Why this study matters
26% of those surveyed by Cadem say they consult AI before buying. In each answer, AI names on average 2.3 brands (ChatGPT) or 2.8 (Gemini). This ranking measures which ones, by category.
Source: Cadem. (2026). [Survey on generative AI and purchase decisions in Chile; online panel, n = 1,000, August 6 to 10, 2026]. 26% consult AI before buying. Summary at anda.cl, September 7, 2026.
Methodology
What we measure
Visibility in AI: in how many answers each brand appears. This ranking measures each brand's visibility in ChatGPT and Gemini answers to purchase questions. It does not measure its Google position, its sales or its reputation.
SEO · search engines
- What it delivers
- A list of links the person browses through
- What competes
- Being on the first page for a keyword
- What is measured
- Position, impressions and clicks
- What moves it
- Content, links and technical signals of the site
AEO · answer engines (ChatGPT, Gemini)
- What it delivers
- A written answer that names two or three brands
- What competes
- Being named in the answer to a purchase question
- What is measured
- Rate: out of every 100 answers in the category, how many mention the brand
- What moves it
- What AI reads and cites: in this study, 58% company sites and 16% media
SEO: Search Engine Optimization · AEO: Answer Engine Optimization · the rate is the same visibility Batwise measures daily for its clients.
How we measure it
We asked AI what a shopper would ask.
35categoriesconsumer goods, retail, services and finance
60questions per categoryas a shopper would ask them, without naming brands
2AIs geolocated in ChileChatGPT and Gemini, on their web version without logging in
13daysfrom September 20 to October 2
2,100fixed questionsthe same ones throughout the study
100%geolocated in Chileevery answer, verified one by one
548brands on the listcurated list; name variants count toward the brand
The published figure: out of every 100 answers in the category, how many mention the brand. It is the same visibility the product measures.
Methodology: from the questions to the winner, in eight steps
- Universe. 35 categories, manufacturers or retailers without mixing them. Curated list of 548 brands that get counted; name variants count toward the brand.
- Questions. 60 per category, 2,100 in total, written as a shopper would ask them and without naming brands. Fixed from the start, with no changes afterwards.
- Capture. ChatGPT and Gemini, anonymous web geolocated in Chile. One answer per question and engine each day for thirteen days.
- Counting. Each brand on the list is counted once per answer. 98.1% precision over 210 answers reviewed by an independent language model.
- Rate. Out of every 100 answers with content in the category, how many mention the brand. ChatGPT and Gemini combined: 54,550 answers.
- Variability. 95% interval by resampling the 60 questions 2,000 times: the unit is the question, not the answer.
- Winner. Clear #1 if its difference with the #2 gives p below 0.05. If not, the tiebreaker is the brand named first most often.
- Validation. Question halves (95.7%), 20 seeds and 20,000 resamples (0 changes), independent review.
Validating the method: six tests of the statistical calculation
| What we tested and how | Result | What it means |
|---|
| We split the 60 questions of each category into two random halves, 200 times. When one half had a winner with a clear lead, we checked whether that brand is also #1 in the other | 95.7% | Winners with a clear lead do not depend on which questions were chosen (measured with the 13 days) |
| Stability of the 95% confidence interval (Monte Carlo error): 20 repetitions of the calculation, each with a different random seed and 20,000 resamples instead of 2,000 | 0 of 35 | No winner changes: they do not depend on the randomness of the calculation |
| Counting precision: an independent language model read 210 random answers, in batches of 25 and without seeing what the system had counted, and marked which brands each one names | 98.1% | Out of every 100 brands the system counted in an answer, 98 are actually named in the text of that answer; the other 2 are counting errors. The 72 pairs where the two disagree go to human review |
| Counting coverage: of all the real mentions of listed brands in those answers, how many the system found (before correcting it: 91.7%) | 98.0% | Out of every 100 times an answer names a listed brand, the system counts 98; it misses 2, and it is the same across all days |
| We calculated the relative error (coefficient of variation) of the leader's rate in each category | 7.5% | Median. In 34 of 35 categories the error is below 15%: the leader's rate is precise |
| We verified that every answer is located in Chile and comes from the same 2,100 questions, which did not change during the study | 100% | All 54,550 answers meet both conditions |
Only the second test refers to the 95% confidence interval; the 95.7% in the first one is a different measure.
Results
The 35 winners
Winning brands by category.
Findings
Finding 1: the winner is #1 in both ChatGPT and Gemini in 32 of 35 categories
32/35categories where the winner is #1 in both AIs
90%of the Top 5 matches between ChatGPT and Gemini
2.80brands per answer in Gemini, versus 2.26 in ChatGPT
The exceptions are Delivery apps, Fashion and Clinics: there the winner is #1 in Gemini, but not in ChatGPT.
ChatGPT and Gemini separately, 13 days, from September 20 to October 2.
Finding 2: of the sources ChatGPT and Gemini cite, 58% are company sites and 16% are news media
58%of citations are company sites
16%are media and editorial sites
1 in 3times ChatGPT cites the site of the brand it recommends; Gemini, 1 in 13
A citation is the link the answer shows as its source; we counted 153,817. When ChatGPT recommends a brand, 1 in every 3 times it links to its site. In pension funds and private health insurers, 4 in every 10 citations go to the official sites of the regulators that oversee those companies (Superintendencia de Pensiones and Superintendencia de Salud), not to the brands.
