murmur

murmur · sector study · September 2026

What AI tools answer when an HR team looks for benefits cards

25 questions asked across 4 AI tools, with 98 responses saved as screenshots. No question names a provider: these are the questions decision-makers type before requesting a proposal.

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: 5 names appear in more than half of responses

Caju is cited most often, in 80 of 98 responses, but the gap to those below is small: there are 5 providers above half, and there is a 30-point spread from the first to the last of them. AI distributes recommendations in this market, and a response naming only one provider is rare.

providerin how many of the 98 responsesshare of all mentionsaverage place in the listin how many of the 25 questions
Caju80 82%17.9%2.025
Flash75 77%16.7%2.325
Swile70 71%15.6%3.225
Pluxee (formerly Sodexo)59 60%13.2%3.825
Ticket51 52%11.4%4.022
Alelo48 49%10.7%3.823
iFood Benefits37 38%8.3%4.321
VR Benefits16 16%3.6%3.414
Alymente10 10%2.2%3.99
Bee Vale1 1%0.2%5.01
Vee Benefits1 1%0.2%7.01
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 11 providers total 448 mentions, and this column divides that total. The fourth is the place where the name usually appears in the list: 1.0 would mean always appearing first. Ben Visa Vale and Biz were measured and did not appear once, so they are not in the table.
02 · order within the response

Appearing is one thing; being the first name is another

AI almost never answers with just one provider: it averages 4.6 names per response. That is why where a name appears matters as much as appearing at all. Caju opens the list most often, with an average place of 2.0. iFood Benefits is at the other extreme: it appears in 37 responses and enters at 4.3rd on average, after the reader has already 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 arriving 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 changes in size depending on where the question is asked. Pluxee (formerly Sodexo) is the strongest case: it appears in 83% of Perplexity responses and 32% 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
23 responses
Caju25202015
Flash19202214
Swile23162110
Pluxee (formerly Sodexo)8151719
Ticket10141017
Alelo1017813
iFood Benefits166510
VR Benefits0349
Alymente6301
Bee Vale0010
Vee Benefits0001
04 · where the answer comes from

Being read by AI almost guarantees being cited by it

In 61 of the 98 responses, it is possible to see which sites the tool consulted. There were 93 different sites, and the most used of all is caju.com.br, Caju’s site, in 16 of the 61. Among the ten most used, 7 are sites belonging to the providers measured and 3 are outside sites, such as benway.com.br and www1.folha.uol.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 95% of the time. When it was not consulted, the name appeared in 37%. It is worth separating what appears in the text without competing for the sale: Elo, Mastercard, Sólides, TOTVS, Visa and Wellhub are card networks or HR software; they were measured and remain outside the table.

siteused as a source in how many of the 61
caju.com.br16
swile.com.br16
ticket.com.br14
benway.com.br13
www1.folha.uol.com.br11
pluxee.com.br10
beneficios.ifood.com.br9
beneficios.swile.com.br8
flashapp.com.br6
biz.com.br5
ajuda.swile.com.br4
alelo.com.br4
providersite used as a sourcename written in the response
Caju18 of 6160 of 73
Flash6 of 6153 of 73
Swile20 of 6149 of 73
Pluxee (formerly Sodexo)11 of 6142 of 73
Ticket14 of 6141 of 73
Alelo4 of 6140 of 73
iFood Benefits9 of 6132 of 73
VR Benefits1 of 6112 of 73
Alymente2 of 6110 of 73
Bee Vale0 of 610 of 73
Vee Benefits0 of 611 of 73
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 we captured, so it is excluded from this count, and only this count.
05 · question categories

Question categories make little difference at the top

The 25 questions covered five categories: general recommendations, company size, benefit type, HR operational challenges, and company circumstances. The leading names recur in almost all of them.

question category1st2nd3rd4th
general recommendation 23 responsesCaju 70%Flash 70%Pluxee (formerly Sodexo) 65%Swile 61%
by company size 16 responsesCaju 81%Flash 81%Swile 69%Pluxee (formerly Sodexo) 62%
by benefit type 20 responsesCaju 90%Flash 90%Swile 70%Ticket 60%
by HR pain point 20 responsesCaju 90%Flash 80%Pluxee (formerly Sodexo) 80%Swile 80%
by company situation 19 responsesCaju 79%Swile 79%Flash 63%Pluxee (formerly Sodexo) 37%

Differences between question categories emerge in the middle of the table, not at the top. For providers outside the leading positions, that is where an opening exists: a company size or benefit type for which AI already names other providers.

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 98. In Perplexity, 23 of 25 were retained: 2 captures failed. Each question was asked in a clean window, in a temporary conversation without history, from an IP in Brazil. We saved every screenshot.
  • No question names a provider. All ask for recommendations, as decision-makers ask before requesting a proposal.
  • 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.
  • Pluxee and Sodexo count as one company: the AI tools themselves write “Pluxee (antiga Sodexo)” — “Pluxee (formerly Sodexo)” in English.
  • VR was counted only when written as “VR Benefícios”. “VR” alone is the Portuguese abbreviation for meal voucher and appears in sentences that do not refer to the company, such as “modelo tradicional (VR/VA separados)” (“traditional model, separate meal/food vouchers”). Its figure is an undercount, never an overcount.
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 the 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 cartao de beneficios flexiveis para empresas
  2. 02qual plataforma de beneficios contratar para minha empresa
  3. 03me indique um cartao de beneficios corporativos
  4. 04qual o melhor cartao multibeneficios em 2026
  5. 05qual empresa de beneficios corporativos vale a pena
  6. 06melhor fornecedor de vale refeicao e vale alimentacao
  7. 07melhor cartao de beneficios para empresa pequena
  8. 08qual plataforma de beneficios indicar para uma startup
  9. 09melhor cartao de beneficios para empresa com 50 funcionarios
  10. 10qual fornecedor de beneficios para empresa com mais de 500 funcionarios
  11. 11melhor cartao de beneficios com auxilio home office
  12. 12qual cartao de beneficios inclui mobilidade e combustivel
  13. 13melhor plataforma de beneficios com saude e bem estar
  14. 14qual cartao de beneficios permite o colaborador escolher a categoria
  15. 15melhor cartao de beneficios que atende o PAT
  16. 16qual plataforma de beneficios integra com o sistema de folha
  17. 17qual fornecedor de beneficios cobra a menor taxa
  18. 18qual cartao de beneficios tem a maior rede de aceitacao
  19. 19qual plataforma de beneficios tem o melhor aplicativo para o colaborador
  20. 20qual empresa de beneficios tem o melhor atendimento para o RH
  21. 21vale a pena trocar o vale refeicao tradicional por cartao flexivel e por qual
  22. 22qual fornecedor indicar para migrar de plataforma de beneficios
  23. 23melhor cartao de beneficios para empresa com equipe remota
  24. 24qual plataforma de beneficios para quem tem funcionario em varios estados
  25. 25melhor cartao de beneficios para industria e trabalho presencial