Blog · Schema markup for AI search platforms
Schema Strategies for Different AI Models: What Works
Dharini Shah · September 7, 2026
AI platforms do not all use schema markup the same way, but the evidence does not support building a separate schema strategy for each model. Google says no special markup is needed to appear in AI Overviews or AI Mode. Microsoft has said schema helps its language models understand content. Tests suggest ChatGPT and Perplexity may treat JSON-LD as ordinary page text when they read it, if they see it at all. And for shopping answers, product feeds matter more than on-page markup.
The practical strategy is therefore one accurate, consistent structured data layer that serves search engines well, plus platform-specific feeds where commerce is involved. This article explains what each major platform has said or what testing shows, where the evidence is thin, and which schema types are worth prioritizing.
What schema markup does, and what it does not do
Schema markup is structured data, usually written in JSON-LD, that describes the entities on a page using the Schema.org vocabulary: an organization, a product, an article, an event or a local business. Traditional search engines use it to understand pages and to generate rich results such as product details or review stars.
For AI systems, there are three possible roles:
- Indirect role through search indexes. Systems that retrieve from Google or Bing benefit from whatever understanding those indexes derive from structured data.
- Direct parsing. A model reading a page could interpret the JSON-LD as structured facts.
- Text role. A model could read the JSON-LD as ordinary text, gaining the facts without using the structure.
The public evidence supports the first role clearly, offers mixed signals on the other two and does not show large citation gains from adding schema.
What each platform has said or shown
Google AI Overviews and AI Mode
Google's documentation on AI features states: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." Google's AI features draw on its core search systems, so structured data helps in the same ways it helps regular search: understanding entities and powering eligible rich results.
Two recent changes matter for planning:
- FAQ rich results are gone. Google stopped showing FAQ rich results in May 2026 and removed related Search Console reporting in June 2026. FAQPage markup can stay on pages without causing problems, but it no longer produces a visible search feature. HowTo rich results were removed earlier, in 2023.
- Shopping depends on Merchant Center. Google's shopping experiences, including in AI Mode, rely heavily on product data from Google Merchant Center alongside on-page Product structured data.
Strategy for Google: keep structured data that supports active rich results, such as Product, Organization, Article, LocalBusiness, Review snippets and Breadcrumb, accurate and consistent with visible content. Do not expect markup to unlock AI Overview inclusion by itself.
Microsoft Copilot and Bing
Microsoft is the one major AI provider that has spoken positively about schema for its language models. Fabrice Canel of Bing said at SMX Munich that schema markup helps Microsoft's LLMs understand content. He also recommended IndexNow for notifying Bing about fresh content.
Strategy for Copilot: well-formed JSON-LD for your core entities, validated and consistent with page content, plus IndexNow for timely updates. Bing Webmaster Tools is worth setting up if you have not already.
ChatGPT
OpenAI has not published guidance saying ChatGPT search uses Schema.org markup for ranking or citation. Its publisher documentation focuses on crawler access through OAI-SearchBot.
Testing gives mixed signals. SEO consultant Mark Williams-Cook ran a test, published in February 2026, in which an address placed only inside invalid JSON-LD was still picked up by ChatGPT and Perplexity. His conclusion was that the systems read whatever is in the HTML and that valid schema did not matter. The Ahrefs schema study, by contrast, referenced findings that major AI systems extract only visible HTML during direct retrieval. The honest conclusion is that behavior likely varies by system and fetch method and may change over time.
For commerce, OpenAI's Agentic Commerce Protocol, announced with Stripe in September 2025, lets merchants provide product data and enable purchases in ChatGPT. OpenAI states that product results are "organic and unsponsored, ranked purely on relevance to the user."
Strategy for ChatGPT: make sure every fact in your structured data also appears in visible text, and never rely on markup to communicate something the page does not say. For ecommerce, evaluate OpenAI's merchant programs and product feed options.
Perplexity
Perplexity's crawler documentation covers PerplexityBot and Perplexity-User but does not describe schema use. The Williams-Cook test suggests Perplexity may read JSON-LD content as text. There is no public evidence of a Perplexity-specific schema advantage.
Strategy for Perplexity: clear visible content, crawler access and consistency between markup and text.
Claude
Anthropic has not published guidance on schema use by Claude's web search. Its public documentation focuses on its crawlers: ClaudeBot, Claude-SearchBot and Claude-User.
Strategy for Claude: the same baseline as for ChatGPT and Perplexity.
Summary by platform
| Platform | Official position on schema | Evidence from testing | Recommended focus |
|---|---|---|---|
| Google AI Overviews and AI Mode | No special markup needed for AI features | Ahrefs found no clear citation effect from adding schema | Rich-result schema, Merchant Center for products |
| Microsoft Copilot and Bing | Schema helps Microsoft's LLMs understand content | Limited public testing | Well-formed JSON-LD for core entities, IndexNow |
| ChatGPT | No public schema guidance | Mixed: may read JSON-LD as text, or only visible HTML | Visible text matching markup, commerce feeds for shopping |
| Perplexity | No public schema guidance | May read JSON-LD as text | Visible text, crawler access |
| Claude | No public schema guidance | Little public testing | Visible text, crawler access |
Which schema types to prioritize
Given that evidence, prioritize schema types that describe your core entities accurately and that search engines actively use.
Organization. Describes your company: name, URL, logo, description and sameAs links to official profiles such as LinkedIn, Crunchbase, Wikipedia or Wikidata where they exist. This helps search engines connect your brand to its profiles and supports consistent entity understanding.
