Trust in AI: 55% in Brazil vs. 46% worldwide (KPMG)

Trust in AI in Brazil is above the world average: 55% of Brazilians say they are willing to trust artificial intelligence systems, against 46% across 47 countries, according to the study Trust, attitudes and use of artificial intelligence: A global study 2025, led by the University of Melbourne with KPMG. The survey reached 48,340 people online between November 2024 and mid-January 2025, about a thousand per country, and treats trust as the willingness, at least in part, to rely on what AI produces and to share information with it. The same study shows that, between 2022 and late 2024, Brazilians’ willingness to rely on AI systems fell from 67% to 53%, while the share who feel worried rose from 49% to 75%, the largest jump among the 17 countries compared. I read these numbers as a license with conditions. Brazilian customers accept talking to AI, and 84% of people in the global study say they would trust a system more if a person could correct, override or challenge what it decides. For anyone using AI in customer service, sales and marketing, that condition weighs more than the trust percentage.

At Iasbeck&Co we implement CRM and AI agents in B2B operations, and the question I hear most before an agent starts talking to customers is whether customers will accept it. I gathered here the surveys with sample, method and date, and my reading for executives, team leaders and HR. Brazilians’ use of AI is covered in Brazilians and AI, and the reaction to automated service in consumers and AI in customer service.

How much do Brazilians trust AI?

In the Brazil snapshot of the University of Melbourne and KPMG study, 55% of Brazilians are willing to trust AI and 86% accept or approve of its use. That puts Brazil ninth among the 47 countries, behind Nigeria (79%), India (76%), Egypt (71%) and China (68%), and ahead of the United States (41%), Germany (32%) and Japan (28%). Using the IMF classification, 57% say they are willing to trust AI in emerging economies, and 39% in advanced ones.

Indicator (KPMG and University of Melbourne, 2025)	Brazil	Average of 47 countries
Willing to trust AI	55%	46%
Accept or approve of AI use (at least to a moderate degree)	86%	72%
High acceptance	44%	33%
Concerned about negative outcomes from AI	79%	79%
Have experienced or observed a negative outcome	43%	43%
Have formal or informal AI training	47%	39%
At work, have relied on AI output without checking it	69%	66%
At work, have made mistakes because of AI	49%	56%

Source: global report and Brazil snapshot. The global training figure is the complement of the 61% with no training at all.

Brazilians trust AI more than the average and worry about it in the same proportion, and the country snapshot describes them as optimistic, excited and worried. The risk most Brazilians say they have already felt is the loss of human interaction and connection, cited by 54%. In the same snapshot, 66% say AI regulation is necessary and only 19% know the rules in place in the country, the subject of AI regulation.

What “trust” means in this survey

The question was “how willing are you to trust AI”, on a seven-point scale, and the percentage adds up the top three answers. Each person answered about one randomly assigned use, among generative AI, AI in human resources, AI in healthcare and AI in general, and trust changed with the use: across the 47 countries, 52% trust AI for diagnosing disease, 44% trust generative AI and 42% trust AI that screens job applicants. For a company, that difference matters more than the average, because the customer judges a concrete application, such as the assistant that answers questions about an invoice, and the survey measured no case of that kind.

Is trust in AI rising or falling in Brazil?

The surveys point in opposite directions. KPMG measured 17 countries in 2022 and again in late 2024, and on average the share who see AI as trustworthy fell from 63% to 56%, willingness to rely on it fell from 52% to 43%, and worry rose from 49% to 62%. In Brazil, willingness to rely on AI went from 67% to 53%, one of the two largest drops, alongside Japan, and worry went from 49% to 75%. The authors suggest that, with more use, people learn the tool’s limits and think harder before relying on it.

The 2025 Edelman Trust Barometer Flash Poll, a special edition of the barometer devoted to AI, shows the opposite. In an online survey run from October 17 to 27, 2025, with about a thousand people in each of five countries, 67% of Brazilians said they trust AI, 14 points above what Edelman measured in November 2023. The question is different (“how much do you trust this technology”, on a nine-point scale), so is the time window, and Edelman asks about AI in general, with no specific use.

