Cited Is Not Recommended
An AI platform can use your pages as a source and still tell the buyer to purchase from your competitor. Research by Lily Ray found that when AI answers cited vendors' own pages, those vendors were still excluded from the answer's recommendations in 69% of cases in June 2026 retests — and 83% by September 2026. Citation is the weakest of the three visibility signals. That is why our audits score mention rate, citation rate, and recommendation rate as three independent numbers, and why any report that shows you only a citation count is measuring the wrong thing.
What the research found
Lily Ray, founder of Algorythmic and one of the most careful measurers in this field, runs what she calls a citation audit: the same buyer prompt is run across multiple AI platforms, multiple times, and two things are recorded separately — which pages appear in the answer's sources, and which brands the answer's text actually recommends. Run that way, a pattern emerges that citation dashboards never show. Vendor pages are cited as background — for specifications, definitions, pricing facts — while the recommendation sentence names a different brand, usually one backed by reviews, comparisons, and community discussion.
The June 2026 round of her retests found vendors excluded from recommendations in 69% of the cases where their own pages were cited. By the September 2026 round, the exclusion rate had risen to 83%. The direction matters as much as the numbers: the gap between being cited and being recommended is widening, not closing. Her work also tracked the pages vendors write to rank themselves — the “best tools in our category” listicles where the author conveniently finishes first. Citations to those self-ranking listicles fell 38%, evidence that the platforms are learning to discount a source whose author grades their own homework.
Why answers behave this way
An AI answer is assembled from two different trust pools. For facts, it reaches for whoever states the fact most clearly — often the vendor. For verdicts, it reaches for whoever sounds least interested in the outcome — reviewers, comparison sites, forum threads, publications. A vendor page is treated as a claim to be checked, not a judgment to be repeated. This is not a bug a vendor can patch with better keywords. It is the platforms doing, crudely, what a careful buyer does: listen to the seller for specifications, and to everyone else for the decision.
It also explains why “we got cited” is such a seductive metric. Citations are easy to earn with factual, well-structured pages, and they feel like winning. But a citation sitting under an answer that recommends your rival is not visibility. It is unpaid research assistance for your competitor's sale.
How we measure it
This is our method, and it is built directly on the distinction above. Every audit records three rates per platform, separately for the brand and for each product or SKU. Mention rate: in how many answers does the brand appear at all? Citation rate: in how many answers do the brand's own pages appear in the sources? Recommendation rate: in how many answers does the text actually put the brand forward — recommendation language in the sentences around the mention, such as “best for”, “top choice”, or an explicit pick? A brand can score 100% on the first two and near zero on the third. We have measured exactly that pattern, and it is the most common serious finding in our audits.
The three rates also fail differently, which is why they must never be blended. Low mention rate is an awareness problem: the platform does not associate your name with the category. Low citation rate is a content problem: your pages do not state facts in a form the platform can lift. Low recommendation rate is a trust problem: third parties have not vouched for you in the places the platform reads. Three problems, three different fix lists. One blended score would hide which lever to pull.
The bottom line
Treat citation as a supporting metric, never the headline. The headline is the recommendation rate, because the recommendation sentence is where the buyer's decision happens. If your current GEO reporting shows citations and calls it visibility, ask for the other two rates. If the vendor cannot produce them, you are not being measured — you are being reassured.
Sources & method. Citation-versus-recommendation figures (69% exclusion, June 2026; 83%, September 2026; self-ranking listicle citations −38%) are from published research by Lily Ray (Algorythmic), based on repeated citation audits across AI platforms. The three-rate measurement method — mention, citation, recommendation scored independently — is our own audit standard, live in our tracking workspace since its v4 release.
© 2026 Jun Liu · JUN & GEO. This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0) — https://creativecommons.org/licenses/by-sa/4.0/. You may share and adapt it with attribution and under the same license.