References in what we publish
The numbers beside the headings
On a page we publish to your own site you may see a small [1] or [2] next to a section heading, and a numbered Sources list at the foot of the page. The numbers run in the order the sections appear, and each one links to its entry. A number against a section says one specific thing: this section was assembled from that piece of material. It is attribution at the level of the section and no finer. We do not mark individual sentences, because nothing here knows which sentence rests on which source, and a number on a sentence would claim that it did.
Whose sources these are
They are yours. The Sources block lists the material we built and published for your business — the guide, the FAQ, the specification, the data sheet — and the page says so in as many words: none of it is independent of the business. That sentence is deliberate. A numbered reference list is a convention borrowed from academic writing, where it means "somebody else's work supports this". Ours means "here is the material this page was assembled from, all of it yours". A list that looks like corroboration has an obligation to say when it is not.
Why you will not see numbers on most pages
Two reasons, both intended. They only appear where there is prior material to assemble. The first page we publish about a subject has nothing behind it yet, so there is nothing to number. The pages that carry references are the later ones, once several pieces of evidence about the same subject exist. And they only appear on pages published to your own domain. A Google Doc cannot carry structured markup at all, and docs.google.com has been cited once across the many thousands of documents published there — so inline references on a Doc would be provenance nobody reads. That is a deliberate omission rather than an oversight.
Outside references — and why they are almost always absent
Separately from the numbering, every document we generate is asked for genuine outside references: a real standard, study, paper, piece of official documentation or published figure that a statement in the document actually rests on. The instruction is deliberately hard to satisfy. It asks only for sources the model is certain of, by title, publisher and address — and it says outright that an empty list is the correct answer when nothing is cited. For most businesses, most of the time, it is. A caterer writing about pan sizes and a local agency writing about their service area are not resting on published standards, and a references list that filled up anyway would be a list of invented papers. An invented reference is worse than no reference, so the bar sits where it does on purpose. If you work in a regulated or technical field you will see these fill in more often.
Every address is fetched before it goes out
A reference with no address is not a reference, it is a claim that a source exists. Those are dropped at the point the document is read. Everything surviving that is then FETCHED before the page is published, and dropped if it does not answer. A model that obeys the letter of the instruction and cites a plausible-looking address that turns out to be a dead link has done the same damage as one that invented a paper outright. We would rather publish three references than four with one dead among them.
References sit above the link back to your site, so that when a document is later assembled into one of your own pages the link is stripped and the references travel with it.