My reading puts the two series together. Brazilians came to trust AI more as a technology, because they use it and see results, and became more cautious about relying on it for a concrete decision, because they have also seen it fail. Edelman finds the effect of knowledge: being well informed about AI raises the likelihood of embracing the technology by 17.45%. For an operation, this means customers arrive willing to try and with no patience for a mistake nobody fixes.

Brazil, the United States and the world: what Edelman and Pew say

In the Edelman poll, trust in AI ranges from 87% in China and 67% in Brazil to 39% in Germany, 36% in the United Kingdom and 32% in the United States. On the growing use of AI, 35% of Brazilians say they embrace it and 24% reject it, and in the United States the ratio flips, with 17% against 49%.

The Pew Research Center complicates the picture. In its 2025 survey of 25 countries, with face-to-face interviews in Brazil, 48% of Brazilians said they are more concerned than excited about the growing use of AI, 37% equally concerned and excited, and 10% more excited, according to the results by country. The median across the 25 countries is 34%, and Brazil sits in the most concerned group, with the United States, Italy, Australia and Greece.

Survey	Question	Brazil	United States	Reference
KPMG and Melbourne (Nov 2024 to Jan 2025)	Willing to trust AI	55%	41%	46% across 47 countries
Edelman Flash Poll (Oct 2025)	Trust AI	67%	32%	87% in China
Edelman Flash Poll (Oct 2025)	Embrace or reject the growing use of AI	35% embrace, 24% reject	17% embrace, 49% reject	China: 54% embrace
Pew (Jan to Apr 2025; US in Jun 2025)	More concerned than excited	48%	50%	Median of 25 countries: 34%

Pew’s question forces people to pick a dominant feeling, while KPMG’s measures a willingness that coexists with worry. Method matters too: KPMG and Edelman use online panels, and KPMG notes that, in emerging economies, online respondents tend to be more educated than the population. Pew interviewed people face to face in Brazil and reached those who use the internet less, a group that reports more concern.

What Americans say about companies

In Pew’s September 2025 survey, with 5,023 adults, 50% said they were more concerned than excited, up from 37% in 2021, 76% say it is extremely or very important to know whether text, a photo or a video was made by AI or by a person, and 53% are not confident they can tell the difference. In an earlier Pew survey, with 5,410 adults in August 2024, 59% said they have little or no confidence in US companies to develop and use AI responsibly.

Why do Brazilians trust AI and worry about it at the same time?

In the Brazil snapshot, 91% expect benefits from AI and 80% say they have already felt some, while 79% are concerned about negative outcomes, and across the 47 countries the loss of human connection worries 83% and is the risk most people say they have already felt (55%). Trust also depends on who uses the AI. In the KPMG study, 39% of people in emerging economies have high confidence in commercial organizations, such as retailers and banks, to use AI in the public’s interest, and 25% have low confidence. In advanced economies, the ratio flips, with 23% against 40%. In the Edelman poll, 78% of Brazilians trust “someone like me” to tell the truth about generative AI, and 54% trust CEOs. Companies enter the conversation with some credit in Brazil, and with less credit than the experience of someone the customer knows.

