GPTBot, ClaudeBot and PerplexityBot, which makes the site invisible to the assistant.
Crawlers do not run scripts. A site that draws everything with JavaScript sends them an empty page.
The three things every "is it open now, where is it" question needs, in text or structured data.
As HTML text. A PDF is a maybe; a photo of the menu is a no.
A phone number is not one, and neither is a LINE chat. An agent needs an action it can execute on the user's behalf.
Schema.org markup that describes the venue, plus the link tags that point an agent at a clean profile and a booking tool.
An AI assistant reads a website the way a crawler does: raw HTML, no JavaScript, no images. An AI-ready site puts its hours, address, phone, menu with prices and a way to book into that HTML, allows the AI crawlers in robots.txt, and describes itself with Schema.org structured data so an assistant can answer questions about the venue and, where possible, complete a booking.
Weavify fetches your home page the way GPTBot does, without running any JavaScript and honouring your robots.txt, then looks for ten things an assistant needs. Each is marked pass, warn or fail and adds up to a score out of ten. The same evidence is regrouped into five questions a guest actually asks, so you can see which ones your site answers and which it leaves open. It reads one menu page if it finds one. Nothing is stored except a one-hour cache of the result.
No. The check measures whether the answers are on your site in a form an assistant can read. Which venue an assistant names is its own decision, and no score, listing or markup can control that. Two separate things are being measured here as well: whether the crawlers are allowed in at all, and whether the answers are there once they are. A site can fail the first and pass the rest.
Recommendation only needs the assistant to find a mention of the venue. Booking needs an action it can execute: a structured endpoint, not a phone number, a LINE chat or a web form. Without one, the conversation ends at "here is their number". A booking path is also not live availability: it tells an assistant where a reservation is made, not whether a table is free tonight.
Poorly. Some assistants open PDFs, most skip them during live web retrieval, and none quote prices reliably from one. A menu published as an HTML page with prices next to each item is what an assistant can read and repeat accurately. A menu that is an image is invisible.
Not to an AI assistant. Facebook and Instagram serve crawlers an empty shell and block the agents ChatGPT and Perplexity use for live retrieval. A page on your own domain, or a claimed Weavify listing, gives the assistant something it can read.
weavify