We asked the same question to Google and ChatGPT: 60 purchase intentions in Spanish, 3 consecutive days, with the protocol closed before collecting the first data. The result dismantles a widely held belief: only 13.7% of the brands that ChatGPT recommends have their website in the top 10 organic results on Google for that same question. 73% appear within the listings that do rank.
There is a question that almost every client asks us: "I am already in the top 10 of Google, does that guarantee that ChatGPT will recommend me?"
And so far the honest answer has been: we don't know. In Spanish, no one had measured it.
Ahrefs demonstrated in English that AI cites pages that Google does not rank. But it measured URLs, not brands. And your business doesn't care if ChatGPT cites your URL: it cares if it recommends you.
So we measured it. With the same logic as the entire series: frozen questions before collecting, repetitions on different days, confidence intervals, and a published dataset so anyone can audit us. The complete methodology is at the end of the article, in section 08.
An important clarification before starting: when we say "top 10 of Google" we always talk about the classic organic results. Without ads and without the AI Overviews response. We compare the traditional ranking against ChatGPT.
The gateway to ChatGPT is not your ranking: it’s the lists that rank. ChatGPT does not recommend you because you rank; it recommends you because the listings that rank talk about you.
Before looking at brands, we validate the method by repeating the Ahrefs measurement with our data: what percentage of the URLs that ChatGPT cites rank in Google organic for that same question?
The pattern is almost identical. And our 8.7% (with a confidence interval of 6.4 to 11.1) is even below the 12% measured in English.
At the domain level, the numbers rise: 19.2% in the top 10 and 39.2% in the top 100. The story doesn’t change: ChatGPT does not cite what Google ranks.
For an independent measurement to reproduce the industry benchmark is the validity check of the method. Now, onto what no one had measured.
We define two possible paths between Google and a ChatGPT recommendation.
Direct path: your website ranks in the organic top 10 and ChatGPT recommends you.
Indirect path: your website does not rank, but the listings and comparisons that do rank mention you, and ChatGPT recommends you.
The intervals do not overlap. This is not a nuance: it is a fivefold difference.
Why does this happen? Because for a question like "best X", the top 10 organic are almost never brand websites. They are comparisons from media, specialized blogs, and forums.
ChatGPT searches, reads those lists, and derives its recommendations from them. Your website may be perfectly optimized: if you are not on the lists, for ChatGPT, it's as if you do not exist.
86% of the brands that ChatGPT recommends do not have their website in the top 10 organic on Google. 73% are mentioned on the pages that do. Visibility in AI is played out in third-party content.
The most illustrative case of the mechanism can be found in our own category. In "best tools to measure a brand's visibility in AI", one of ChatGPT's recurring recommendations is a tool whose blog systematically publishes lists like "the 12 best visibility tools in AI" where they include themselves.
The listings rank, ChatGPT reads them, the brand enters. The indirect path, functioning in real time.
The direct path does not weigh the same in all categories.
Descriptive data by category. The sample size per cell does not allow for fine intervals.
If you sell physical consumer products, the direct path practically does not exist: 4.7%.
Manufacturers do not rank for "best running shoes." Specialized magazines and blogs rank. In finance, however, banks and brokers manage to sneak their domain in one out of four cases.
The reverse question also has an answer, with a nuance that is worth gold.
If ChatGPT already considers you part of the conversation in your category and also ranks, it recommended you that same day in 77.2% of observations (interval of 69.0 to 86.0).
But counting all brand domains that appear in a top 10, including those that Google ranks and ChatGPT ignores, the rate drops to 24.3%.
Translation: ranking helps if you are already part of the conversation, but it doesn't get you into it. The conversation is built in the lists.
We repeat the same 60 questions for 3 consecutive days just for this. If ChatGPT always recommended the same thing, measuring one day would be enough.
That's not the case, not by a long shot.
493 brand-question pairs observed in the 3 waves. The similarity between the lists of consecutive days (Jaccard index) averages 0.58 to 0.61 out of 1.
Almost half of the brands that ChatGPT recommends today were not recommended yesterday and will not be recommended tomorrow. Auditing your visibility in AI just once is like looking at a blurry photo of a moving object.
This connects with what Semrush documented at the source level: Reddit went from 60% to 10% of ChatGPT citations in six weeks.
We added the missing data: at the level of recommended brands, the landscape changes daily.
