Why AI engines do not cite you yet, even with a perfect SEO score

By Attiq Ur Rehman Khalid · 2 November 2026

A site can score 100 for SEO, carry every schema type an audit asks for, and still never be named by an AI answer engine. We know, because that site is ours. For six months our pages passed every check Iris runs, and for six months the citation check came back 0 of 8. This week a national broadcaster we audited for the first time was named in 3 of 8 answers on its first check, with a page that scored 64 for content. The difference is not the score. This article is about what the difference actually is, and what to do in the order that works.

What does an AI engine do before it names anyone?

An answer engine with web search does not read the web. It runs a search, takes the handful of results that come back for its own phrasing of the question, reads those, and writes an answer that names some of them. Everything that follows comes from that one fact.

• If your page is not in the top results for the question the engine actually typed, the engine never sees it. No amount of on-page quality fixes that, because quality is judged only after a page is read.
• The engine's phrasing is rarely your phrasing. It asks the way a stranger would: "best live meeting copilot for sales calls", not your product name. If nobody has written your name next to that phrase, you are absent from the results.
• Engines prefer pages that list several options. A comparison, a directory, a "top tools" roundup or a forum thread answers the question in one place, so it ranks for it and gets quoted. Your own homepage answers a different question: "what is this company?"
• Two engines with the same question return different sources, and the same engine returns different sources on different days. One check is a reading, not a verdict.

Our site had the first problem. It was well built and almost nobody else had written about it, so for the questions a buyer asks, it did not exist.

Why did a 64-point page get cited when a 100-point page did not?

The broadcaster has been written about for decades. Wikipedia, press, directories of media outlets and other broadcasters all mention it by name next to the words an engine would search for. When the engine asked where to find reliable news from that region, the broadcaster was in the results before the engine read a single page. Its weak headings and missing lists cost it points in our audit and nothing in the citation.

That is the uncomfortable order of things: being found comes first, being readable comes second. Most AI-search advice reverses it, because readability is the part a tool can fix for you.

What should you do first, second and third?

1. Get your name onto the pages engines already quote. Find out which pages are cited for your category's questions and get listed there: software directories, comparison articles, industry roundups, the forum threads where people ask your question. Iris shows these as "named instead of you" and "named alongside you" after every check, so the list is not a guess.
2. Write the one sentence that says what you are. Engines and directories both need "X is a Y for Z" in plain words, in your title, your description and your Organization schema. A page that describes itself as "the future of work" is not matched to any question.
3. Answer the real questions on your own site. Add a page, or a section, whose headings are the questions people ask in your category and whose answers are short, factual and specific. Not a FAQ about your pricing: the questions a buyer types before they know you exist.
4. Ask customers to review you where engines look. A handful of reviews on a listing site does more for citations than a redesign, because the review page ranks for "X reviews" and "X alternatives" and the engine reads it.
5. Fix the on-page points last. Lists, question headings, schema and llms.txt decide how much of a page the engine uses once it has it. They are worth doing; they are not what gets you in the door.

How long does it take, and how do you know it is working?

Listings and reviews are read by search indexes within days, and answer engines pick them up from there. We do not have a measured average yet, so we will not quote one: expect the first named answer some weeks after you are listed in two or three places, and expect it to be one answer out of eight, not four.

• Run the same category's questions every week, not different ones. A changing question set makes the number unreadable.
• Keep the per-engine tally. Being named by one engine and not another tells you which index you are missing from.
• Watch "named instead of you". When a rival drops off that list and you appear, you have taken their slot; when both of you appear, you have joined the list, which is the normal first step.
• Record the date of the first citation. It is the one number in AI search that only moves forward.

Iris keeps that timeline for every site it checks: each check as a bar, the first citation as a dated line, the engines and the exact question that named you. Six months at zero was a fact about our listings, not our pages. We are starting the fix in the order above, and this page will carry the date of our first citation when it lands.