Structured data and Google AI Mode: does schema markup actually help you get cited?

Key takeaways

• SALT.agency analyzed 107,352 URLs cited in Google AI Mode and found no single schema type beyond standard baseline markup (Organization, WebPage, Article) gave a measurable citation advantage.
• Otterly.AI ran a 3-month controlled experiment across 7 AI search platforms and found 6 out of 7 couldn't even correctly fetch schema markup when directly asked.
• Schema still matters for traditional Google rich results and AI Overviews within Google's own ecosystem, just not as a direct citation signal in AI Mode.
• The reason is architectural: LLMs tokenize raw HTML and don't read JSON-LD the way a traditional crawler does.
• The better path to AI citations is content quality, topical authority, and entity clarity -- not adding more schema types.

---

There's a persistent belief in the SEO world that schema markup is a kind of secret handshake with AI search engines. Add the right structured data, and Google's AI Mode will recognize your page as authoritative and cite it. It's a tidy story. It's also probably wrong.

Two separate research projects have now looked at this question with real data, and the results are worth sitting with -- especially if you've been spending time implementing FAQPage, HowTo, or Product schema specifically to boost AI visibility.

What SALT.agency found across 107,352 cited URLs

SALT.agency pulled schema data from every URL cited in Google AI Mode across a sample of 107,352 websites. The goal was simple: do pages that get cited in AI Mode have different schema profiles than pages that don't?

Here's what the schema distribution looked like across those cited pages:

The top schemas are exactly what you'd find on any reasonably well-built content site. Organization, WebPage, Article, BreadcrumbList -- these aren't sophisticated GEO tactics. They're the baseline of technical SEO hygiene that most sites already have.

The most common schema "triplets" (connected relationships between schema types) were equally mundane:

• WebPage → mainEntity → Article
• Article → author → Person
• Organization → url → Homepage

Nothing exotic. No clever schema combinations that seemed to unlock citation status. SALT.agency's conclusion: "no single type beyond the standard set appears to give a measurable advantage."

What Otterly.AI found in a controlled experiment

Otterly.AI took a different approach. Rather than analyzing existing citations, they ran a prospective experiment: implement five schema types on live pages, then track what happens across 319 prompts on 7 AI search platforms over three months.

The results were striking in a different way. Six out of seven AI platforms couldn't correctly fetch or interpret the schema markup when directly asked about it. Only Gemini retrieved the correct JSON-LD data. Google AI Mode actually hallucinated a "Service" schema type that wasn't implemented on the page at all.

The experiment did find meaningful gains within Google's own ecosystem: SERP features increased by 377% and AI Overviews appearances went up 1,500% after three months. But for AI search platforms outside Google (ChatGPT, Perplexity, Claude, etc.), the direct impact was "much more limited."

Otterly.AI's conclusion: "Schema markup did not appear to directly influence AI Search citation behavior in a measurable way."

Why this makes architectural sense

The reason schema doesn't work the same way for AI search isn't a bug or an oversight. It's a fundamental difference in how these systems process web pages.

Traditional search crawlers were built to read structured data. Schema markup was invented precisely because early algorithms couldn't reliably extract meaning from unstructured text. You needed to explicitly tell Google "this is an organization" or "this is a review" because it couldn't figure that out on its own.

Large language models don't have that problem. They tokenize raw HTML -- breaking the entire page into smaller units the model processes together. A JSON-LD block like this:

{
  "@type": "Organization",
  "name": "Example Corp",
  "url": "https://example.com"
}

Gets tokenized the same way as any other text on the page. The model isn't reading it as structured metadata with special authority. It's reading it as text, alongside everything else. If your page clearly says "Example Corp is a company that does X" in plain prose, that signal is just as legible to an LLM as the schema version.

This is the core architectural point that makes schema a weaker lever for AI citation than it is for traditional SEO.

The Ahrefs data point

Ahrefs tracked 1,885 pages that added JSON-LD schema between August and a later observation period and found no correlation between schema implementation and increased AI citations. Their framing was direct: "The data does not support the claim that you need schema to get cited more by AI."

That's three independent data sources -- SALT.agency, Otterly.AI, and Ahrefs -- pointing in the same direction.

Where schema still matters (and it does)

None of this means you should strip schema from your site. That would be a mistake for a few reasons.

Within Google's own ecosystem, schema clearly still works. The Otterly.AI experiment showed a 1,500% increase in AI Overviews appearances after implementation. Google AI Overviews and traditional rich results are still driven by structured data signals, and those features drive real traffic.

Schema also helps with entity recognition. Google's Knowledge Graph relies heavily on structured data to understand who you are, what you do, and how your organization relates to other entities. That entity clarity has downstream effects on how confidently Google's systems treat your content -- even if it's not a direct citation trigger in AI Mode.

The practical guidance from the SALT.agency data is worth taking seriously: Organization, WebPage/Article, BreadcrumbList, and Author (Person) schemas are present in the vast majority of AI Mode citations. Make sure you have those. They're not citation drivers, but their absence could create gaps in how AI systems understand your content's context and provenance.

What actually drives AI Mode citations

If schema isn't the lever, what is? The research points toward a few things that matter more:

Content that directly answers the query. AI Mode is retrieving sources that contain the specific information needed to answer a prompt. Pages that clearly, comprehensively answer a question are more likely to be cited than pages that gesture at a topic.

Topical authority and entity clarity. Google's systems need to understand what your site is about and that you're a credible source on that topic. This comes from consistent content coverage, clear authorship, and strong internal linking -- not from schema types.

Being cited elsewhere. AI systems, including Google's, draw on signals about where content is discussed and referenced. Pages cited in other authoritative contexts tend to surface more in AI-generated answers.

Avoiding content that AI can't use. Pages with thin content, poor structure, or content buried behind JavaScript rendering are harder for AI systems to process -- schema or no schema.

Tools like Promptwatch can show you which prompts your competitors are being cited for but you're not, which is a much more actionable starting point than guessing which schema types to add.

The honest answer to the original question

Does schema markup help you get cited in Google AI Mode? The honest answer is: not directly, not in the way many people assume.

The data from three independent sources is consistent. Schema adoption among cited pages looks exactly like schema adoption across well-built sites in general. There's no special schema type that unlocks AI Mode citations. The LLMs processing these pages don't read JSON-LD as privileged metadata.

What schema does do is keep your technical foundation solid, help Google understand your entities, and improve your eligibility for rich results and AI Overviews within Google's own search features. Those are real benefits worth maintaining.

But if you're implementing FAQPage schema specifically because you think it will get you cited in AI Mode, the data suggests you're optimizing for the wrong thing. The time is better spent on content quality, topical depth, and understanding which specific questions AI models are answering in your space -- and whether your content actually answers them.

Tools worth knowing for AI search visibility

If you're trying to understand and improve your AI search presence more broadly, a few platforms are worth looking at:

BrightEdge has published research suggesting schema correlates with stronger brand presence in AI Overviews specifically -- which aligns with the Google-ecosystem findings above.

Semrush's AI Visibility Toolkit tracks how brands appear across AI search engines, useful for benchmarking your current citation footprint.

Ahrefs Brand Radar tracks brand mentions across AI search engines and was the source of the 1,885-page schema study referenced above.

Profound is an enterprise-focused platform for tracking AI search visibility across models, useful if you need deeper analytics on citation patterns.

The broader point: schema is one small piece of a much larger picture. Understanding where you're visible, where you're not, and what content gaps are costing you citations is a more productive focus than chasing schema types that the data doesn't support.