Beyond the Model: AI is only as trustworthy as its sources, specially in Emerging Markets

Everyone wants to talk about the model: which large language model is fastest, which one reasons best, and which one “hallucinates” least. It is an easy conversation to have, because models are visible, you can test them, compare them and benchmark them in a spreadsheet.

What is much harder to see, and often more important, is what the model is working with in the first place.

My work focuses on the content layer that sits behind AI: where information comes from, whether it is authoritative and current, how it is assessed, and whether it can support a real business decision. In emerging markets, that expertise matters because the most valuable signal is often local, difficult to discover and easy to misread from outside.

The gap nobody talks about

Ask a general-purpose AI tool a detailed question about a mature, well-documented market and it may do a reasonable job. There is decades’ worth of widely available, indexed, structured and predominantly English-language content for it to draw on.

Ask the same tool a specific question about a privately held, mid-sized company in Colombia, or the ownership structure of a Southeast Asian conglomerate, and the limitations become much more visible. The open web rarely offers the same depth or consistency across these markets. Valuable information may be published only in local languages, use terminology that is highly specific to a market, or sit in specialist sources and archives that are not publicly available or readily discoverable. What is public can also be outdated, incomplete, poorly attributed or copied from a source with no clear accountability.

This is the part of the AI story that is often skipped: a model can only work with the content it can access and interpret in context, and not all content is equally credible. Language matters: relevant signals can be missed not because they do not exist, but because they sit outside the globally visible, predominantly English-language information ecosystem. Fluency cannot turn weak evidence into reliable insight; a polished answer can still be a confident-sounding guess.

Why curation is the real differentiator

At EMIS, our teams have spent years building relationships with local information providers, regulators, exchanges and specialist publishers across emerging and frontier markets. Many of these sources, or their archives, are not available on the open web. Our content experts know where to look, which sources carry authority locally, how to assess a newer or less familiar publication, and how to evaluate information shaped by local terminology and publishing practices. A source does not have to be globally famous to be credible; sometimes its local specificity is exactly what makes it valuable. Curation is not collecting more information. It is knowing what is credible enough to use.

That is the unglamorous, human work behind every AskISI answer, and it does not stop at finding and evaluating sources. We identify gaps we know exist in each market, connect fragmented signals, produce content where reliable local coverage is missing, and create metadata that gives structure to the information held in documents. That metadata turns a collection of reports, filings and news into a roadmap: it makes information discoverable, connects companies, industries and events, and helps users and AI systems understand how the pieces fit together. AskISI is grounded in that curated and structured base, built and maintained across markets and languages.

The distinction matters because business decisions demand a higher standard than “it sounds plausible.” No business leader would knowingly base an investment, market-entry or risk decision on an anonymous post or a stale webpage; using AI should not lower that standard. In markets where public coverage is fragmented, credible sourcing and traceability are essential. They are also a source of differentiation: access to distinctive local content, combined with the expertise to interpret it, can produce insights that an open-web-only tool cannot simply replicate.

Trust has to be built before the AI even starts writing

The industry conversation about AI trust tends to focus on the moment of output: is the answer accurate, is it hallucinating, and can it be verified? Those are the right questions, but they come too late if the underlying content was never trustworthy to begin with.

Our view is simple: trust cannot be bolted onto an AI system after the fact. It has to be built into the content layer first – through authoritative sourcing, traceability, quality control and human judgement – long before a model generates a sentence. That means identifying what deserves to be trusted and understanding what it means in context. It is why AskISI can answer questions about markets where good information is hardest to find, grounded in sources users can identify and assess for themselves.

The smartest model in the world is still only as good as the content it is allowed to read. In emerging markets, that is not a footnote – it is the foundation of a credible answer.

By Cristina Bustamante, Director of Content Licensing & Partnerships, ISI Markets

Cristina Bustamante is Director of Content Licensing & Partnerships at ISI Markets, where she leads the strategy to expand access to authoritative local content and high-quality business information across emerging markets. Based in London, she has more than a decade of experience in content, underpinned by a background in economic journalism at leading media outlets.

---

Learn more about EMIS, leading provider of reliable industry and company intelligence across emerging markets, and explore its recently launched AI assistant ASkISI.