Our site scores 100 out of 100 on technical SEO. In three months, AI answer engines cited it zero times. Not few. Zero. This is what we found when we went looking for the reason, and what we would check first on any site with the same problem.
Where the number came from
Bing Webmaster Tools has a panel most people never open: AI Performance. It reports how often Copilot cites your site, and how many of your pages it has quoted. Ours read 0 citations and 0 cited pages, over 90 days, with 23 URLs successfully crawled and indexed.
Indexing was not the problem. Being findable and being quotable are different things.
What is an AI citation?
When someone asks ChatGPT, Perplexity, Claude or Google's AI Overviews a question, the answer names a handful of sources. Being named is a citation. It is not a ranking, not a click, and not an ad. The engine simply decided you were worth quoting.
Two different things get measured, and confusing them wastes months:
- AEO — answer engine optimisation is a prediction: is this page structured so an engine could quote it? Clear headings, structured data, direct answers.
- GEO — generative engine optimisation is a measurement: did an engine actually name you when a real buyer question was asked?
You can score perfectly on the first and zero on the second. We did.
How to measure it yourself
- Open Bing Webmaster Tools and check AI Performance. ChatGPT search runs on Bing's index; if Bing has not crawled you, ChatGPT cannot cite you regardless of your score.
- Ask an AI assistant the questions your buyers type — the category question, not your brand name. "Best tool for X", not "what is [your product]".
- Write down who it names instead of you. That list is your real competitive set.
When we did this for our meeting copilot, the engines named otter.ai, fireflies.ai, read.ai and krisp. All established, all discussed widely elsewhere.
Why perfect pages do not get cited
Answer engines quote what other sites say about you. A page you wrote about yourself is a claim; a comparison post, a directory listing or a forum answer that names you is evidence. The models weight evidence.
This is uncomfortable if you have just spent weeks on schema and headings, but it explains the data: our on-page work was finished and our citation count did not move, because nothing off our own domain mentioned us.
What actually moves the number
- Get listed where comparisons live. AlternativeTo, G2 and Capterra pages are structured as "alternatives to X" — exactly the shape an engine lifts when someone asks for options.
- Publish original data. Numbers nobody else has are the most quotable thing on the internet. This article is an example; our AI Citation Index is another.
- Answer the question in the first sentence under the heading. Engines lift self-contained passages. A heading phrased as a question, answered immediately beneath it, is far easier to quote than the same fact buried in paragraph three.
- Fix entity confusion before anything else. If an AI describes a different
product when you ask about yours by name, no amount of content will help until it knows which
company you are. Add a
disambiguatingDescriptionto your schema stating plainly what you are not. - Keep it current. Models favour recently updated pages, which is why a visible last-updated date matters more than it used to.
What we are doing about it
We publish the AI Citation Index: measured citation rates by category, from live engine answers, with the methodology published and free re-measurement for any company listed. Not estimates — if we have not run it, we show nothing.
We will report our own number as it moves, including if it does not.
Check your own gap
Iris scores any page free, with no signup: an SEO score, an AI-search readiness score, and a check of whether engines actually cite you. The gap between the second and third numbers is the entire job.