To understand these numbers, it helps to know how ChatGPT works with internal search.
The architecture is called RAG (Retrieval-Augmented Generation) and has two phases:
When AI teams evaluate a RAG, they measure the two phases separately: what it retrieves and how stable that retrieval is, and what it generates from the retrieved content.
This experiment is exactly that, applied from the outside to the most used RAG in the world:
The practical consequence: you cannot optimize "for the ChatGPT algorithm" as you did for Google's. You can optimize what its retrieval finds: the third-party content where your brand appears.
The logic is simple: ask the same question to Google and ChatGPT, on the same day, and compare who appears on each side.
About the statistics. Three decisions separate this experiment from an anecdote with graphs.
The unit of analysis is the question, not the citation. Citations from the same response go in a herd: if you treat each one as an independent observation, you artificially inflate significance. This is the common error in large-scale studies. All our calculations group by intention.
Bootstrap intervals. For each metric, we resampled the 60 complete questions 10,000 times and recalculated the result. The range where 95% of those recalculations fall is the confidence interval. When we say 73.0% with an interval of 65.9 to 79.9, we mean: if you repeated the experiment with other questions from these categories, you would expect to fall within that range.
Pre-closed protocol. The hypotheses, questions, and analysis rules were set and sealed with a hash before collecting the first data. No room to choose later the metric that fits best.
The three preregistered hypotheses were confirmed: the Ahrefs pattern replicates in Spanish, the direct path is minority (below the third we set as a threshold), and the indirect path is majority.
For volatility, we used the Jaccard index: it divides the brands that two lists share by all those that appear in either of the two. It is worth 1 if the lists are identical and 0 if they share nothing.
We do not ask you to believe us: check the data. We publish the complete dataset under CC0 license: the 360 captures, the 2,127 citations with positions, the 1,063 brand recommendations, the top 100 from Google of each wave, and the brand dictionary.
5 CSV files + brand dictionary · CC0 License · If you use it, we appreciate the link.Álvarez, S. and Cardozo, S. (2026). Experiment 5: overlap between the organic top 10 of Google and the brands recommended by ChatGPT in Spanish. CreceRank. 60 purchase intentions, 3 waves, July 2026. Available at crecerank.com
If you cite a specific data point, these are the three that summarize the study: 13.7% of direct path, 73.0% of indirect path, 44% of fleeting recommendations.
Yes, but not as you think. Ranking your own domain explains a minority of the recommendations (13.7% in our data). What really correlates is being mentioned within the listings and comparisons that rank in the top 10 organic: that's where 73% of the brands that ChatGPT recommends are. The SEO that works for AI is, above all, the SEO of other pages.
If you are a relevant brand in the category and your domain is in the top 10 organic, ChatGPT recommended you that same day in 77% of our observations. But don't confuse correlation with a gateway: 86% of the recommended brands were not in the top 10 with their own domain.
No. The comparison is against the classic top 10 organic: ads are excluded and the AI Overviews response was recorded as separate data, not counted in the ranking. We wanted to answer a specific question: whether traditional SEO gets you into ChatGPT or not.
RAG stands for Retrieval-Augmented Generation: the system first retrieves relevant documents (retrieval) and then drafts the response based on them (generation). ChatGPT with web search is a RAG whose retrieval is the search. This experiment audits that retrieval from the outside: what it retrieves, what it recommends from it, and how stable it is over time.
Faster than assumed. We repeated the same 60 questions for 3 consecutive days: only 35% of the brands appeared all 3 days, and 44% appeared only one day. That's why any visibility audit in AI done just once becomes outdated almost instantly.
Because the reference studies do not declare language or geography, and extrapolate from American English. Our Experiment 4 demonstrated that the sources that AI cites in Spanish are different. This experiment adds that the overlap between Google and ChatGPT in Spanish (8.7%) is even lower than the 12% measured in English.
Identify the listings and comparisons that rank in the top 10 organic for the commercial questions in your category and work to appear in them: this is the path through which 73% of the recommended brands enter. Maintain your SEO (the direct path exists and adds up), get reviews and mentions in the sources that ChatGPT consults, and monitor in series, not with isolated snapshots.
This study maps the general landscape. Your specific case (your brand, your competitors, your questions) is another experiment, and it's the one that matters. CreceRank runs it for you every day.
