Bing’s LLMs are confirmed to read your schema markup, Google says it “depends” and calls most SEO claims about it “wishful thinking,” and OpenAI has only confirmed using structured data for shopping feeds — everything else circulating about schema and AI citations is inference, not documented fact.
Ask five SEOs whether JSON-LD schema markup improves your odds of being cited by ChatGPT, Perplexity, or Google’s AI Overviews, and you’ll get five confident answers backed by five different multipliers — 2.3x, 2.5x, 3.2x, “40% higher,” “0.546x.” None of these figures trace back to a published methodology, sample size, or control group. That inconsistency is itself the finding: the claims move around because nobody agreed on how to measure them. What’s actually on the record from the people who run these systems is much narrower, and worth separating from the noise.
What each AI platform has actually said
Only one major AI platform has gone on record confirming that its language models use schema markup as a distinct signal, separate from the page’s visible text. Fabrice Canel, Microsoft’s principal product manager for Bing, said as much on stage at SMX Munich, confirmed afterward in a LinkedIn thread and reported by Search Engine Land on March 20, 2025: Bing’s LLMs use schema to help understand content, and “Gen AIs value fresh content in particular, partly as a reference check of their LLM training data.”
Google has never issued that kind of confirmation. The closest thing on record is Google Search team member John Mueller’s personal reply on Reddit, reported by Search Engine Roundtable on January 2, 2026, explicitly flagged as “not official guidance”: “the short answer is yes, no, and it depends.” Mueller said some features genuinely need structured data — pricing, shipping, and availability for Shopping results are “basically impossible to read in high fidelity & accurately from a text page” — while for most other schema types “there’s a lot of wishful thinking. Always has been, and will continue to be. Your ‘best geo insurance comparison site’ isn’t going to rank better by adding insurance markup.” Google has also repeated, multiple times since October 2023, that structured data is not a ranking factor; it only governs eligibility for rich-result display.
OpenAI’s only public confirmation is narrower still: ChatGPT Search’s shopping results, rolled out in April 2025 per Search Engine Land / Wired, are “powered by structured data from third-party sources.” That’s a commerce-feed integration, not a statement about how ChatGPT parses JSON-LD on an ordinary blog post or documentation page. Perplexity has issued no public statement on schema use at all.
What an actual test found
The one piece of independent testing on this question points the other way from the marketing claims. SEO researcher Mark Williams-Cook built a fake company page for a made-up brand (“DUCKYEA”), and instead of putting the company’s address in visible text, he embedded it only inside invalid, made-up JSON-LD schema. He then prompted both ChatGPT and Perplexity for the address. Both models returned it — reported by Search Engine Roundtable on February 6, 2026, following an earlier version of the same test. Since the schema wasn’t valid, Williams-Cook concluded the models were reading it as ordinary page text, not as a validated, structured signal: “the LLM agent is simply picking up whatever you are listing in the HTML. It does not matter if it is valid schema… schema is not being used in the explicit sense it was designed for with those systems.”
That’s consistent with what would happen if these models treat JSON-LD blocks the same way they treat any other string of text on the page — useful as content, but not parsed against the schema.org vocabulary to check whether the markup is even legal.
| Platform / test | Documented position | Source | Date |
|---|---|---|---|
| Bing / Copilot | Confirmed: schema helps its LLMs understand content | Fabrice Canel via Search Engine Land | Mar 20, 2025 |
| Google Search / AI Overviews | No official statement; personal Mueller view is “it depends” — some features need it, most schema-for-ranking claims are “wishful thinking” | John Mueller via Search Engine Roundtable | Jan 2, 2026 |
| OpenAI / ChatGPT | Confirmed use of structured product feeds for Shopping results only | Search Engine Land / Wired | Apr 2025 |
| Perplexity | No public statement on schema use | — | — |
| Independent test (ChatGPT + Perplexity) | Both surfaced data from invalid, fake JSON-LD identically to visible text — suggests no schema validation, just text ingestion | Mark Williams-Cook via Search Engine Roundtable | Feb 6, 2026 |
The rich-results signal is also shrinking on Google itself
Whatever confidence schema markup buys you, it’s worth noting that even Google’s own rich-result program — the most concrete, measurable payoff schema has ever offered — has been narrowing for years. Google restricted FAQ rich results to “well-known, authoritative government and health websites” in August 2023 and deprecated HowTo rich results the same year (mobile in August, desktop in September). Search Engine Journal reported that FAQ rich results stopped appearing in Google Search entirely on May 7, 2026, with the FAQ search-appearance filter, the rich-result report, and Rich Results Test support for it removed the following month. FAQPage remains a technically valid schema.org type, and Google has said it will still parse the markup to understand a page — but the visible payoff most sites built FAQ schema for is gone.
That deprecation matters for the AI-visibility framing specifically, because a lot of “schema helps you get cited by AI” advice is really recycled “schema gets you a rich result” advice from the pre-2023 SEO playbook, applied to a new set of engines without new evidence.
What’s actually worth implementing
None of this means schema markup is worthless — it means the case for it should rest on what’s documented, not on borrowed multipliers. Google’s own developer documentation still treats JSON-LD as its only recommended structured-data format, and the highest-leverage types across current guidance remain Organization (with sameAs links to verified profiles), Article or BlogPosting (with datePublished, dateModified, author, and publisher), and Product for commerce pages. These are the types every confirmed use case — Bing’s content understanding, Google’s Shopping-style rich results, OpenAI’s product feeds — actually touches. FAQPage and HowTo are safe to keep (they won’t hurt anything) but shouldn’t be prioritized on the promise of AI citation gains that no platform has confirmed.
Schema markup also can’t do its job if crawlers can’t reach the page it’s sitting on in the first place — see our breakdown of AI crawler permissions and robots.txt rules for GPTBot, ClaudeBot, and PerplexityBot. And schema addresses a different layer of the same problem that llms.txt tries to solve: schema structures the page for machines that are already reading it, while llms.txt tries to hand a model a curated summary instead of making it parse the page at all. Both remain unproven at scale for citation outcomes, which is the throughline of everything in our AI visibility research hub: platforms confirm mechanisms selectively, and most of the rest is inference dressed up as certainty. The same evidence bar applies to the failure-mode side of these platforms — our ledger of documented failure modes across frontier models uses the same rule: no vibes, only sourced claims.
FAQ
Does schema markup guarantee my content gets cited by ChatGPT or Perplexity?
No. Neither company has confirmed that general-purpose JSON-LD (as opposed to commerce feeds, in OpenAI’s case) affects citation odds, and one independent test found both models reading invalid schema identically to plain text.
Does adding schema improve my Google ranking?
Google has repeated since at least October 2023 that structured data is not a ranking factor. It affects eligibility for rich-result display in Google Search, and per Mueller’s January 2026 comments, may help specific Google features like Shopping parse pricing and availability data.
Is it worth adding FAQPage schema in 2026?
For Google specifically, the visible rich-result payoff is gone as of May 7, 2026 for all but a small set of authoritative government and health sites. The markup is still valid and harmless, but it shouldn’t be a priority investment based on rich-result or AI-citation promises.
Last updated July 27, 2026. This page is refreshed as benchmarks and scores move.