murmur

murmur · sector study · September 2026

What AI tools answer when a retailer looks for a card terminal

25 questions asked across 4 AI tools, with 100 responses saved as screenshots. No question names a brand: these are the questions a business owner types before choosing.

This study does not pick a winner. The count is the same for everyone: the name is written in the response or it is not. Anyone appearing zero times is not included on this page, and no company commissioned, reviewed or paid for this study.

01 · who appears

No one dominates: 6 names appear in more than half of responses

Ton is cited most often, in 84 of 100 responses, but the gap to those below is small: there are 6 providers appearing in more than half of responses, with a 32-point spread from first to last. AI distributes recommendations in this market, and a response naming only one provider is rare.

providerin how many of the 100 responsesshare of all mentionsaverage positionin how many of the 25 questions
Ton84 84%16.3%2.425
Mercado Pago78 78%15.1%3.625
PagBank76 76%14.8%3.825
InfinitePay75 75%14.6%3.125
SumUp65 65%12.6%3.923
Stone52 52%10.1%3.023
Cielo37 37%7.2%4.221
Getnet20 20%3.9%5.014
Rede9 9%1.7%3.15
C6 Pay8 8%1.6%5.67
PicPay3 3%0.6%6.03
SafraPay2 2%0.4%7.52
Asaas1 1%0.2%7.01
Bin1 1%0.2%6.01
Cora1 1%0.2%6.01
Nubank1 1%0.2%7.01
Sicredi1 1%0.2%5.01
Vero1 1%0.2%6.01
How to read it. The second column shows how many responses contained the name. The third is the provider’s share of everything cited: the 18 providers total 515 mentions, and this column divides that total. The fourth is where the name usually appears in the list: 1.0 would mean always appearing first. Granito and Caixa were measured and did not appear once, so they are not in the table.
02 · order within the response

Appearing is one thing; being named first is another

AI almost never answers with just one provider: it averages 5.2 names per response. That is why where a name appears matters as much as appearing at all. Ton is named first most often, with an average position of 2.4. Getnet is at the other extreme: it appears in 20 responses with an average position of 5.0, after the reader has seen other names.

For providers in the middle of the table, this is where the difference lies between being considered and merely mentioned. A list has a beginning, and anyone at its end competes with names the reader has already seen.

03 · one tool is not another

Each tool has its own ranking

The same provider’s visibility changes depending on which tool is asked. Cielo is the strongest case: it appears in 56% of Claude responses and 8% of ChatGPT responses. For anyone outside the top positions, this means there are tools where they can compete and tools where they cannot.

providerChatGPT
25 responses
Claude
25 responses
Google
25 responses
Perplexity
25 responses
Ton20192223
Mercado Pago22182018
PagBank19182019
InfinitePay17181921
SumUp18141221
Stone8161513
Cielo214813
Getnet31142
Rede2412
C6 Pay0503
PicPay0003
SafraPay0002
Asaas0100
Bin0001
Cora0100
Nubank0001
Sicredi0001
Vero0100
04 · where the answer comes from

Being read by AI almost guarantees being cited by it

In 57 of 100 responses, it is possible to see which sites the tool consulted. There were 71 different sites, and the most used is maquinatop.com.br, in 14 of 57. Among the ten most used, 4 are sites of the providers measured and 6 are external, such as maquinatop.com.br and maquininhacerta.com.br.

What the tool reads also appears in what it writes: when the provider’s site was consulted, its name appeared in the response 98% of the time. When it was not consulted, the name appeared in 28%. It is worth separating what appears in the text without competing to sell the terminal: Reclame Aqui, a reputation site, and Ingenico, which makes the machine. Both were measured and remain outside the table.

siteused as a source in how many of the 57
maquinatop.com.br14
infinitepay.io13
maquininhacerta.com.br11
mercadopago.com.br9
maquininhareview.com.br8
calculadoradetaxas.com.br7
idinheiro.com.br7
blog.cielo.com.br6
guiafranquiasdesucesso.com6
blog.ton.com.br4
softpos24.com4
guiadoinvestidor.com.br3
providersite used as a sourcename written in the response
Ton5 of 5762 of 75
Mercado Pago9 of 5758 of 75
PagBank3 of 5756 of 75
InfinitePay13 of 5756 of 75
SumUp2 of 5753 of 75
Stone2 of 5737 of 75
Cielo6 of 5729 of 75
Getnet2 of 5716 of 75
Rede0 of 578 of 75
C6 Pay1 of 578 of 75
PicPay2 of 573 of 75
SafraPay0 of 572 of 75
Asaas0 of 571 of 75
Bin0 of 571 of 75
Cora0 of 571 of 75
Nubank0 of 571 of 75
Sicredi0 of 571 of 75
Vero0 of 571 of 75
What this section does not cover. This section considers only ChatGPT, Claude and Perplexity, which show the list of sites consulted. Google does not show that list in the responses captured, so it is excluded from this count only.
05 · asking for an alternative

Asking for an alternative widens the list, but only slightly

Five of the 25 questions do not ask for the best; they ask for another. In English translation, they ask things such as ‘is there a less-known one that is worthwhile?’ and ‘which one from a smaller company is worth it?’. The idea was to see whether the tool retains names it does not give when the question is direct.

