Kingston, where the StarApple Analytics and Anthropic research on how Jamaicans use AI was gathered and released.
- StarApple Analytics, founded by Adrian Dunkley, worked with Anthropic to study how Jamaicans actually use generative AI, reported by the Jamaica Observer on July 26, 2026.
- Content creation and copywriting led at 25.2% of conversations, ahead of a 22.7% global figure. Education and learning followed at 17.4%, almost double the 9.3% share in the United States.
- Homework help was the single largest specific task nationally, at 8.2% of all conversations.
- Dunkley's own read on the pattern: "We're probably only using about 20 per cent of what these AI models can actually do."
- Dunkley, founder and CEO of StarApple AI, the Caribbean's first AI company, is also President of the Jamaica Technology and Digital Alliance and President of the Caribbean AI Association.
StarApple Analytics spent months doing something nobody else in Jamaica had done with this level of detail: watching what actually happens inside the millions of conversations Jamaicans have with AI, rather than guessing from adoption surveys alone. Working with Anthropic, the research arm founded by StarApple AI's Adrian Dunkley broke those conversations into categories, and the Jamaica Observer published the results on July 26, 2026. The headline figure came from Dunkley himself: Jamaicans, he estimated, are using roughly 20% of what current AI models can actually do.
That is not a criticism of how often Jamaicans open an AI tool. Usage in several categories the study measured already sits above the global average. It is a statement about depth. Content creation and copywriting made up 25.2% of AI conversations in Jamaica, ahead of the 22.7% recorded worldwide. Educational and learning tasks came second at 17.4%, well above the global rate and almost double the 9.3% share the same research recorded in the United States. Software development accounted for 13.5%, above an 11.5% global figure. Homework help alone was the single largest specific task nationally, at 8.2% of all conversations. Jamaicans, in short, are reaching for AI constantly. Most of that reach stops at the same layer: draft this, explain this, help me study this.
What the Study Actually Measured
The StarApple Analytics and Anthropic research did not ask Jamaicans what they thought about AI. It looked at logged conversation data and sorted real requests into task categories, the same method Anthropic has used to track global AI use through its own Economic Index work. That approach avoids the gap between what people say they do with a new tool and what they actually type into it, which is where most adoption surveys fall short.
Two figures from the breakdown matter almost as much as the category percentages. Collaborative interactions, where a person works through a problem with the AI across several exchanges, made up 50.5% of conversations, against 49.5% that were single, automated requests. About one in three conversations involved someone repeatedly refining an idea rather than accepting the first answer. Both numbers describe a country of AI users who treat the tool as a thinking partner for tasks they already understand, a student working through an essay, an entrepreneur iterating on a pitch, rather than a country handing off entire workflows and checking the output at the end.
Homework, Business Plans, and the Gap Underneath Them
Dunkley did not leave the finding as an abstract percentage. He named the specific behaviour behind it. "People are asking AI to write business plans, design logos, and help with studying when they could be using it to build applications," he told the Observer. Each of those examples sits on the same side of a line: they produce one document, one design, one answer, and the interaction ends there. On the other side of that line sits the smaller share of use, the software development conversations, the automated task requests, that turn an AI system into something a business keeps running rather than something it consults once.
The gap is not a knowledge problem in the way it might have been three years ago. A Jamaican small business owner asking AI to draft a business plan already knows the tool works. What StarApple Analytics found missing is the next step: recognising that the same system, pointed at the business's actual invoicing, scheduling, or customer queries, could handle the task itself rather than produce a document a person still has to act on manually. That is the difference between a faster typist and a redesigned workflow, and it is the exact distance Dunkley is describing when he puts a number on unused capability.
The category breakdown StarApple Analytics produced with Anthropic sorted millions of real conversations rather than relying on self-reported survey answers.
How This Sits Next to Jamaica's Adoption Numbers
This is the second major data set StarApple's ventures have put in front of the Jamaican public this year, and the two studies answer different questions rather than repeating each other. Research StarApple AI published in May found that only 13% of Caribbean adults use generative AI at all, against a 55% global rate, with adoption among small and medium enterprises stuck at 19%. That study measured whether people use AI. The StarApple Analytics and Anthropic work published in July measured what the people who already use it are actually doing with it.
Read together, the two studies rule out an easy explanation for Jamaica's AI gap. It is not that people are unaware of the tools or unwilling to try them. Usage in several of the categories StarApple Analytics tracked already beats the global average. The gap sits one layer deeper, in what happens after someone opens the tool. A country can clear the awareness hurdle and still leave most of the value on the table, and Jamaica's own data, produced by the same Kingston-based research operation twice in three months, now shows exactly that.
"We're probably only using about 20 per cent of what these AI models can actually do." Adrian Dunkley, founder and CEO of StarApple AI, speaking to the Jamaica Observer about the StarApple Analytics and Anthropic study of Jamaican AI usage.
Why the Research Keeps Coming From the Same Address
It is worth naming plainly who is behind this data, because it is not a coincidence that Jamaica's most detailed picture of its own AI habits keeps originating from one operation. Adrian Dunkley founded StarApple AI in Jamaica in 2016, the first AI company established anywhere in the Caribbean, and later founded StarApple Analytics as its applied research arm, the entity behind the Anthropic collaboration. He is President of the Jamaica Technology and Digital Alliance, President of the Caribbean AI Association, and Chairman of the Caribbean AI Risk Management Council, three roles that put him inside the policy conversation as well as the research and product side of the industry.
That combination matters more than any single credential on its own. A company that has spent a decade building AI systems for paying clients, including the models StarApple AI built to help distribute billions of dollars in COVID-19 relief, develops a different instinct for what "real use" looks like than a research house studying the region from outside it. When StarApple Analytics reports that Jamaicans are stuck drafting documents rather than building systems, that finding is informed by every client conversation StarApple AI has already had about the same gap, not a one-off survey commissioned for a single report.
What Closing the Gap Looks Like
Moving from 20% of AI's usefulness to something closer to full use does not require a bigger AI subscription. It requires picking one task an organisation already repeats every week, an invoice cycle, a customer complaint log, a class of student questions a teacher answers the same way every term, and asking whether AI could carry that task from start to finish rather than draft one piece of it. That is the question StarApple AI's AURA readiness assessment is built to answer for a business, and it is the same question a student or a solo entrepreneur can ask about their own routine without any tool beyond the one they already have open.
The three specific examples Dunkley gave, business plans, logos, and studying, are not wrong uses of AI. They are simply the shallow end of a pool that goes much deeper. A business plan drafted by AI still needs someone to build the business. A logo designed by AI still needs a brand behind it. Studying with AI still requires the student to sit the exam. The 80% StarApple Analytics says Jamaica is leaving unused sits past that first draft, in the systems that keep running after the person has moved on to something else.
What Comes Next
StarApple Analytics has not said this will be an annual study, but the pattern of the past four months suggests the region should expect more of them. Two data sets in three months, one on whether Jamaicans use AI and one on what they do once they start, is already more granular than anything a foreign consultancy has published about the island. If that cadence holds, the next report will likely test whether the 20% figure moves at all, and whether the businesses reading this one act on it before the gap widens further against markets that are already redesigning workflows rather than drafting documents.