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Example Software Inc.",
"url": "https://www.example.com",
"logo": "https://www.example.com/logo.png",
"description": "Example Software makes scheduling tools for dental clinics.",
"sameAs": [
"https://www.linkedin.com/company/example-software",
"https://www.crunchbase.com/organization/example-software"
]
}
Product and Offer. Essential for ecommerce: name, brand, GTIN where applicable, price, currency, availability and reviews. Keep it synchronized with your product feeds.
SoftwareApplication. Useful for software products: application category, operating system and offers.
Article. For editorial content: headline, author with a link to an author page, datePublished and dateModified. Accurate dates support freshness signals, which matter for some AI assistants.
LocalBusiness. For location-based businesses: address, hours, phone and service area, matched exactly to your business profiles.
FAQPage. Harmless to keep, but no longer produces Google rich results. The visible question-and-answer content is what AI systems read, so focus on writing strong answers rather than on the markup.
The consistency rule
The most important schema principle for AI search is consistency. Your structured data, visible page content, product feeds, business profiles and third-party listings should all state the same facts. Contradictions, such as a different price in schema than on the page, or a different company description in markup than on your LinkedIn profile, create exactly the kind of conflicting signals that lead to inaccurate AI answers.
A practical approach is to generate structured data from the same source of truth that produces your visible content and feeds, rather than maintaining markup by hand.
Common mistakes
Treating schema as an AI ranking lever. The evidence does not show large AI citation gains from schema alone.
Hiding facts in markup. If important information exists only in JSON-LD, some AI systems may never see it.
Letting markup drift. Stale prices or outdated descriptions in schema contradict your visible content.
Building model-specific markup. There is no public evidence that any AI platform rewards a special schema variant.
Ignoring feeds for ecommerce. For shopping answers, Merchant Center and AI commerce feeds typically matter more than page markup.
A hypothetical example
A hypothetical online furniture retailer spends a month adding extensive schema and sees no change in AI shopping answers. An audit shows its Merchant Center feed has missing GTINs and inconsistent availability, its product pages show different sale prices from the feed, and its Organization markup uses an old company name. The team fixes the feed, synchronizes prices across the feed, markup and pages from one product database, and corrects the Organization data. The value came from consistency and feed quality, not from adding more schema types.
How Bob Builds AI helps
Bob Builds AI's Brand Memory keeps company and product facts in one place so that content, markup and profiles stay consistent. Agent Analytics helps teams understand how AI crawlers and agents access their sites, and the AI search readiness resources cover the technical foundations. For an industry-specific example, see schema markup for law firms.
FAQ
Do AI models like ChatGPT read schema markup?
The evidence is mixed. A February 2026 test by Mark Williams-Cook found ChatGPT and Perplexity picked up information placed only in JSON-LD, suggesting they read it as page text. Other research indicates some AI systems extract only visible HTML during retrieval. OpenAI has not published schema guidance. Keep every important fact in visible text as well.
Does Google require schema for AI Overviews?
No. Google states that you do not need new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode. Standard SEO eligibility applies. Structured data still supports Google's rich results and entity understanding, which indirectly supports the same systems.
Should I remove FAQ schema after Google dropped FAQ rich results?
You do not need to. Google stopped showing FAQ rich results in May 2026 but says FAQ markup can remain on pages without causing problems. It no longer produces a search feature, so do not invest heavily in adding it. The visible question-and-answer content remains useful for readers and AI systems.
Does Bing Copilot use schema markup?
Microsoft has said it does. Bing's Fabrice Canel stated at SMX Munich that schema markup helps Microsoft's LLMs understand content. That makes well-formed JSON-LD for your core entities worthwhile if Copilot matters to your audience, alongside IndexNow for timely content updates.
What schema types matter most for AI search?
Prioritize schema that accurately describes your core entities: Organization with sameAs links, Product and Offer for ecommerce, SoftwareApplication for software, Article with author and dates for editorial content, and LocalBusiness for location-based businesses. Accuracy and consistency with visible content matter more than the number of types.
Does adding schema increase AI citations?
Not reliably, based on current evidence. An Ahrefs study of 1,885 pages that added schema found no clear citation effect for pages already appearing in AI answers. Schema may still help pages that are not yet visible, and it supports search features, but it should not be treated as a primary AI visibility tactic.
How is schema different from product feeds for AI shopping?
Schema describes products on your web pages. Product feeds, such as Google Merchant Center feeds or feeds submitted for AI commerce programs, send structured product data directly to platforms. For shopping experiences in Google and ChatGPT, feeds are typically the main data source, so keep feeds and on-page schema synchronized.
Should I create different schema for different AI models?
No. There is no public evidence that any AI platform rewards model-specific schema variants. One accurate, validated structured data layer that matches your visible content, plus platform feeds for commerce, is the most defensible approach.
Conclusion
Different AI platforms treat schema differently, but not in ways that justify model-specific markup. Google says markup is not required for its AI features, Microsoft says schema helps its models, and testing suggests other assistants read JSON-LD as text or not at all. The one consistent lesson is that accuracy and consistency beat volume.
The next step is a consistency audit: compare your Organization and Product markup, visible page content, product feeds and business profiles, and fix every contradiction. Then keep all of them generated from one source of truth. Bob Builds AI can help you maintain that single version of your brand facts and monitor whether AI answers reflect it.