It does, but only slightly. In the 24 general-recommendation responses there are 10 different brands; in the 20 asking for alternatives there are 14. The 4 appearing only there are Asaas, Cora, SafraPay and Vero, each appearing in one or two responses. At the same time, the big names rise in this category instead of making room: Stone goes from 38% to 75%; Mercado Pago from 75% to 95%.

For anyone outside the leading positions, the finding is harsh: the door exists but is narrow. Being remembered when a buyer requests an alternative is not the same as entering the main list, and most questions fall on that list.

question category1st2nd3rd4th
general recommendation 24 responsesTon 88%PagBank 79%Mercado Pago 75%InfinitePay 67%
alternative request 20 responsesMercado Pago 95%PagBank 85%Stone 75%InfinitePay 70%
by business type 20 responsesTon 85%Mercado Pago 80%PagBank 75%InfinitePay 60%
by fee and timing 20 responsesInfinitePay 95%Ton 95%SumUp 75%PagBank 70%
by revenue 16 responsesInfinitePay 88%SumUp 81%Ton 81%Mercado Pago 75%
06 · how we measure

How we measure

  • We asked the 25 questions in 4 tools, once each, in September 2026. That would yield 100 responses; we retained 100. Each question was asked in a clean window, temporary conversation without history, from an IP in Brazil. We saved every screenshot.
  • No question names a provider. All ask for recommendations, as a business owner asks before choosing.
  • We count only when AI writes the provider’s name in the response. When the name appears only in the list of sites consulted, it does not enter this count: that is something else, measured separately in section 04.
  • We count only what the tool wrote. Site cards, videos and map listings around the response are excluded because AI is not choosing the name there.
  • The names in the table came from the responses themselves. We did not begin with a ready-made list, so no outside provider ‘should’ be there: if it was not cited once, it does not appear.
  • Moderninha and Minizinha count as PagBank, while Point counts as Mercado Pago: they are terminal lines, and AI tools always write the company name alongside them.
  • Ton and Stone are the same company and were counted separately. AI tools know this and write ‘Ton, da Stone’ (Portuguese for ‘Ton, from Stone’), but recommend one or the other according to business size. Adding them would hide precisely which one the tool chooses. Both figures are in the table for anyone who wants to add them.
07 · limits

What this study does not say

  • Nothing about service quality. What is measured is what AI tools answer, not who has the best acceptance network or lowest fee.
  • Nothing about how much each thing is searched. The 25 questions reflect how decision-makers ask, not a sample weighted by search volume.
  • Nothing about what AI answers for you. Responses vary with the preceding conversation, region and day. This is a September 2026 snapshot in a clean window.
  • Nothing about how to change this result. This material measures; it is not an action plan.
08 · the questions

The 25 questions, in full

Original Portuguese research evidence, preserved as recorded.

  1. 01melhor maquininha de cartao para pequeno negocio
  2. 02qual maquininha de cartao vale a pena em 2026
  3. 03me indique uma maquininha de cartao boa e barata
  4. 04qual maquininha de cartao e a mais confiavel
  5. 05qual a melhor maquininha para comecar a vender
  6. 06qual maquininha tem o melhor custo beneficio hoje
  7. 07existe alguma maquininha menos conhecida que valha a pena
  8. 08qual maquininha alternativa as grandes do mercado
  9. 09qual maquininha de empresa menor compensa usar
  10. 10tem maquininha nova no mercado que esteja valendo a pena
  11. 11qual maquininha pouca gente usa mas e boa
  12. 12melhor maquininha para MEI
  13. 13qual maquininha para quem vende como autonomo
  14. 14melhor maquininha para restaurante e delivery
  15. 15qual maquininha para loja de rua
  16. 16melhor maquininha para quem vende em feira e evento
  17. 17qual maquininha cobra a menor taxa de debito e credito
  18. 18qual maquininha tem a melhor taxa de antecipacao
  19. 19qual maquininha nao cobra aluguel nem mensalidade
  20. 20qual maquininha cai o dinheiro na hora
  21. 21qual maquininha aceita pix com a menor taxa
  22. 22qual maquininha para quem vende pouco por mes
  23. 23qual maquininha para quem fatura alto e quer negociar taxa
  24. 24vale a pena trocar de maquininha e por qual
  25. 25qual aplicativo transforma o celular em maquininha