Cute Robot: what an assistant can read, and what it cannot check

A sitemap is the list of pages you are telling machines you have. Without one, discovery leans on links, yours and other people's, so a page nothing points at is the one most likely to be missed.

Structured data is the machine-readable version of the facts already on your pages: who you are, what you sell, where you work. Without it those facts have to be inferred from prose, and inference is where you get described as something you are not.

One document answering every address means there is nothing page-shaped to read: an assistant asking for a service page, a location page or a typo gets the same few words back, and cannot tell any of them apart.

Blocking a training crawler is a decision a business is entitled to make. It is worth knowing it was made, because it is usually a security product's default rather than a choice, and it decides whether the site can be learned from at all.

Without robots.txt a crawler has no instruction file to read. It is also where the sitemap is declared, so its absence usually costs both.

The canonical tag is where you say which address is the real one. Without it a search engine picks one itself, and the same page reached two ways can end up counted twice.

Open Graph tags are how you choose the title, description and image shown when one of your addresses is shared. Without them, whatever is quoting the page chooses for you.

The H1 is the page's own statement of what it is. None leaves it unstated, several leave it ambiguous.

A page with almost no words has little to answer with, so it is rarely the thing that gets quoted.

llms.txt is a short plain-English summary of what the business is and what it sells, written for assistants. It is not yet a standard and nobody should promise results from it, but it costs one file and it is the one place you get to state the business in your own words.

Very short titles carry no information, very long ones get cut mid-phrase.

Some sites send an almost empty page and then draw the content with code once it is open in a browser. People see a full page. An assistant that only reads what was sent sees the empty one.

A buyer cannot judge an offer without knowing what it costs, and a site that does not say invites them to ask elsewhere.

An assistant can check what a business says about itself against a listing somebody else controls. A site that links none of them gives it nothing to check against.

Structured data is the machine-readable version of facts already on the page. Without it those facts have to be guessed from prose, and guessing is where a business gets described as something it is not.

A rating on your own website is a claim, not evidence. What makes it evidence is a link to the place the reviews live, where a stranger can count them.

Where a name, address, phone or list of services differs between a website and its profiles, the difference is what gets noticed, and it makes every other claim look less certain.