Structured data is information in a fixed, machine-readable format, and unstructured data is everything else, such as ordinary page text. Add schema markup to a page and the common assumption is that AI systems now understand it perfectly. That is not quite right, and it is not quite wrong. Structured data and well-written prose do different jobs, and the evidence on whether structured data helps you get cited by AI is thinner than most articles suggest.
This post explains what structured data is good at, what only prose can do, what Google says about schema and AI features, what a large 2026 study found when pages added schema, and what to do about it. We re-read Google’s guide to optimizing for generative AI features (last updated 10 July 2026) and its page on AI features on 11 October 2026, and we audited the structured data on our own site. Where something is our own judgment, we say so. For writing the prose side well, see our post on passage ranking and standalone paragraphs.
Quick summary, if you’re short on time:
- Google’s guide to generative AI features says structured data isn’t required for generative AI search, and that there’s no special schema.org markup you need to add. It adds that it is a good idea to keep using it as part of your overall SEO, because it helps with eligibility for rich results.
- Google also says you can ignore overfocusing on structured data, along with llms.txt files, chunking content, rewriting content just for AI systems and seeking inauthentic mentions.
- An Ahrefs study published in May 2026 found that pages cited by AI were almost three times more likely to have JSON-LD, but that 1,885 pages which added JSON-LD showed no major uplift in citations on Google AI Overviews, AI Mode or ChatGPT compared with matched control pages.
- Use structured data for exact facts, and unstructured data (your prose), such as name, address, hours, price and dates, for reasoning and explanation. Keep them consistent.
- On our own site, all 132 pages in our sitemap carry structured data. We cannot show that it changes AI citations, and we do not claim it does.
Jump to a section:
- The plain definitions
- What each one is good at
- What Google says
- What the evidence shows
- Using both properly
- What we found on our own site
- A prompt to check your content mix
- Questions about structured data and AI
Structured vs unstructured data: the plain definitions
Before the website angle, here is the general difference, because the same words mean slightly different things in different fields.
- Structured data is organised in a fixed format that a computer can read without interpretation, usually as rows and columns, such as a table of customers in a database or a spreadsheet. Every field has a known type: a date, a number, a name.
- Unstructured data has no fixed format. Emails, documents, web page text, images, audio and video are all unstructured. It holds most of the information businesses have, but a program has to interpret it, for example with language understanding or machine learning, before it can be used like a table.
- Semi-structured data sits in between. It has labels or tags but not a rigid table. JSON and XML files are the usual example, and so is schema markup on a web page.
| Structured | Semi-structured | Unstructured | |
|---|---|---|---|
| Example | A customer table with name, phone and city columns | A JSON file or schema markup on a page | An email, a blog post or a PDF |
| Where it lives | A database or spreadsheet | Files, APIs, page code | Documents, web pages, media files or a data lake |
| How it is analysed | Direct queries | Parsing the labels | Language processing or machine learning algorithms |
| Strength | Exact, fast and reliable | Flexible with some structure | Rich in meaning and detail |
These are the general definitions, and they are common knowledge in data work, not claims from Google. Now to what this means for a website, where your visible page text is unstructured data and your schema markup is structured, or semi-structured, data about it.
What structured and unstructured data are each good at
| Structured data (schema markup) | Unstructured data (your prose) | |
|---|---|---|
| Best at | Exact facts with no interpretation: a name, address, hours, price, rating, date | Reasoning, comparison, process, nuance and edge cases |
| Who reads it | Machines, directly | People first, and machines through language understanding |
| Weakness | Cannot explain why you are different or how your pricing works | Facts buried in sentences can be harder to extract reliably |
| Risk | Out-of-date or wrong markup, which contradicts the page | Vague writing that never states the fact plainly |
The difference between structured and unstructured data is the key to this topic, and the two do not compete. A good page states its key facts plainly in the text and, where it helps, repeats them in markup, so that a person and a machine see the same thing.
What Google says
Google’s guide to optimizing for generative AI features is direct on this. Under the things it says you can ignore for Google Search, it lists “overfocusing on structured data”. It says structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add. But it says it is a good idea to continue using it as part of your overall SEO strategy, as it helps with being eligible for rich results on Google Search.
Google’s page on AI features says something similar: you don’t need to create new machine-readable files, AI text files or markup to appear in these features, and there is no special schema.org structured data you need to add. It adds that the best practices for SEO remain relevant for AI features, and that to be eligible as a supporting link a page must be indexed and eligible to be shown in Search with a snippet.
So Google’s position is: use accurate structured data because it supports rich results, and do not expect it to be a separate lever for AI answers. Note that this is Google’s own AI features. It does not describe how other AI assistants work.
What the evidence shows
Many articles claim schema increases AI citations. The best public test we know of, which is the one to quote, is a study by Ahrefs published on 11 May 2026. Here is what it reports:
- The correlation. From an analysis of 6 million URLs, AI-cited pages were almost three times more likely to have JSON-LD than non-cited pages, and 53% of AI-cited pages had schema.
- The caveat. Ahrefs says that could easily be correlation and not cause, because schema tends to live on better-maintained, more technically sophisticated sites that also publish stronger content and earn more links.
