Citation vs recommendation in AI: Lily Ray's study
Translated from the Spanish original. Leer en español
An AI can cite your content and recommend another company. What Lily Ray found in 100 B2B queries and how to measure AI visibility without fooling yourself.
In brief
- An AI citing your page and recommending your brand are two different facts, with very different commercial value.
- Lily Ray reviewed 100 B2B "best tools" queries in Google AI Overviews: in roughly 69% of cases, the pages where a brand proclaimed itself the best were cited without the brand being recommended.
- The study is observational and focuses mostly on B2B software; it is not a universal figure and doesn't prove a single signal.
- It is worth measuring four levels: presence, citation, recommendation and outcome.
The finding: your comparison can work for others
Lily Ray has raised a distinction that affects anyone doing GEO, AEO or any of the acronyms piling up next to SEO: an AI using your content is not the same as recommending you. Put like that it seems obvious, and yet in practice we confuse it every day.
Her study in Search Engine Journal examined 100 B2B queries of the best [category] type, that is, "best tools", in Google AI Overviews, with three rounds of checks between April and June 2026. For each one she asked two questions: whether the brand's page was cited and whether the brand was recommended.
The result: in roughly 69% of cases, the pages where a brand declared itself the best option ended up cited, but the brand wasn't recommended. Google took the content as a source without buying the article's commercial conclusion.
You write a comparison to prove why you are better and Google can use it to recommend precisely the others. Guerrilla marketing, but against yourself.
That is, Google can trust a page without trusting that the company signing it is the best choice for the user.
Four levels to measure AI visibility
We already measure things like whether ChatGPT cites us, whether we show up in AI Overviews or whether Perplexity names us. Fine, but what is missing is asking what role we play inside the answer: being a source is one thing and being the solution is another.
| Level | Question it answers |
|---|---|
| Presence | Does our brand or our URL appear? |
| Citation | Is our content used as a source? The system sees useful information on the page to build the answer. |
| Recommendation | Does our brand, product or service appear among the options advised to the user? |
| Outcome | Does that generate a qualified visit, a booking, a lead or a sale? |
If you cannot see all the columns, scroll the table sideways.
If Google uses our content, it means it finds it, understands it and judges it useful, which is already a good sign. If it also recommends the brand, we are talking about another level: trust in the entity, recognition, reputation, external consensus and the ability to influence a decision. It isn't achieved with a highly optimized page alone, unfortunately for anyone who thinks repeating "we are the industry leaders" three times in the first paragraph equals authority.
You can accumulate citations and receive few recommendations, or be recommended and still get few clicks because the decision has already been made inside the answer. Measuring gets more entertaining; Google Analytics had been asking for company for years. To work on visibility from the writing side, read how to write so AI cites your website or mentions your brand.
Comparisons: don't drop them, make them credible
My position is not to abandon comparisons: they can be among the most useful formats for search engines and generative systems. The flaw is in the ones with the impartiality of a mother talking about her son.
If you publish "The 10 best car rental companies" and by chance you are the best in price, coverage, service, fuel, deposit, customer service, vehicles and probably also parallel parking, the user may suspect something. Google too.
A truly useful comparison needs:
- Clear and verifiable criteria.
- Methodology, data and update date.
- Context, advantages and disadvantages.
- Acknowledging scenarios where another option may be better.
There is an uncomfortable risk: if you list five competitors and the system attributes stronger external signals to them than to you, your page helps Google understand the market and recommend someone else.
Travel and car rental: the case where your guide informs and someone else sells
Think of queries like "best car rental companies in Málaga", "which hotel comparison site is more reliable" or "which rental company has the fewest problems with the deposit". A company publishes an excellent guide that describes each provider and compares prices, excess, deposits, fuel, coverage and user experience. Google uses it as a source, great.
Then it checks that another company adds up thousands of reviews, editorial mentions, constant presence across several sources and much more brand recognition. Result: your page provides the information and your competitor gets the recommendation. So much for the work ending on your own website.
Authority is not manufactured only at home
You control what you say about yourself, but not entirely what the environment says:
- Reviews and mentions.
- Links and media coverage.
- Forums and external comparisons.
- Brand recognition and presence in other sources.
A brand can repeat a hundred times that it is the best; if a hundred independent sources say something similar, the signal is another one. On the weight of third-party sources, see why your competitors appear more than you in ChatGPT, Gemini and Perplexity.
How to apply it
- Choose a sample of relevant queries for your business, of the "best [category]" type.
- Separate in your report mentions, citations and final recommendations.
- Note which brands appear recommended and why (reviews, coverage, reputation, links).
- Review your comparisons with the list of verifiable criteria, methodology, date and scenarios where another option is better.
- Measure all the way to the outcome: qualified visit, booking, lead or sale.
Careful with the 69%
Better not to raise that percentage to a magic figure. The study is observational and focuses mostly on B2B software; we don't know whether it repeats the same in travel, car rental, ecommerce or other sectors, and there is no single signal that explains which brand is recommended. Authority, popularity, reviews, links, media coverage, product quality and reputation may weigh in, surely several at once.
The valuable thing about the study is not the 69%, but that it forces us to separate two concepts we usually mix without thinking. Applied to brand searches, you have it in what Google says about your brand and where it gets it from.
Frequently asked questions
What is the difference between an AI citing me and recommending me?
Citing you is using your page as a source because it contains useful information. Recommending you is your brand, product or service appearing among the options the system proposes to the user.
What did Lily Ray find?
After reviewing 100 B2B queries in Google AI Overviews, between April and June 2026, she saw that in roughly 69% of cases the articles in which a brand presented itself as the best option appeared cited, but the brand was not recommended.
Is the 69% valid for any sector?
It is not known. The study is observational and focuses mostly on B2B software, and we can't say it repeats the same in travel, car rental or ecommerce.
To finish your coffee
If an AI uses you as a source, it doesn't necessarily trust you as an option. In business terms, that difference weighs.
If you want to review how your brand appears in generative answers, tell me your case. And if you want to write content that is easier to cite, download the GEO writing guide (in Spanish).
Source: Lily Ray, analysis reported by Search Engine Journal. B2B study on citation and recommendation in Google AI Overviews, April-June 2026.
