Schema Markup Barely Moves the Needle for AI Recommendations — Authority and Clarity Do
The SEO industry has spent the last year telling brands that schema markup is the key to AI visibility. Add JSON-LD structured data to your pages, implement FAQ schema, create an llms.txt file, and AI platforms will recommend your brand. It sounds logical. It is mostly wrong.
A study published by First Page Sage in September 2026 measured the actual impact of structured data on AI recommendations across 4,213 commercial prompts sent to ChatGPT, Google Gemini, and Claude. The researchers analyzed 1,089 brands across 14 industries, ran 657 agentic shortlisting and purchasing tasks, and reviewed an average of 37 third-party sources per brand.
The headline finding: schema markup raised recommendation rates by 3.7 percentage points on its face — but by only 0.4 points after controlling for authority signals. The brands that had schema markup also tended to be the brands with stronger authority. The schema itself was almost irrelevant.
This study does not mean structured data is useless. It means the industry has been optimizing the wrong kind of structured data.
The Numbers That Matter
The First Page Sage study measured "recommendation rate" — the share of relevant commercial prompts in which an AI platform named a brand as a recommended option. Here are the findings that should reshape how brands think about AI visibility.
Authority signals dominate standard AI search. In standard AI search (non-agentic prompts), authority signals accounted for 80.4% of recommendation influence. List mentions — appearing on curated "best of" and comparison lists across the web — contributed 33.1% of that influence on their own. Being mentioned on authoritative third-party sources matters more than anything you do on your own website.
Content clarity is the threshold that separates visible brands from invisible ones. The study scored each brand's "clarity" — how clearly and specifically it communicates what it offers — on a 1 to 10 scale. Brands scoring 5 or 6 achieved an 11.2% recommendation rate. Brands scoring 7 or 8 jumped to 23.6%. That is a 2.1x increase from crossing a clarity threshold. No schema markup, no technical optimization, no structured data implementation — just describing what you do more clearly.
Third-party consistency outweighs on-site consistency. Brands with consistent information across third-party sources achieved a 17.1% recommendation rate. Brands with consistent on-site information but inconsistent third-party data achieved 13.7%. The AI platforms cross-reference what your website says against what directories, review sites, and industry lists say about you. If the information does not match, confidence drops.
FAQ schema actually hurts. After controlling for authority, FAQ schema reduced recommendation rates by 0.3 points. The researchers hypothesize that FAQ schema may signal low-authority content — the kind of thin "frequently asked questions" pages that weaker brands create to appear comprehensive. Whatever the mechanism, the data is clear: FAQ schema is not helping.
llms.txt files are neutral. The llms.txt file format — designed specifically to help AI models understand website structure — showed a -0.1 point impact after controlling for authority. Functionally zero. The file that was created explicitly for AI consumption does not measurably improve AI recommendations.
Product schema is the exception. In e-commerce, Product schema markup raised recommendation rates by 4.8 points even after controlling for authority. This is the one type of structured data where the markup itself carries independent value — because it provides specific, comparable attributes (price, availability, specifications) that AI agents need to make purchasing recommendations.
Why Schema Markup Does Not Work the Way People Think
The logic behind schema markup for AI visibility seems straightforward: AI models need structured data to understand your brand, schema provides structured data, therefore schema improves AI understanding.
The flaw is in the first premise. AI models do not need your schema markup to understand your brand. They have already read your website, your competitors' websites, review sites, industry publications, social media, Wikipedia, Crunchbase, and dozens of other sources. By the time a user asks ChatGPT to recommend a product in your category, the model has already formed a representation of your brand from the entire web.
Schema markup tells the model the same things it already knows — your name, your address, your product categories — in a format it does not prefer for building brand representations. The model's training data is natural language. Its strongest signals come from natural language mentions, citations, and descriptions across the web.
This does not mean schema is valueless. It means schema's value is narrow and specific: it helps when it provides information the model cannot easily extract from natural language. Product specifications, pricing, inventory status, booking availability — these are the structured data points that AI models genuinely need in machine-readable format because they require precision and real-time accuracy.
Agentic Tasks Change the Equation — Slightly
The study also ran 657 agentic tasks — prompts that asked AI models to shortlist vendors, compare products, and recommend purchases. In these agentic contexts, structured data's influence increased to 4.6% versus 0.9% in standard search.
The reason is mechanical. When an AI agent is tasked with actually completing a purchase or booking, it needs machine-readable paths to do so. A booking widget that requires JavaScript rendering is harder for an agent to interact with than a structured API endpoint. A product listing with machine-readable price and availability attributes is easier to compare than one where those details are embedded in marketing copy.
The study found that machine-readable booking and checkout paths produced a 17.4-point increase in agentic shortlist rates over clear text alone (36.8% versus 19.4%). This is the largest measurable impact of any structured data factor in the entire study — and it has nothing to do with schema markup. It is about making your transaction infrastructure accessible to agents.
