Schema markup won't get you cited by AI. Here is what it actually does.
Ahrefs tracked 1,885 pages that added schema markup and watched what happened to their AI citations. The answer: almost nothing. On Google AI Mode the lift was 2.4 percent. On ChatGPT, 2.2 percent. On AI Overviews it went slightly negative. Numbers that size are indistinguishable from noise.
That should bother anyone who has been told to ship FAQ schema to win in AI search. So let us settle what structured data really does for generative engines, what it does not, and where it is still worth the effort.
The correlation that fooled everyone
There is a real, repeatable finding behind the schema hype. Across 6 million URLs, pages cited by AI were about three times more likely to carry JSON-LD than pages that were not. Vendors quote that 3x everywhere. It is true. It is also confounded.
Schema does not live on random pages. It lives on better-maintained, more technical sites. The same sites publish deeper content, earn more links, and build more brand authority. Those are the things that win citations. Strip the schema out and the rest of the signals still carry the page through. Correlation, not cause. Ahrefs ran the actual causal test, matched control pages and a difference-in-differences model, and the lift collapsed.
What happens to your JSON-LD before the model ever sees it
Here is the part most schema advice skips. When an AI engine fetches your page at answer time, it does not hand the raw HTML to the model. It converts the page to clean text or Markdown first, and that conversion throws away `<script>` tags. Your JSON-LD lives in a `<script type="application/ld+json">` block. It gets stripped on the way in.
Two controlled tests show this plainly. A German agency, searchVIU, built identical pages that differed only in their markup and asked five systems to pull a fact placed only inside the schema. When a price existed only in JSON-LD, zero of five systems retrieved it. Otterly ran a similar direct-fetch test across seven platforms: six could not read the schema when asked, and Google AI Mode hallucinated a schema type that was not even on the page.
The takeaway is uncomfortable but clean. For most AI assistants, your schema is not in the room. They are reading the words a human reads.
The two engines that actually read it
Not every platform throws the markup away. Two are on record or clearly observed reading structured data, and both matter.
Microsoft is the only major player to say it out loud. At SMX Munich in 2025, Bing principal product manager Fabrice Canel confirmed schema markup helps Copilot's models understand content. Bing has parsed Schema.org since 2018. Gemini is the second. In the direct-fetch tests, it was the lone system that read raw JSON-LD correctly, which fits Google's wider stance on structured data.
Google itself is blunt about the limit. Its own documentation states there is no special schema for AI features and that structured data is not required to appear in AI Overviews or AI Mode. Read that twice. The company building the most-used AI search surface says you do not need schema to show up in it.
| Engine | Reads your JSON-LD? | How it really gets your content |
|---|---|---|
| Google AI Overviews / AI Mode | Yes, via Gemini and the index | Retrieves from the Search index. Schema helps eligibility and rich results, not direct selection. |
| Bing Copilot | Yes, confirmed | Built on Bing's index, which parses schema. The one platform that says it uses markup. |
| ChatGPT search | No, reads rendered text | Bing-grounded retrieval, then reads visible page text. Any schema benefit is indirect, through Bing. |
| Perplexity | No, reads rendered text | Own crawler and index. Pulls the visible text and leans on community sources. |
| Claude | No, reads rendered text | Web search returns page text. No documented use of structured data. |
So why ship schema at all
Because the indirect path is real, and one direct path still holds.
The indirect path is plain SEO. Schema earns rich results and helps Google and Bing understand your pages. Those two indexes are exactly what AI Overviews and ChatGPT search retrieve from. AI Overviews are grounded in Google's ranking systems, and roughly 97 percent of them cite a source from the top 20 organic results. If schema helps you rank and get indexed, it buys you a ticket into the pool the model draws from. The benefit is two steps removed, but it is not imaginary.
The direct path is commerce. Price, availability and shipping are genuinely hard to read out of prose, and this is the one place even Google's John Mueller says structured data does heavy lifting. Product schema and merchant feeds are how that data reaches AI shopping answers. If you sell things, this is not optional.
Where schema does nothing is as a standalone trick to earn citations. Do not sell it that way. Do not buy it that way.
The FAQ and HowTo era is over
If your AI-search plan leans on FAQ and HowTo schema, update it. Google retired the HowTo rich result in 2023 and finished removing the FAQ rich result by 2026. The types still validate and they do no harm as clean question-and-answer structure, but they no longer earn a visual feature, and they were never a citation lever to begin with. The win that people credited to FAQ schema was always the answer-first writing underneath it.
What to actually do
Keep schema. Treat it as cheap infrastructure, not a growth channel. The priority order that holds up:
Organization and Person to anchor your brand and authors as entities, with `sameAs` to verified profiles. This is your entity foundation.
Article on editorial pages, with an accurate `dateModified`.
Product on commercial pages, kept in sync with live price and stock.
Validate every block, and make sure the markup matches what a reader can see on the page. Mismatched schema gets you demoted, not promoted. Everything past that, the FAQPage and HowTo blocks, is optional housekeeping.
Then put your real budget where the citations actually come from: depth, named authorship, brand mentions, and earned links. We have written the playbook for that in how to get your brand cited in ChatGPT, Gemini and Perplexity, and the case for why server-rendered HTML beats your clever markup for AI crawlers. If you want the wider argument on how these disciplines fit, start with GEO vs SEO.
Schema markup is a good habit. It is just not the lever you were told it was.
---
Want to know how AI engines actually see your site? Run a free AI Visibility Audit to see where you show up across ChatGPT, Gemini, Perplexity and AI Overviews, and what is really driving it. Or talk to us about building the program.