Summary table of the surveys cited

Source	Year	Sample and method	Main finding
Gillespie and colleagues, University of Melbourne and KPMG	2025	48,340 people in 47 countries, online panel, 1,001 to 1,098 per country, Nov 2024 to Jan 2025	55% of Brazilians and 46% worldwide willing to trust AI; 83% would trust more with assurance mechanisms
Edelman Trust Barometer Flash Poll	2025	Over 5,000 people in 5 countries, online, Oct 17 to 27, 2025, margin of 3.1 points	67% of Brazilians trust AI, against 32% of Americans; up 14 points in Brazil since 2023
Pew Research Center, 25 countries	2025	28,333 adults in 24 countries, Jan to Apr 2025, face to face in Brazil, plus 5,023 in the US	48% of Brazilians and 50% of Americans more concerned than excited
Pew Research Center, Americans and AI	2025	5,023 adults, probability panel, Jun 2025	76% want to know whether content was made by AI
Pew Research Center, public and experts	2025	5,410 adults in Aug 2024 and 1,013 experts	59% have little or no confidence in companies

What these data do not say

All the numbers are self-reported, and behavior in front of an assistant that gets an invoice amount wrong may be different. No survey measured trust in a specific company or channel, and KPMG did not include customer service among the randomly assigned uses. The data point closest to your operation is the one you measure with your own customers.

The questions are not equivalent either. Willingness to trust, trust in the technology and dominant feeling measure different things, which is why KPMG shows a decline and Edelman a rise in the same Brazil, and the 55% from one cannot be compared with the 67% from the other. Since online panels in emerging economies reach more educated people, the Brazilian trust figure may be higher than what the whole population feels.

The relationships reported are correlations: people who know more about AI trust it more, and people who already trust it also tend to use it more. KPMG funded the study, and the report states that the university team conducted the research independently. Edelman is a communications firm that publishes the barometer through its research institute. Fieldwork runs from late 2024 to October 2025, and attitude numbers age fast.

What changes for customer service, sales and marketing?

The most useful part of the KPMG study for operators is the list of conditions. The researchers asked what would make people trust an AI system more, and 83% agreed with the set of mechanisms. Each item becomes a decision before you switch on an agent.

Condition for trust (KPMG, % who would trust more)	How it shows up in the operation
Right to opt out of having one’s data used by the system (86%)	A clear option to keep one’s data out of personalization and agents
Accuracy and reliability are monitored (84%)	Sample audits of the agent’s answers, with a correction rate
Employees are trained in responsible and safe use (84%)	Training for those who operate and those who review, before the agent goes live
Human intervention to correct, override or challenge the output (84%)	A path to a person in every channel and review before a proposal goes out
Clarity on who is accountable when something goes wrong (82%)	A named owner for each agent
Assurance by an independent third party (74%)	External audit, which weighs less than the items above

Customer service: say it is AI and guarantee the person

In customer service, human intervention weighs the most, tied to the risk Brazilians most often say they have already felt. The assistant introduces itself as an assistant in the first message, says what it can solve and shows how to reach a person without the customer repeating the story. The handoff rules are in AI in customer service. The data add a metric: besides the automated resolution rate, track how many customers asked for a person and how long they waited.

Sales: the salesperson signs what the AI wrote

In B2B sales, the buyer’s trust lies with the salesperson and the company, and a wrong number in a proposal written with AI spends that credit in one go. The 59% of Americans who distrust companies’ use of AI suggest that customers blame the company for the mistake, not the tool. Everything the AI writes that reaches the customer goes through the salesperson, who checks numbers, dates and promises and answers for the text, and meeting recording goes in the invitation, with the option to decline. The kinds of errors AI makes and how the reviewer finds them are covered in AI hallucination.

Marketing: data, personalization and synthetic content

In marketing, the 86% condition speaks to personalization: customers accept more relevant communication and want to be able to say no to the use of their data. Pew’s data calls for another decision, because 76% of Americans want to know what was made by AI and more than half cannot tell. A synthetic image presented as a photo of a real customer burns trust when someone finds out, and the team needs to write down where it flags the use of AI, as in generated images and chats.

The team’s trust: when the problem is trusting too much

The KPMG study shows a risk from the other direction. Among Brazilians who work, 69% have relied on AI output without checking its accuracy, 49% have made mistakes at work because of it and 51% have presented AI-generated content as their own. Across the 47 countries, those who do not check add up to 66%, and those who made mistakes, 56%. The customer distrusts as a precaution, part of the team trusts without criteria, and the two meet when a wrong answer goes out under the company’s name. The team needs calibrated trust: knowing in which task the AI fails and what to check before sending. The written rule for that is in AI use policy, and the 47% of Brazilians with some AI training show how much work is left.

