Being quotable: how AI assistants read your site

Answer engines cite pages they can parse, attribute and trust. What that means in practice, and which of your pages currently fail it.

SEO · Published: 9 September 2026 · 4 min read · Ольга Мирошник

A growing share of visits now begins with an assistant summarising the web rather than a list of links. The mechanics are less mysterious than the discourse suggests: an assistant fetches pages, extracts what it can parse, and cites the sources it can attribute. Every one of those verbs is a technical requirement you can check.

Can it fetch the page?

Most assistants crawl without executing JavaScript, or execute it with a short budget. A page that renders its content client-side may be, to that crawler, an empty shell with a loading spinner. Server-render or prerender the text you want quoted — this is the same requirement search engines had, applied more strictly.

Then check what your robots.txt says. Assistant crawlers use their own user agents, and a blanket Disallow written for scrapers years ago may now be excluding you from answers. That is a policy decision worth making deliberately rather than inheriting.

Can it parse the meaning?

Extraction favours documents with visible structure:

• One h1, then h2 and h3 in order. Skipped levels and headings chosen for their font size make the outline unreadable.
• The answer near the question. A heading phrased as a question with the answer in the first paragraph below it is the most citable shape there is.
• Facts in text, not in images. A price, a spec table or a deadline rendered as a picture is invisible to extraction. So is text inside a video.
• Structured data that matches the page. Product, Article, FAQPage and Organization markup remove ambiguity about what a number means.

Can it attribute you?

Assistants cite sources they can name. That means an explicit author, a publication and modification date, a stable canonical URL, and an Organization block that says who you are. Pages with no author and no date are the first to be summarised without credit — the content is used, the link is not shown.

Does it trust the page?

Trust here is mundane: does the page contradict itself, is the date plausible, does the linked policy exist, do the internal links resolve. A page full of broken links and a copyright notice from 2019 reads as abandoned to a model for the same reason it does to a person.

What to check on your own site

Take your twenty most valuable pages and ask, for each:

is the main text in the served HTML, without JS?
is there one h1 and a sensible heading order?
does a question in a heading have its answer in the next paragraph?
are the key facts text rather than image?
is there an author, a date and a self-referencing canonical?
is structured data present and consistent with the visible content?

Most sites fail three or four of these on most pages, and every one of them is a small, mechanical fix. None of it is a new discipline — it is the same crawlability, structure and honesty that search rewarded, now with a stricter reader that will happily answer from somebody else's page if yours is hard to read.