- The test. It tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched them against 4,000 control pages, and measured citation changes.
- The result. Adding schema produced no major uplift in citations on any platform. It reported a 4.6% decline for Google AI Overviews, which it called small but statistically significant and said might reflect other factors, and changes of +2.4% for AI Mode and +2.2% for ChatGPT that were statistically indistinguishable from zero.
Ahrefs’ own conclusion is that it cannot tell whether schema did a tiny bit of good or nothing at all. Read that as: the claim that schema gets you cited is not proven. It does not show that schema is useless. It still helps with rich results, and it keeps your facts consistent. We are describing one study, and a different method could find something else.
Using structured and unstructured data together
The best pages combine structured and unstructured data so that the same facts appear in the text and in the markup. Unstructured data carries the explanation, and structured data gives the fixed facts a machine can read exactly.
- State the key facts in plain text first, as unstructured data that people can read. Your name, address, phone number, hours, main services and prices belong in visible text, where people and machines can both read them.
- Mark up only what is on the page. The structured data should match the visible content. If the markup says one price and the page says another, you have created a contradiction.
- Choose the specific type. Use the schema.org type that really describes the thing, such as LocalBusiness or a subtype, Article or FAQPage, and keep it current.
- Do not hide explanation in fields. Reasons and comparisons belong in readable paragraphs under clear headings.
- Test it. Use the Rich Results Test and the Schema Markup Validator, and fix errors.
- Review it when things change. Prices, hours and addresses change, and the markup must change with them.
Our post on LocalBusiness schema goes through the local business markup in detail, and our post on schema markup for AI visibility covers which types are worth adding.
What we found on our own site
We looked at the structured data on every page in our sitemap on 11 October 2026, using the local copy of zunelo.in:
| What we measured | Result |
|---|---|
| Pages checked | 132 |
| Pages with JSON-LD structured data | 132 |
| Organization, WebSite and Place on every page | 132 |
| BlogPosting and Person (author) on blog posts | 91 of 91 |
| FAQPage on pages with an FAQ section | 127 |
| Service markup on service pages | 27 |
So we are as thorough with markup as most sites could reasonably be. We still cannot tell you it changes how AI tools cite us, and we would not claim it. We keep it because it is accurate, it supports rich result eligibility where Google offers it, and it keeps our facts consistent. Note that Google removed the FAQ rich result from Search, so our FAQ markup is there for consistency and for other systems, and we do not rely on it for a visible feature. This is a judgment about our own practice, and your situation may differ.
A prompt to check your content mix
An AI tool can compare your page text with your markup and flag differences. It cannot fetch your live page, so paste both and check the findings.
Act as a careful content and structured data reviewer. Compare the text of my page with its JSON-LD markup and list differences. Use only what I paste.
My details:
– What the page is, and who it is for: [one line]
– The visible text of the page: [paste]
– The JSON-LD from the page source: [paste]Please do the following:
1. List every fact that appears in the markup but not in the visible text, such as a price, a phone number or opening hours.
2. List every fact that appears in both but with a different value.
3. List key facts that appear only in the text and could sensibly be added to the markup, such as address or hours.
4. List any place where the text is vague about something the markup states precisely.
5. Say what you could not check.Rules:
– Use only my pasted text and markup. Do not invent facts or schema properties.
– Do not promise that markup will increase AI citations.
– Write in plain, simple English.
Fix mismatches in favour of what is true, then test the markup again before you publish.
Questions about structured data and AI
Does structured data help AI search?
Google says structured data isn’t required for generative AI search and that there’s no special schema.org markup to add, though it helps with rich results. A 2026 Ahrefs study found no major uplift in AI citations when pages added JSON-LD.
Do pages with schema get cited more by AI?
In an Ahrefs analysis of 6 million URLs, AI-cited pages were almost three times more likely to have JSON-LD. Its own follow-up test of pages that added schema found no major uplift, which suggests the first result reflects better-maintained sites, not schema.
Should I remove my schema markup?
No. Google says to keep using structured data as part of your SEO because it helps with rich result eligibility. Keep it accurate and in line with the page.
What should go in structured data and what in prose?
Exact facts, such as name, address, hours, price and dates, belong in both. Reasoning, comparison, process and nuance belong in clear prose under good headings.
Is wrong schema worse than no schema?
Markup that contradicts the visible page creates confusion for people and machines. Our advice is to keep it accurate and in line with the page, or remove it.
How do I test my structured data?
Use Google’s Rich Results Test for features Google supports, and the Schema Markup Validator for the general vocabulary. Fix errors, then check the live page.
Is there special schema for AI Overviews?
No. Google says there is no special schema.org structured data you need to add to appear in AI Overviews or AI Mode, and that the best practices for SEO remain relevant.
Where this matters
Use structured data for what it does well: exact, accurate facts and rich result eligibility. Use prose for explanation, and keep the two consistent. Do not expect markup alone to win AI citations, because the best evidence we have says it does not.
If you would like your markup and content reviewed, see our AI SEO and GEO service, or send us your website address and we will tell you what we see.
