Answers
Why is my documentation not cited by AI assistants?
Short answer
Why is my documentation not cited by AI assistants?
Documentation answers a different question from the one a recommendation query asks. Docs are written for somebody who has already chosen you and needs to know how something works, so they explain parameters and return shapes and never once say who else a reader might have picked or why you would suit them better. A recommendation query needs a page that compares, and a page that compares is almost never a vendor's own docs. That is why an assistant with your entire documentation set available will still answer a choosing question from a third-party roundup. Docs are excellent at holding a customer and structurally unable to win a decision.
What docs can and cannot win
Not a quality problem. A choosing question needs a page that compares, and documentation never compares.
01
Your docs answer the question after the decision
Documentation is written at a specific moment in a reader's life, which is after they have picked you. Everything about the form follows from that: it assumes the product, it explains mechanics, and it never argues. A developer asking which API to use is standing before that moment, and nothing in your docs is addressed to them, so retrieval that is looking for a page weighing options finds nothing to take.
This is not a defect to fix by rewriting the docs. Docs that started arguing against competitors would be worse documentation and would still lose, because the pages that win those queries are trusted precisely for not being written by a vendor.
Where each page answers
Documentation is written for the reader below the line. Recommendation queries are asked above it.
02
What being cited from docs actually looks like
Docs do get cited, just for a narrower class of question. When somebody asks how to authenticate, what a rate limit is, or what a specific field returns, your reference page is often the best answer available and gets used as one. That is real and worth having, and it is also why teams conclude their docs are working: they see citations, and the citations are all on questions asked by people who already chose them.
The distinction worth drawing internally is between citations that hold a customer and citations that win one. Both count, they are not substitutes, and a dashboard that adds them together will tell you things are fine while you lose every recommendation query.
Count these apart, not together
Shape, not a measured ratio for your product. A dashboard that adds the two together reads healthy while every choosing query is lost.
03
The one docs change that does help
There is a version of this work that pays, and it is not a rewrite. Assistants lift cleanly from pages that state a self-contained answer in one place rather than assembling it across a table, three headings and an aside. Most reference documentation fails that test not because it is wrong but because the answer to a plain question is distributed across the page, and an extractor takes a fragment that reads as incomplete.
Giving the handful of genuinely common questions a direct paragraph each, in your own docs, in your own words, is cheap and it works on the questions docs can actually win. It does not move recommendation queries, and expecting it to is how the effort gets miscounted as a failure.
What an extractor can take
Same facts either way. Only the second one survives being lifted out of the page it came from.
Asked next
The questions that follow this one
Should I add llms.txt to my documentation?
It costs nothing and it is not the constraint. Recommendation answers are assembled from third-party pages, which our own measurement puts at 91.5 percent of 901 citations, and no file on your domain changes which of those pages get retrieved.
Do assistants read my docs at all?
Frequently, for how-to and reference questions from people who already chose you. That is genuine value and it is a different job from being recommended, so it is worth counting separately rather than folding both into one visibility number.
Would better SEO on my docs help?
It helps the questions docs can win. It does not make a docs page the right retrieval result for a question about which product to choose, because that question needs a page that compares and docs do not compare.
What should we write instead?
Direct answers to the questions buyers ask before choosing, on a surface built for it, plus accurate presence on the third-party pages already answering those questions. This page is an example of the first half.
Where to spend the effort
Free gap report
When the model answers,
be the one it names
Tell us your category and the questions your buyers ask. We run them against live AI answers and walk you through what came back. If you are already winning, we will tell you that too.
12 queries · 5 repeats each · 5 working days
Free · Walked through live · Five working days
Queries we run
Cited instead of you
reddit.com38%
g2.com26%
news.ycombinator.com21%91.5% of citations point off-site