This finding aligns with what the broader agentic commerce infrastructure is demanding. The Universal Commerce Protocol and Agentic Commerce Protocol both require structured product and transaction data. But they require operational structured data — pricing feeds, inventory APIs, checkout endpoints — not SEO-oriented schema markup.
What Actually Drives AI Recommendations
If schema markup is not the answer, what is? The study points to three factors that together account for the vast majority of AI recommendation influence.
1. Authority Through Third-Party Mentions
Appearing on curated lists, being cited in industry publications, getting reviewed on authoritative platforms — these signals account for the largest share of recommendation influence. AI models treat the web as a citation network. Brands that are mentioned frequently and positively across trusted sources get recommended. Brands that are not, do not.
This is not new in concept — it is Google's original PageRank insight applied to AI. But the mechanism is different. Google counts backlinks. AI models count mentions, context, and sentiment across their entire training corpus. A brand mentioned in a Wall Street Journal article, three industry comparison posts, and five Trustpilot reviews has a different authority footprint than a brand with the same schema markup but no third-party presence.
For brand data, this means your brand identity needs to be consistent and accurate everywhere it appears — not just on your website. Directories, review platforms, social profiles, industry databases, and comparison sites all feed the authority signal that AI models use to decide whether to recommend you.
2. Content Clarity Above the Threshold
The 2.1x jump in recommendation rates between clarity scores of 5-6 and 7-8 suggests a threshold effect. Below a certain level of clarity, AI models are uncertain about what your brand actually offers and hedge their recommendations. Above that threshold, the model has enough confidence to name you as a specific recommendation.
Clarity is not about having more content. It is about having more specific content. A brand that says "we provide innovative solutions for modern businesses" scores low on clarity. A brand that says "we extract logos, colors, and fonts from any website URL and return structured JSON through a REST API" scores high.
The implication for brand data: your brand description, product descriptions, and value propositions need to be concrete and specific enough that an AI model can confidently categorize what you do, who you serve, and how you differ from alternatives. Vague positioning is not just bad marketing — it is an AI visibility penalty.
3. Cross-Platform Consistency
When the information on your website matches the information on third-party sources, AI models have higher confidence in recommending you. When there are discrepancies — different addresses, different service descriptions, inconsistent pricing — confidence drops.
This is where brand data infrastructure becomes critical. A brand intelligence API can extract your current brand data — logo, colors, description, contact information, social profiles — and verify it against what appears across the web. Inconsistencies that you might never notice manually become visible when compared systematically.
The study found that third-party consistency (17.1% recommendation rate) outperformed on-site consistency (13.7%). Your own website being well-structured matters, but the information about your brand scattered across the rest of the internet matters more.
What Brands Should Actually Do
The First Page Sage study suggests a reordering of priorities for brands that want AI platforms to recommend them.
Stop treating schema markup as an AI visibility strategy. Schema markup has SEO value for traditional search and provides specific utility for e-commerce product data. But implementing extensive schema markup specifically to improve AI recommendations is optimizing a factor that accounts for less than 1% of recommendation influence in standard search.
Invest in third-party presence. Get listed on authoritative industry directories. Pursue reviews on trusted platforms. Contribute to comparison content and best-of lists in your category. Build the citation network that AI models actually use to evaluate authority.
Raise your clarity score. Audit every page on your website for specificity. Replace vague value propositions with concrete descriptions. Ensure that an AI model reading your homepage can immediately understand what you sell, who you sell it to, and what makes you different. The clarity threshold is binary — you are either above it or below it — and the gap between the two is a 2x difference in recommendation rates.
Make your brand data consistent everywhere. Audit your brand information across directories, review sites, social profiles, and industry databases. Ensure your name, address, description, and offerings match across every platform. Use brand data APIs to systematically verify consistency rather than manually checking each source.
Build machine-readable transaction paths for agentic commerce. If your business involves transactions — e-commerce, bookings, subscriptions — invest in making those paths accessible to AI agents. This is the one area where structured data produces large, measurable gains. Implement product feeds, booking APIs, and checkout endpoints that agents can interact with programmatically.
Monitor your AI recommendation rate. The traditional SEO metrics — rankings, organic traffic, click-through rates — do not capture AI visibility. Track how often and where AI platforms mention your brand in response to relevant commercial prompts. This is the metric that matters for the next wave of brand discovery.
The Schema Markup Industry Will Not Like This
An entire ecosystem of tools, consultants, and agencies has been built around the premise that schema markup drives AI visibility. The First Page Sage study challenges that premise directly. Schema markup is a useful web standard with specific, narrow applications. It is not the lever that determines whether AI platforms recommend your brand.
The actual levers — authority, clarity, and consistency — are harder to optimize than adding JSON-LD to your pages. They require building a real brand presence across the web, writing content that is genuinely specific and useful, and maintaining accurate brand data everywhere your brand appears.
That is harder work. It is also what actually moves the numbers.
Build your AI-ready brand identity. Extract your brand fingerprint and ensure your brand data is consistent across every platform where AI models look.