The Schema-for-AI Claim Everyone Repeats (and Who Actually Tested It)

If an agency or consultant tells you to "add schema for AI visibility," ask them one question: where's the test? The advice is everywhere, but as Seroundtable reported, almost nobody had actually measured the claim — early independent experiments found no consistent evidence that schema directly improves AI search visibility, despite how widely it gets repeated.

That's a real problem for local businesses spending money on GEO. You can't prioritize fixes when the tactics themselves are unverified. So we looked at what evidence does exist — and it's more substantial than the advice industry gives it credit for.

Here's the evidence base behind this article:

• A controlled 7-platform experiment from OtterlyAI running 319 prompts across seven AI platforms from December 7, 2025 to March 7, 2026, tracking whether platforms could actually fetch and use schema markup.
• A/B extraction tests by practitioners like Julio C. Guevara, who ran extraction attempts "hundreds of times" with varied prompts to see whether LLMs could pull product data from structured data alone.
• A three-page controlled test published on Search Engine Land comparing pages with good schema, poor schema, and no schema for AI Overview appearance.
• Google's official guidance from its AI optimization guide, which addresses the question directly rather than leaving it to speculation.

One caveat matters before we go further: zero peer-reviewed studies exist on schema markup's impact on AI search visibility, per a Search Engine Land analysis. Everything we have comes from practitioner experiments, first-party tests, and official documentation. The field is also young — ChatGPT search only launched in October 2024.

That's exactly why measurement beats assumptions. At Briefsy, we treat AI visibility as something you track across ChatGPT, Perplexity, and Gemini — not something you optimize on faith. What follows is what the actual tests found.

Why Schema Fails on ChatGPT, Perplexity, and Claude — and Where It Still Works

If you've been told that adding schema markup is a shortcut to AI search visibility, the technical reality is sobering. The machinery that large language models use to read web pages actively works against JSON-LD structured data — and most AI platforms never even attempt to fetch it.

Two independent experiments explain why. Mark Williams-Cook's testing showed that when LLMs tokenize content into character sequences, schema gets effectively "destroyed" — the words "type" and "Organization" in "@type": "Organization" become indistinguishable from regular page text, according to experiments reported by Search Engine Roundtable. Separately, OtterlyAI's controlled experiment found that HTML-to-Markdown extraction pipelines often ignore or strip <script> tags entirely, removing JSON-LD before the model ever sees the page.

The fetch test results were equally blunt. Across 319 prompts and seven platforms between December 2025 and March 2026, six of seven AI platforms failed to fetch schema markup when directly prompted — only Gemini succeeded. Google AI Mode fared worse than failure: it hallucinated schema, confidently describing a "Service" schema that was never implemented on the test page. As OtterlyAI put it, "a platform that confidently returns wrong structured data is not reading schema."

A/B extraction testing confirmed the pattern. When researchers ran extraction attempts "hundreds of times" with varied prompts, LLMs could retrieve product details like price, colors, and SKU numbers only from pages where the information appeared as visible text — never from structured data alone.

Where does schema still work? In Google's traditional ecosystem, the measured gains are real:

• One site saw SERP features increase +377% over three months after adding schema
• AI Overviews appearances jumped +1,500% in the same period, though attributed to Google's index effects rather than schema directly
• A three-page controlled test found only the well-implemented schema page appeared in an AI Overview — though authors call the result "promising, but inconclusive"

That three-page test is the tentative bright spot. The page with complete, quality schema ranked for six keywords with a best position of 3 and earned an AI Overview appearance, while the poor-schema page ranked but never surfaced in AI Overviews. The authors themselves caution it could reflect unseen variables or "a coin-flip-style fluke of the Google algorithm."

The practical takeaway: schema is infrastructure for Google's parsers, not a lever for ChatGPT, Perplexity, or Claude — none of which have confirmed they preserve schema during crawling. This is exactly why tracking actual AI visibility matters more than checking technical boxes; it's the approach behind Briefsy's AI-search visibility checks across ChatGPT, Perplexity, and Gemini, which measure what platforms actually cite rather than what's in your source code. Watch your real citation rates, invest in visible on-page content, and treat schema as supporting groundwork — not your GEO strategy.

What Actually Drives AI Citations: Visible Content Over Markup

Here's the encouraging part: while schema markup underperforms as a direct AI visibility lever, the things that actually drive citations are well within your control. The evidence consistently points to one priority above all others — the visible text on your page.

The mechanics explain why. When Mark Williams-Cook ran extraction tests "hundreds of times" with varied prompts, LLMs could retrieve product details like price, colors, and SKU numbers only from pages where the information appeared as visible text — never from structured data alone, as reported testing showed. AI search platforms are built to process unstructured text; schema was invented for parsers that couldn't.

Google's own guidance reinforces this hierarchy. According to Google's AI optimization guide, unique, compelling content "will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide." Google also debunks the adjacent tactics that circulate in SEO communities — no llms.txt, no AI text files, no special "chunking" of content required.

The citation data backs this up. A December 2024 Search/Atlas study found no correlation between schema markup coverage and AI citation rates, according to an industry analysis that also notes zero peer-reviewed studies exist on schema's impact on AI search visibility to date.

