If your brand appears when someone asks ChatGPT but disappears in Gemini, it’s not an error or bad luck. At CreceRank we ran 100 searches across four AI engines, three times each, and the result was clear: two models agree on only 9% of the brands they recommend. Each AI creates its own universe. Here are the 4 reasons why this happens and what you can do to be included in all of them.
More and more people are asking an AI "what is the best agency", "what software suits me", or "who should I buy from". The answer from that model determines which brands exist for that person and which do not. So we asked ourselves a simple question: if you make the same search to four different artificial intelligences, do they recommend the same brands?
To answer it without opinions, at CreceRank we chose a specific market with many players: marketing agencies in Argentina. It is the ideal category to see if the AIs converge or disperse. This is not the first time we have tested AI in this way. We already did it when analyzing what sources the AI uses in Spanish.
"best digital marketing agency in Argentina"
"marketing agency for e-commerce"
"marketing agency for fintech"
"marketing agency with good price-quality ratio"
AI models are probabilistic: each response can change. Without repetition, it is impossible to know if two models think differently or if the same AI is simply unstable. The three runs were what allowed us to separate the signal from the noise.
The average overlap between models is 9%. In other words, out of every ten brands that appear in a recommendation, nine do not repeat in the other model. AI Overview and Gemini are the most similar, with 18%, which makes sense because both are from Google and share the same underlying index. Perplexity, on the other hand, matches with all in less than 4%.
| Pair of models | Match (Jaccard) |
|---|---|
| AI Overview vs Gemini | 18% |
| AI Overview vs ChatGPT | 10% |
| ChatGPT vs Gemini | 9% |
| Gemini vs Perplexity | 4% |
| ChatGPT vs Perplexity | 3% |
| AI Overview vs Perplexity | 3% |
% of shared brands between each pair of models, average of 100 searches.
The most revealing aspect is not the average, but the structure. We gathered all the brands mentioned by the four models and it resulted in a universe of over 900 distinct names. Out of every 100 recommended brands, 81 appeared in only one engine and in none of the other three. Only 16 brands, 1.7%, appeared in all four at once. There is a small core that all recognize, but around it, there is a very long tail of brands that exist in the mind of only one engine.
You might think this only happens to small brands and that being a leader ensures your presence. That's not the case. Seonet, the most recommended brand in the entire study, appears on average in this way:
| Model | Searches where it appears (average per run, out of 100) |
|---|---|
| Gemini | 52 / 100 |
| ChatGPT | 42 / 100 |
| AI Overview | 40 / 100 |
| Perplexity | 5 / 100 |
Presence of the leading brand (Seonet) in 100 searches, average per run.
There are two readings. First, the leader does not always appear: in its best model, Gemini, it shows up in half of the relevant searches, and in ChatGPT and Google in four out of ten. Second, being a leader depends on each engine: Seonet dominates in Gemini and is almost invisible in Perplexity, with 5 out of 100. Adding the four engines, the number one reaches 82% of searches, never 100%, and appears in just 5% across all four at once.
If the number one brand, measuring correctly, does not exceed 82% considering all four AIs, any brand that looks at its presence with a single query is looking at itself in a broken mirror. Presence in AI is a distribution, not a yes or no.
Before comparing models with each other, it is worth asking something more basic: if I make the same search to ChatGPT three times in a row, does it give me the same brands? The answer is no. We measure the internal consistency of each model, that is, how similar one response is to the next in the face of the identical search.
Translated: if you ask ChatGPT the same thing twice, around three out of four brands it mentions change between one response and the other. The most stable is AI Overview, with 39%, and even then it does not reach half. Perplexity is the most chaotic, with 21%, and it only names brands in 38 out of 100 searches all three times.
Two models cannot match each other more than each one matches itself. If ChatGPT repeats itself by 26%, it is impossible for it to match Gemini by 90%. That 9% does not mean that the AIs think opposite things: it means that a good part of the disagreement is noise, not criteria. When we filter only the stable brands, the overlap rises from 9% to 12%.
If each model builds its own universe, it is worth asking what its preferred brands are. Here appeared the clearest division of the entire experiment. ChatGPT, Gemini, and AI Overview share a similar taste: they recommend local Argentine agencies, specialists, with a strong digital presence. Perplexity does the opposite and goes for the large global networks.
The contrast is brutal with a single example: Seonet, which appears in 70 of the 100 searches for Gemini, appears in only 9 for Perplexity. They are two different editorial criteria. Three models reward local authority and specialization. One rewards size and global brand recognition. For a medium local brand, that defines the strategy: you will enter ChatGPT, Gemini, and Google before Perplexity, where you compete against international holdings.
It’s not that your brand is worse for Gemini. It’s that Gemini measures something different than ChatGPT. Optimizing for a single engine leaves you out of the criteria of the other three.
The overlap is not even; it depends a lot on how the search is formulated. In generic searches, like "best marketing agency", the match rises to 13%: the AIs go for the safe bets and repeat the most well-known names. As soon as the search becomes specific, a particular service, a limited budget, a niche industry, the match drops to half.
The difference is statistically solid, it’s not just a sampling coincidence. And we close with a fact that sums it all up: when two models do recommend, they agree on who comes first only 9% of the time. Between ChatGPT and Perplexity, it’s 2%. The reading is straightforward: in the generic, you compete against everyone and the same ones always win. In the specific, the game is much more open, because no model has a consolidated answer yet.
If you care about appearing when an AI recommends, and you should care because more and more purchasing decisions start there, this is what changes the way of working.
What matters about the exercise is the method: repeat, count, and measure instead of settling for a single answer. It’s exactly the logic with which we at CreceRank measure the visibility of brands within AI, and the reason we set up the entire experiment in our own tool.
No. In a CreceRank experiment with 100 searches across four engines, two AI models agree on average on only 9% of the brands they recommend. 81% of the brands appear in only one model and only 1.7% in all four at once.
Because each AI engine uses different criteria for recommendations. ChatGPT, Gemini, and AI Overview tend to favor local and specialized brands, while Perplexity favors global brands with press authority. Additionally, the models are unstable, so a brand may appear in one query and disappear in the next.
Because AI models are probabilistic by design. ChatGPT agrees with itself only 26% of the time when repeating the same search, so it is expected that the list of brands changes between one query and the next. That’s why a single query is not representative.
It depends on your profile. If you are a local brand, AI Overview and Gemini are advisable because they favor local players and share indexing with Google. Perplexity favors global brands with press authority, so it requires a strategy of mentions and external citations.
By repeating each relevant search several times in each engine and recording how often your brand appears. A single query is not enough due to the instability of the responses. At CreceRank, we automate that measurement in ChatGPT, Gemini, AI Overview, and Perplexity.
Niche, at least to start. In generic searches, the agreement between models is 13% and established names always win. In niche searches, it drops to 7%, which means less consensus and more room for you to become the reference answer in your category.
What we did in this experiment (asking hundreds of times and measuring the result) is done by CreceRank with your brand: it tracks how you appear in ChatGPT, Perplexity, Google AI Overview and Gemini, with data from Spain and LATAM.
No credit card required · Prompts in Spanish · Data from Spain and LATAM
