The difference between structured and unstructured data, the third category most explanations leave out, and why the phrase means something narrower in search

Structured data is information organised into a predefined model, so that every item has named fields and a known type. Unstructured data is information with no such model, where the meaning sits in the content rather than in a schema: the text of an email, a photograph, an audio recording, a PDF. The single practical difference is that structured data can be queried directly and unstructured data has to be processed before it can be queried at all. Most explanations stop there and miss the category that holds a large share of real data, which is semi structured: information carrying tags or keys that describe it, without a fixed table behind it. JSON, XML and most API responses are semi structured, and so is the kind of structured data a marketer adds to a web page.

The three categories side by side

	Structured	Semi structured	Unstructured
Where the shape is defined	In a schema, before the data arrives	In the data itself, as tags or keys	Nowhere
Typical form	Database tables, spreadsheets, CSV	JSON, XML, log lines, API responses	Documents, email bodies, images, audio, video
How you query it	Directly, with a query language	By path or key, with tooling that understands the format	Only after extraction, indexing or a model reads it
What happens when the shape changes	The schema has to change, deliberately	New keys can appear without breaking the rest	Nothing, because there is no shape to break
Share of a typical organisation's data	The minority, and the part everybody reports on	Growing, and mostly in motion between systems	The majority, and the part nobody reports on

The same fact, written three ways

Take one piece of information: a customer named Dana Price bought a product on the twelfth of March for 240 dollars and was unhappy with the delivery.

• Structured. A row in an orders table: customer_id, order_date 2026-03-12, amount 240.00, currency USD, status delivered. Every field is named and typed. You can total the month's revenue in one query and you learn nothing at all about the complaint.
• Semi structured. A JSON object with the same fields plus a nested array of support interactions, each with a channel and a timestamp. New fields can be added by whoever writes the next version without anybody redesigning a table, which is why systems talk to each other in this format.
• Unstructured. The email Dana actually sent. It contains the complaint, the tone, the reason she may not order again, and the specific thing that went wrong, and none of that is in either of the first two forms. To count how many customers complained about delivery this month you have to read every email or have something read them for you.

That example is the whole argument in miniature. The structured version is what gets reported because it is easy to count. The unstructured version is where the reason lives. Almost every organisation has good numbers about what happened and poor numbers about why, and this is the mechanical cause.

What each category is genuinely good for

• Structured data is for arithmetic and for anything that has to be consistent. Finance, inventory, transactions, anything audited. The rigidity is the feature: a typed field cannot quietly become something else.
• Semi structured data is for exchange and for things that evolve. It is the lingua franca between systems precisely because it tolerates change on one side without breaking the other.
• Unstructured data is for meaning, and it has become usable in a way it was not five years ago. Language models and vector search have made it possible to query a body of documents by meaning rather than by keyword, which turns an archive from storage into an asset.

Why a marketer meets the same words in a narrower sense

This is the point where the general definition and the search definition diverge, and it causes real confusion for anyone who learned the words in one context and then meets them in the other.

In search, "structured data" means something specific: markup added to a web page, in a standard vocabulary, telling a machine what the things on the page are. A price is a price, a review is a review, an opening time is an opening time. It is written in JSON-LD, which is semi structured in the general taxonomy, and it is added to a page whose main content is unstructured prose. So a page with structured data is an unstructured document carrying a semi structured description of itself, and both senses of the phrase are correct within their own field.

The practical significance for a marketer is that the markup is the only part of a page whose meaning is not being inferred. Everything else about the page has to be interpreted; the markup is asserted. That is why it has an outsized effect on how a page is displayed and quoted relative to the effort it takes, and why getting it wrong is worse than omitting it. How to add it on the most common platform, with or without a plugin, is in what structured data does and how to add it, and a copyable block for one of the most commonly needed types is in a copyable SoftwareApplication block and the properties that matter.

Turning unstructured data into something you can count

The question that actually arrives in a business is not which category a thing belongs to. It is how to get a number out of a pile of text. Four approaches, in increasing order of effort.

1. Add structure at the point of entry. The cheapest fix by a wide margin. A dropdown on a support form produces a countable field; a free text box produces an archive. Decide this before the data exists, because retrofitting is the expensive version of the same work.
2. Tag a sample by hand. Two hundred records read and labelled by a person gives you a distribution you can act on and a training set for anything automated later. This is underrated and it is usually an afternoon.
3. Extract fields automatically. Pull named entities, dates, amounts and categories out of documents into a table. Reliable for well formed documents such as invoices, less so for free prose.
4. Index for meaning rather than words. Embed the documents and query them by similarity, so a search for delivery complaints returns the email that said the parcel arrived broken without the word complaint in it. This is what makes a large unstructured archive answerable.

Questions about structured and unstructured data

What is the main difference between structured and unstructured data?

Structured data has a predefined model with named, typed fields, so it can be queried directly. Unstructured data has no model, so it has to be processed before any query can run against it. The shape is the difference, and everything else follows from it.

Is JSON structured or unstructured?

Neither, strictly. JSON is semi structured: it carries keys that describe its values without a fixed schema sitting behind it. That is why systems exchange data in it, and why it is the format used for the structured data markup added to web pages.

Is a PDF structured data?

Generally no. A PDF is a layout format and its text is unstructured, which is why extracting a table from one is harder than it looks. A PDF generated from a database may contain structured information presented unstructurally, which is the worst of both and extremely common in business reporting.

Is email structured or unstructured?

Both at once, usefully. The headers are structured, with named fields for sender, recipient, subject and timestamp. The body is unstructured. That is why you can count emails by sender in seconds and counting them by topic is a project.

Why does unstructured data matter more now?

Because it became queryable. Searching a large body of documents by meaning rather than by exact keyword is a recent capability, and it changes an archive from something you store into something you can ask questions of.

Does structured data in SEO mean the same thing?

It means something narrower. In search it refers to markup added to a page in a standard vocabulary, declaring what the things on the page are. Both uses are correct in their own field, and the search one describes a specific technique rather than a category of data.

The one part of a page whose meaning is declared rather than guessed

Everything above is a distinction about shape, and it has a direct consequence for anybody publishing on the web. The systems that now answer questions for your customers read pages as unstructured text and infer what they mean, which they do well and not perfectly. The markup on a page is the only part they do not have to infer, because it is stated: this figure is a price, in this currency, valid until this date; this person holds this role; this business is open at these hours. When an assistant repeats a fact about your company and gets it wrong, the fix is almost never more prose, it is asserting the fact in a form that leaves nothing to infer. AI Knows Us reads what the assistants currently say about a business, and shows which pages and which sources each answer was assembled from, which is how you find out whether your declared facts are the ones being used.