What leadership does with these numbers

1. List the points where AI reaches the customer. Website and WhatsApp assistant, agent-drafted emails, proposals, campaign images and meeting recordings, each with an owner.

2. Write the transparency sentence for each channel. Define how the assistant introduces itself and what the invitation says about recording, with legal review where there is a legal or contractual obligation.

3. Guarantee the path to a person and measure it. Track requests for a human, waiting time and complaints from people who could not get past the bot.

4. Name who is accountable for each agent. The agent owner answers for the instructions and permissions, and the area’s leadership answers for what it released without item-by-item approval.

5. Audit by sample every week at the start. Record the correction rate and space out the audits once the rate drops and stays low.

6. Measure your customers’ trust. Add a question about AI-assisted service to the satisfaction survey and compare customers served by the agent with those served by people.

7. Review every quarter. Take the indicators from steps 3, 5 and 6 to the executive team and decide where to expand use and where to put a person back.

What HR and L&D do with these data

HR sees the problem from both sides of the KPMG survey. On one side, two in three professionals have used AI without checking the output, and half have presented AI-generated content as their own. On the other, customers say they would trust more if they knew the company trains its people to use AI responsibly, an item 84% pointed to in the global study. Training teaches the team to doubt with method, checking numbers, sources and promises, and to explain to the customer what the AI did and what a person reviewed. To measure it, HR tracks the correction rate in audits, incidents reported by the team and the trust question in the satisfaction survey, before and after the training.

The data show Brazilian customers willing to trust AI and watching for three things: knowing it is AI, reaching a person and having someone accountable for mistakes. Preparing the operation for that is the work of the in-company applied AI training, built on the company’s processes and channels, with groups for leadership, customer service, sales and marketing. It delivers the prompt library by area and the agent architecture, with what each agent reads, what it runs on its own and who approves the rest.

Frequently asked questions

Do Brazilians trust AI?

They trust it more than the world average. In the University of Melbourne and KPMG study, 55% of Brazilians say they are willing to trust AI, against 46% across 47 countries, and in the October 2025 Edelman Flash Poll, 67% say they trust it. Worry is also high: 79% are concerned about negative outcomes, and in Pew’s survey 48% say they are more concerned than excited.

Is trust in AI rising or falling?

It depends on the question. KPMG measured a drop in Brazilians’ willingness to rely on AI, from 67% to 53% between 2022 and late 2024, and Edelman measured a 14-point rise in trust in the technology between November 2023 and October 2025. Both series point to a public that uses AI more and has become more selective about when to rely on it.

Why do Brazilians trust AI more than Americans?

The studies show the pattern without proving the cause. KPMG finds more trust, use and AI training in emerging economies, and Edelman links trust to use and knowledge. Part of the gap may also come from method, because online panels in emerging countries tend to reach more educated people.

Does a company have to tell customers it is using AI?

The legal obligation depends on the case, the channel and the contracts, and the company’s legal team is responsible for that call. The attitude data recommend telling them: 76% of Americans say it is very important to know whether content was made by AI, and people trust more when they know who is accountable for the system.

Should customers have the option to talk to a person?

In the data, it is one of the most cited conditions for trust: 84% of people in the global study would trust more a system that lets a human correct, override or challenge the output. In customer service, that means a path to a person in every AI channel, with the case summary, and a metric for how long people wait.

How do I prepare the team to use AI with customers without losing their trust?

Start with review: the team learns what to check before sending, such as numbers, dates, promises and sources, and practices with the company’s real cases. Then train the conversation with the customer, with each channel’s transparency sentence and the handoff to a person. Measure the effect through the correction rate in audits and the trust question in the satisfaction survey.