LATAM already uses AI more than the United States. What's missing isn't adoption, it's direction

Published: June 2026 · By Wen López

The short answer: the region is not behind on AI, it's ahead. Mexico (66%), Brazil (74%) and Argentina (65%) surpass the United States (40%) in generative AI usage, according to the Google and Ipsos study with 21,000 interviews across 21 countries. What's missing is not adoption: it's systems. 71% of Mexicans say they want to do more with AI but lack confidence in how to use it, the highest percentage of the 21 countries measured. And the experimental evidence is consistent: the pattern that wins is a human directing plus AI executing, not AI alone. In this article: what to actually automate in a small startup, with serious studies, and what not to believe.

The LATAM paradox: high adoption, low systems

The Our Life with AI study (Google and Ipsos, ~1,000 interviews per country) breaks the cliché of the lagging region: Mexico uses generative AI 26 points more than the United States. The Latin American Artificial Intelligence Index (ILIA 2025, by CENIA with CEPAL and the IDB) adds another layer: LATAM generates 14% of global visits to AI tools with only 11% of the world's internet users. The region over-uses AI.

And here is the paradox: that same region receives barely 1.12% of global AI investment. And 71% of Mexicans admit they want to do more with AI but need more confidence in how to use it.

Using ChatGPT is not the same as operating with AI. The difference between the two is a system: defined workflows, human direction and measurement. That is exactly what the experimental evidence confirms.

What to actually automate (what serious experiments say)

Not vendor opinions: controlled, published experiments with methodology. This is what they show, in order of strength:

1. Customer responses. The strongest causal study that exists (Brynjolfsson, Li and Raymond, published in the Quarterly Journal of Economics, 5,179 support agents): AI raised productivity 14% on average, and 34% for novices. For experts the effect was nearly zero. AI levels small teams upward.

2. Routine writing. A controlled experiment published in Science (MIT, 453 professionals): 40% less time and 18% higher quality on professional writing tasks. The effect was larger for the less skilled writers.

3. SEO content with human editing. The only peer-reviewed case where AI plus human beat the experts on the real business outcome: AI-generated content with light human editing outperformed SEO specialists in actual rankings (Reisenbichler et al., Marketing Science).

4. Ad copy volume. An MIT experiment with 2,234 participants and a field test with 4.9 million real impressions: the human+AI team produced 50% more ads per person and better text. Key detail: AI alone lost against the duo. The human kept choosing the angle and the image.

Notice the pattern? The same configuration wins in all four studies.

In every serious experiment the same formula wins: human directing, AI executing. AI alone loses. The human alone goes slower. The system is the duo.

It is the formula we run the Lab with and the one we implement in client teams: content engines, landing pages, agent workflows and analytics, always with human direction and measurement.

What not to believe (the part agencies don't tell you)

The same evidence standard applies to the hype. Three data points to keep your feet on the ground:

• Agents are at their peak of expectations. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, and estimates that of thousands of vendors selling "agents", only a fraction offers real ones. It is a prediction, not a measurement, but the direction of the warning is clear.
• The abandonments are already happening. S&P Global surveyed 1,006 professionals: 42% of companies abandoned most of their AI initiatives in 2025, versus 17% the year before. On average, almost half of proofs of concept are discarded before reaching production.
• Even perception deceives. In a METR experiment with expert developers, participants felt 20% faster using AI when they were actually 19% slower. Small sample (16 people) and a specific context, but the lesson is uncomfortable and useful: without measurement, you don't know if AI is helping you.

Why do projects fail? RAND's research on AI failures found that cause #1 is not the technology: it is choosing the wrong problem from leadership. Starting with the tool instead of the bottleneck.

Where to start in a small startup (without an AI department)

With the evidence in hand, this is the order we recommend for a team of 2 to 15 people:

1. Choose the bottleneck, not the tool. Which task eats your hours every week? That's the candidate. (RAND: projects die from choosing the wrong problem.)
2. Start where the evidence is strongest: customer responses, routine writing, content with human editing, ad copy variants.
3. Keep human direction in the loop. A person signs off on what reaches the world. It is the winning formula in the experiments and, by the way, the only thing the law protects.
4. Measure before and after. Without a baseline there is no way to know if you improved (remember the developers who felt faster while being slower).
5. Visibility is part of the system. Operating better is useless if nobody finds you: search already changed shape and your presence in AI answers is a channel you work like any other.

One more data point to close: according to Anthropic's Economic Index (based on millions of real Claude conversations), creating marketing materials is already the second largest task across all API usage, and 77% of that usage is directed automation, not chat. The startups already operating this way are not the future: they are your current competition.

You may also like

• 2026-06Who owns AI-generated content? What the law actually says in 2026
• 2026-06How to show up in ChatGPT and Google's AI summaries: the real data
• 2026-06How much does a commercial cost in 2026? Nobody knows (and that's the problem)

Frequently asked questions

Is LATAM behind in AI adoption?

No. Mexico (66%), Brazil (74%) and Argentina (65%) surpass the United States (40%) in generative AI usage according to Google and Ipsos (21,000 interviews across 21 countries). The region also over-uses AI tools relative to its internet population (ILIA 2025). What's missing is turning that usage into working systems.

Which marketing tasks should be automated first with AI?

The ones with the strongest experimental evidence: customer responses (+14% productivity, +34% for novices), routine writing (40% less time), content with human editing (beat SEO experts in rankings) and ad copy variants (+50% volume). Always with final human review.

Do AI agents work for small companies?

With calibrated expectations. Gartner predicts more than 40% of agentic projects will be canceled by 2027, and 42% of companies already abandoned most of their AI initiatives (S&P Global). The ones that work start from a concrete bottleneck, keep human direction and measure results from day one.

Why do AI projects fail?

According to RAND, cause #1 is choosing the wrong problem: leadership defines poorly what needs solving and starts with the tool instead of the bottleneck. The technology is usually the part that works.

---

Sources: Google · Ipsos, Our Life with AI (topline PDF) · ILIA 2025, CENIA · CEPAL · IDB · Brynjolfsson, Li and Raymond, QJE · Noy and Zhang, Science · Reisenbichler et al., Marketing Science · Ju and Aral, MIT · Gartner, agentic AI prediction · S&P Global, Voice of the Enterprise · METR, developer study · RAND, why AI projects fail · Anthropic Economic Index · HubSpot State of Marketing 2026