153,817 citations in 54,550 answers, 13 days, from September 20 to October 2.
Comparison / benchmark
Comparison with other studies
Cadem Marcas Ciudadanas 2026-I and AI have the same #1 in 10 of 15 comparable categories
Cadem surveys 18,000 people (May 2026) on each brand's positive presence, relevance and contribution. It publishes only the #1 of each category.
Same #1 · 10: Airlines (LATAM Airlines), Delivery apps (PedidosYa), Fuel (Copec), Fast food (McDonald's), Dairy (Colun), Marketplaces (Mercado Libre), Home improvement (Sodimac), Supermarkets (Lider), Department stores (Falabella) and Telecommunications* (Entel).
Different #1 · 5:
| Category | Cadem's #1 | That brand in AI | #1 in AI |
|---|
| Pension funds (AFP) | AFP Habitat | 61.3%, #3 by rate | AFP Capital, 69.0% |
| Banks | BancoEstado | 56.1%, #3 by rate | Santander, 65.9% |
| Clinics | RedSalud | 37.2%, #2 | Clínica Alemana, 30.4% |
| Pharmacies | Dr. Simi | 30.7%, #4 by rate | Cruz Verde, 60.1% |
| Car brands | Tesla | in 16 of 1,560 answers (1.0%) | Toyota, 64.5% |
Retail, fuel and transport brands match. Where they differ, Cadem's brand is in AI but behind another one, except Tesla, which AI almost never names.
cadem.cl/marcas-ciudadanas · *Telecommunications compared with Mobile carriers · Fragrances is not compared: Cadem measures stores (Preunic), we measure brands.
Criteria Total Brands 2026 and AI have the same #1 in 9 of 15 comparable categories
Criteria surveys 4,400 people (June and July 2026) on each brand's relevance across 12 attributes, on a scale from 0 to 1,000. It publishes only the #1 of each category.
Same #1 · 9: Airlines (LATAM Airlines), Delivery apps (PedidosYa), Fuel (Copec), Marketplaces (Mercado Libre), Supermarkets (Lider), Department stores (Falabella), Universities (Universidad de Chile), Home* (Sodimac) and Telecommunications* (Entel).
Different #1 · 6:
| Category | Criteria's #1 | That brand in AI | #1 in AI |
|---|
| Pension funds (AFP) | AFP Modelo | 59.9%, #5 by rate | AFP Capital, 69.0% |
| Banks | BancoEstado | 56.1%, #3 by rate | Santander, 65.9% |
| Fast food | Burger King | 40.7%, #2 by rate | McDonald's, 50.3% |
| Pharmacies | Dr. Simi | 30.7%, #4 by rate | Cruz Verde, 60.1% |
| Private health insurers (Isapres) | Consalud | 58.3%, #3 by rate | Colmena, 67.8% |
| Fashion | H&M | 40.5%, #2 by rate | Falabella, 41.6% |
Dr. Simi is the #1 of all of Total Brands (697 points) and in AI it ranks #4 in Pharmacies. BancoEstado is #1 in Cadem and in Criteria, and #3 by rate in AI.
criteria.cl/total-brands · *Home compared with Home improvement, Telecommunications with Mobile carriers · Health (Achs Salud) is not compared.
Chile 3D 2026 (NIQ) and AI have the same #1 in 1 of 3 comparable categories
Chile 3D (NIQ) measures prestige, presence and affection with more than 5,200 people. NIQ does not publish results: these three winners come from the brands' own press releases.
Same #1 · 1: Telecommunications* (Entel).
Different #1 · 2:
| Category | Chile 3D's #1 | That brand in AI | #1 in AI |
|---|
| Clinics | RedSalud | 37.2%, #2 | Clínica Alemana, 30.4% |
| White goods* | Samsung | 51.9%, #2 by rate | LG, 54.0% |
Chile 3D is the only one of these surveys that measures affection. Its three brands are in AI's Top 2 for their category.
nielseniq.com · press releases from RedSalud, Entel and Samsung · *Telecommunications with Mobile carriers, White goods with Home appliances.
About us
About Batwise
Batwise was born to understand how the rise of technology is changing consumption and behavior paradigms, and in particular how AI is finding and recommending brands. Its value proposition is to give brands visibility into how AI sees and recommends them, and to turn those findings into concrete actions. When comparing these results with studies such as Marcas Ciudadanas, Chile 3D and Total Brands, the most interesting part was not the matches but the gaps: brands that are very strong among consumers do not always have the same presence in ChatGPT or Gemini. That is where Batwise helps: it identifies which signals AI is reading and what brands can do so that their position in these models better reflects their real relevance.
Batwise is a Chilean "AI Search Intelligence" platform backed by Start-Up Chile. Batwise monitors daily the conversations relevant to each business, analyzes where and how a brand appears, which competitors gain visibility and why, and turns those signals into an actionable plan its agents can execute.
This study was carried out in collaboration with Jean Sepúlveda, PhD in Economics and researcher at the School of Economics and Business of UDD.
