What is data literacy?

Data literacy is the ability to read, question and use data: to understand where it came from, what it does and does not show, how reliable it is, and how to act on it sensibly. It is not statistics. It is knowing when a number, a record or a document can bear the weight someone wants to put on it.

Why does data literacy matter for AI agents?

Agents act on facts. A payment, an account change or a reply to a customer is only as sound as the data behind it. When an agent holds an action for a person, the reviewer is mostly judging evidence: is this the right customer, the right amount, the right account, from a source that can be trusted? A reviewer who cannot weigh evidence will approve whatever looks complete.

Data literacy is also how people catch problems agents cannot. A ledger entry that does not match a statement, a date that falls on a weekend when the business is closed, a figure that looks plausible but sits outside the normal range: these are often spotted by a literate person before any rule catches them.

How does it show up in practice?

In a governed platform, every fact reaching an agent arrives stamped with where it came from. The provenance classes, strongest first, are:

• sealed: issued by the system of record itself, such as the core banking ledger, and tamper-evident;
• signed: a named, identifiable party answers for it;
• declared: stated by a customer or person, not verified;
• external: arrived from outside, such as an emailed invoice or a web page.

A data-literate reviewer reads these classes as naturally as a date. They know an external claim cannot authorise a payment, and that declared income needs checking before a credit decision. They ask "where did this come from?" before "does this look right?"

What is an example?

A hospital finance team's agent drafts a supplier payment and holds it because the bank details changed. The new details came from an email, external provenance. The supplier's master record, sealed, still shows the old account. A reviewer with data literacy spots the mismatch, calls the supplier on the number already on file, and rejects the change. Health and banking settings face this daily, which is why sector guides such as Article 4 literacy for agents in healthcare and in banking treat evidence as a core skill.

How is data literacy different from AI literacy?

AI literacy is understanding how AI systems work, what they can and cannot do, and how they are governed. Data literacy is understanding the material they work on. A person can know a great deal about agents and still accept a forged invoice; another can be sharp with data and misread why an agent refused. People who oversee agents need both, and a sound AI literacy programme teaches them together.

Which terms are related?

• Prompt literacy: skill in instructing AI systems, explained in what prompt literacy is.
• Provenance: the record of where a fact came from.
• Decision literacy: reading how evidence became an outcome.

For the full topic, read AI literacy and certification for people who work with agents.