What is schema markup, and why does it matter for AI?
Schema markup for AI is structured data, written in JSON-LD, that states what a page and its business are — organisation, author, FAQs, products — in a machine-readable form engines can extract without guessing.
Schema.org is a shared vocabulary that every major search and AI engine understands. Written as JSON-LD in the page head, it turns prose into discrete facts: this is an Organization, its name is X, its founder is Y, this article was published on Z date. For classic SEO the payoff was rich results. For AI engines, the payoff is cleaner extraction and stronger entity resolution — schema is one of the ways you tell an engine what your entity actually is, which in turn feeds directly into whether you’re recognised as a citable brand at all, a theme covered in our wider AI SEO guide.
The distinction worth holding onto: schema doesn’t make your content better, and it doesn’t make an AI engine like you more. What it does is remove ambiguity. A model reading unmarked prose has to infer who wrote it, when, and what the business behind it actually is. Schema states these facts directly, so the model spends less effort inferring and more confidence repeating them.
Which schema types do AI engines actually use?
In our own share-of-voice work, the schema types that consistently earn their keep are Organization, Person, Article and FAQPage — because they reinforce entity recognition or hand AI engines clean, extractable answers.
| Schema type | What it declares | Observed value for AI visibility | Effort |
|---|---|---|---|
| Organization / Person | Your entity — name, founder, location | High | Low |
| Article / BlogPosting | Author and publish/update dates | Medium–high | Low |
| FAQPage | Clean question-and-answer pairs | High | Low–medium |
| Product / Offer | Price and availability | High for retail | Medium |
| Breadcrumb | Site structure | Low–medium | Low |
| HowTo / Review | Step-by-step or review content | Situational | Medium |
These rankings reflect what we observe in client and internal share-of-voice measurement; treat any harder percentage claim with the same caution.
A practical way to read this table: Organization and Person schema are near-mandatory because they underpin entity recognition site-wide, not just on one page. Article schema is cheap to add and low-risk. FAQPage schema pays off in direct proportion to how genuinely useful your Q&A content is — badly written FAQ pairs marked up in schema are still badly written FAQ pairs. Product and Offer schema matters disproportionately for retail and e-commerce, where AI shopping assistants increasingly parse price and availability directly. Breadcrumb and HowTo schema are worth adding when cheap to implement, but rarely worth a dedicated project.
Which schema is overrated for AI search?
Schema that merely duplicates what the page already states in plain text, or that chases rich-result formats Google has since deprecated, adds little for AI recommendation and is not worth the engineering time.
Two honest warnings. First, don’t over-mark-up thin pages — wrapping a paragraph of filler in Organization and Article schema doesn’t make it more citable, because AI engines still need real content to extract. A schema tag cannot manufacture substance that isn’t there. Second, never let schema state something the visible page doesn’t: a mismatch between marked-up facts and displayed content is a trust risk, not a shortcut, and search engines increasingly cross-check the two. Keep schema as a faithful mirror of the page, not a wish list.
A third, quieter trap: chasing every schema type Schema.org publishes, on the theory that more markup must help. It doesn’t. Effort is finite, and marking up ten low-value types thoroughly is worse use of time than marking up the four or five high-value ones correctly and keeping them accurate as the page changes.
How do you add schema correctly?
Add schema by writing valid JSON-LD in the page head, starting with Organization and Person site-wide, then Article and FAQPage where the content genuinely supports them.
- Use JSON-LD in the page
<head>— not microdata scattered through the HTML body. - Start with Organization and Person schema site-wide; add Article schema to posts and FAQPage schema wherever you have real Q&A.
- Keep every value identical to the visible content and to your entity description elsewhere on the web.
- Validate with a structured-data testing tool before publishing.
- Keep
datePublishedanddateModifiedaccurate — AI Overviews favour fresh, well-maintained content. - Re-check periodically that AI engines describe and cite you correctly after changes.
Schema fixes of this kind are part of NeuralGen’s fixed AI Visibility Audit — £1,800, delivered in five working days. The audit checks existing markup for validation errors, mismatches with visible content, and gaps against the high-value types above, then implements the fixes rather than just listing them.
For most WordPress sites this work is genuinely quick — a plugin or a small custom function can output Organization schema site-wide in an afternoon. The larger time cost is usually validation and keeping FAQPage and Article schema in sync as content changes, which is where an ongoing process, not a one-off task, pays off.
Does FAQPage schema still help AI Overviews?
Yes, for AI extraction, with a caveat: clean question-and-answer pairs are among the easiest content for AI engines to lift, so FAQPage schema remains useful for AEO even though Google has narrowed FAQ rich results in classic search.
The distinction matters. A rich result is a visual enhancement in classic Google search — an expandable FAQ snippet under your listing. An AI citation is different: it’s an AI engine reading your Q&A pairs and reusing them, in whole or in part, inside a generated answer. Google scaling back FAQ rich results says nothing about whether FAQPage schema still helps AI Overviews or ChatGPT parse your content — it does. For the full picture of where this fits alongside everything else, see our AI search optimisation checklist.
Practically, this means the incentive to write good FAQ content hasn’t gone away — it’s just moved. Where a rich result once justified an FAQ section on its own, the reason to invest in one now is that AI engines read it directly. Keep the questions genuinely representative of what people ask, and keep every answer self-contained, because that is precisely the shape an AI engine can lift cleanly.
What about llms.txt — is it schema?
No. llms.txt is a plain-text file at your domain root that points AI crawlers to your most important, canonical content; it complements schema rather than replacing it.
Where schema describes facts within a page, llms.txt is a simple index — a signpost telling AI crawlers which pages matter most and where to find the canonical version of your content. It’s an emerging convention, not a Schema.org standard, and adoption by AI engines is still uneven. Given the low cost of adding it, it’s reasonable insurance rather than a guaranteed ranking factor. Implementing schema and llms.txt properly across a site is one of the practical steps in our AI visibility service.
A typical llms.txt file lists your most important pages in plain Markdown — your homepage, your core service pages, your pillar guides — with a short description of each, so a crawler with limited time or budget can prioritise the pages you actually want represented. It costs almost nothing to write and maintain, which is the main argument for adding it even while support across engines remains inconsistent: the downside risk is close to zero, and the upside is a small, cheap nudge toward the content that matters most.