Guide · Checking citability

Can an AI assistant cite your site?

Two different things can go wrong here, and they have nothing to do with each other. An assistant may be unable to read your page at all, or it may read the page and find nothing worth quoting. This guide checks both.

Two failures that look identical

A fetch failure is a delivery problem, and a quoting failure is a writing problem. From the outside they are indistinguishable. You ask an assistant a question your page answers, and your site is simply not named.

An answer engine is a system that reads sources and writes a reply instead of listing links. It has to finish two steps before it can name you, and either step can end the attempt on its own.

Step What has to be true How the failure shows up
FetchComplete HTML arrives without running scriptsVague description, or none at all
QuoteSome passage stands alone out of contextYou are known, but a competitor is cited

Checking the fetch

Server-side rendering means the HTML arrives complete from the server before any script runs. When your content only appears after JavaScript executes, a crawler that does not execute scripts sees an empty shell.

  1. Turn off JavaScript in your browser and reload the page. Whatever survives is roughly what a plain crawler receives.
  2. View the page source and search it for a sentence you can see on screen. If the sentence is missing, a script drew it.
  3. Read the robots file at the root of your domain. A disallow rule you inherited from a template blocks well-behaved crawlers silently.
  4. Count the words that survived step one. Our AI Accessibility module gives full marks once at least 200 words of real text render without scripts.

That module carries 8% of the overall grade, and it is the cheapest of the five AI-search modules to fix. Most sites that fail it fail on rendering rather than on writing.

Checking whether anything is quotable

Citability is the property of a page that decides whether a model can lift a passage out of it and stand behind the result. Our grader scores it as 12% of the overall grade, which puts it level with Entity Clarity at the top of the AI-search half.

The thresholds below are the ones the code applies, not general advice. Partial credit is real, so a page that misses full marks is rarely at zero.

Dimension Points Full marks Partial credit
Data points20Five or more concrete figures14 for three or four, 7 for one or two
Structured content15Two of list, table, definition list8 for one of the three
Questions and answers15FAQ markup and three question headings10 for either one alone
Definitions10Three sentences that plainly define a term5 for one or two
Self-contained sentences1060% of sentences run 8 to 30 words7 at 40%, 4 at 20%

A figure only counts when it is written as a figure. A percentage, a sum of money or a number with thousands separators reads as data, while a bare integer inside a sentence does not.

The ten-minute version

Run this on the single page you most want quoted. Do not start with the whole site, because the fixes generalise once you have made them once.

  1. Reload with scripts disabled and confirm the main text is still there.
  2. Count the concrete figures in the copy, and get the total to five or more.
  3. Add a table or a list, so the page carries at least two kinds of structure.
  4. Write three real questions as headings, and mark them up as an FAQ.
  5. Add three sentences that define your key terms in plain language.
  6. Break any sentence longer than thirty words into two that stand alone.

What a citation actually requires

A quotable sentence survives being removed from the paragraph around it. If a sentence opens with "this" or "as mentioned above", it cannot be lifted into an answer without dragging its neighbours along.

That is the whole discipline, and it is unglamorous. Write the sentence you would want to see quoted back at you, and put it where a reader can find it.

Common questions.

Can I just ask an assistant whether it can see my site?

You can ask, but the answer is unreliable. A model may describe your site from training data it absorbed months ago, or invent a plausible-sounding description from your domain name alone. Ask it a question only your page answers, then check whether your domain is named as the source.

Does blocking AI crawlers in my robots file hurt me?

It ends the conversation before it starts. A crawler that is disallowed cannot fetch the page, so nothing else on this list matters. Blocking is a legitimate choice, but make it deliberately rather than inheriting it from a template or a plugin default.

Why does a page that ranks well never get quoted?

Ranking rewards the document and quoting rewards the passage. A page can earn a first-page position on links and relevance while containing no definition, no figure and no sentence that stands on its own. The model has nothing it can lift, so it lifts someone else's.

How long after I fix a page before an assistant notices?

Nobody publishes a number, and anyone quoting one is guessing. Retrieval-backed assistants that fetch live pages can reflect a change within days, while answers drawn from training data may lag by many months. Fix the page, then re-check on a schedule instead of waiting for an announcement.

The free scan runs every check above on one page of your site and returns a letter grade with the failures named. No account and no card. Paid teardowns start at $79 and crawl more of the site.

Run a free scan →

Related: what generative engine optimization actually is, and why your business does not appear in AI answers.