So what should you actually do? Think of schema as infrastructure, not a magic bullet — worth maintaining properly, but never at the expense of content work. Your higher-priority levers look like this:

• Put every fact you want cited into visible, well-structured on-page text — prices, specs, answers, and definitions the model can extract directly.
• Invest in depth and originality: Google explicitly ranks unique, compelling content above any technical tactic for long-term generative AI presence.
• Keep schema accurate and matched to visible content, since it still drives rich results eligibility in Google Search.
• Measure outcomes rather than assume them — AI responses are non-deterministic, and visibility shifts are often algorithmic rather than tactic-driven.

That last point matters more than it sounds. In one controlled experiment, brand coverage rose about six percentage points after a schema rollout — but competitors who changed nothing saw equivalent or larger gains, leading the researchers to conclude the lift was algorithmic, not schema-driven. Without tracking, you can easily credit the wrong tactic.

This is exactly the gap Briefsy fills: it monitors your AI search visibility across ChatGPT, Perplexity, and Gemini alongside your competitors, reviews, and Google Business Profile, then delivers a prioritized brief of what changed and what to do next — so your effort goes to the levers that actually move citations, not the ones that merely sound technical.

How to Prioritize: A Practical Schema and AI-Visibility Playbook

So where does that leave you? With a clear set of priorities: keep schema for what it demonstrably does, and stop expecting it to do what the evidence says it can't.

Keep your existing well-implemented schema — but for the right reasons. Schema's proven value is eligibility for rich results like star ratings, prices, and availability, which can expand your SERP real estate, according to Semrush's guidance. One three-month measurement captured +377% in SERP features and +1,500% in AI Overview appearances — gains the authors attribute to Google's shared index, not to AI platforms reading your markup directly (OtterlyAI's experiment).

Beyond maintenance, the highest-leverage schema work is building a connected entity graph rather than isolated markup. Aimee Jurenka recommends using @id and @graph to link Organization, Person, and Article nodes — with worksFor, authoredBy, and sameAs relationships — so brand ownership and authorship stay clear regardless of page layout. This is "infrastructure, not a magic bullet," but it's one of the few things you can control that platforms like Bing and Google AI Overviews explicitly use.

Your practical playbook, in priority order:

• Match markup to visible text. LLMs extract price, colors, and SKU data only from on-page text — never from structured data alone, per extraction tests.
• Skip llms.txt, AI text files, and "chunking" rewrites. Google's own AI optimization guide debunks these tactics outright.
• Invest in unique, compelling content first — Google says it matters more than any technical tactic.
• For products, evaluate AI Merchant programs (ChatGPT's Merchant Program, Perplexity's "Buy with Pro") as potentially better levers than schema.

On measurement: don't rely on assumptions. Use Search Console's Generative AI performance report to track how Google's AI surfaces treat your content, and remember that a December 2024 study found no correlation between schema coverage and AI citation rates — so treat schema audits as hygiene, not a primary GEO lever. Tools like Briefsy that check visibility across ChatGPT, Perplexity, and Gemini weekly can tell you whether any of this work is actually moving the needle.

The honest bottom line: schema quality may modestly help AI Overviews, and it definitely helps Google Search. It does not directly win you citations in ChatGPT or Perplexity. Budget accordingly.

Measure, Don't Assume: Tracking Your AI Visibility Over Time

If there's one lesson from every experiment we've covered, it's this: single-run AI visibility tests can mislead you badly. AI responses are non-deterministic — the same prompt can return different answers on different days — and that volatility makes one-off audits almost useless as evidence.

OtterlyAI's own three-month experiment is the perfect cautionary tale. They tracked their brand coverage climbing from a ~27% November baseline to the low 30s after adding schema, which looked like a win — until they noticed competitors who had changed nothing saw equivalent or larger gains. Their conclusion: "the December lift was algorithmic, not schema-driven." A single before-and-after snapshot would have told a completely false story.

The measurement problem runs deeper than volatility. According to Search Engine Land's analysis, zero peer-reviewed studies exist on schema's impact on AI search visibility, and the field itself is young — ChatGPT search only launched in October 2024. Even the Search Engine Land three-page test that favored well-implemented schema acknowledged its result could reflect "unseen variables" or "a coin-flip-style fluke of the Google algorithm."

So what does responsible tracking look like? A few principles:

• Run the same prompt set repeatedly over weeks, not once — trends matter more than individual answers.
• Check across multiple platforms, since 6 of 7 AI platforms failed to fetch schema and behavior differs sharply between ChatGPT, Perplexity, and Gemini.
• Compare against competitors, so you can tell algorithmic shifts from changes you actually caused.
• Use Google Search Console's Generative AI performance report for AI Overviews and AI Mode data.

This is exactly the measurement mindset behind Briefsy. Instead of a one-time audit that's stale within a month, Briefsy acts as your AI market analyst — running recurring AI-search visibility checks across ChatGPT, Perplexity, and Gemini, watching your competitors and Google Business Profile alongside your own site, and delivering a prioritized brief of what changed and what to do next. When your brand coverage moves, you'll know whether it's your work or the algorithm's.

Treat AI visibility the way you'd treat rankings: as an ongoing metric, monitored over time, interpreted against a baseline. The businesses that win in AI search won't be the ones that added schema once and hoped — they'll be the ones measuring consistently and acting on what the data actually shows.