Schema Markup's Real Impact on AI Search | OtterlyAI Experiment

Schema markup is one of the most consistently recommended tactics in technical SEO, and for good reason. It has a 10+ year track record of helping Google surface rich results, trigger featured snippets, and improve SERP feature eligibility, benefits that remain fully intact today. As AI Search platforms have grown in prominence, a natural question has followed: if schema helps traditional search engines understand content, does it also help AI Search platforms?

To find out, OtterlyAI ran a controlled experiment to test that assumption. We implemented five types of schema markup on live pages, directly prompted on 7 major AI Search platforms to retrieve that structured data, and tracked brand coverage across 319 prompts over a 3 month observation window.

Key Findings (TL;DR)

• 6 out of 7 AI Search platforms were unable to fetch or correctly interpret schema markup when directly asked. Only Gemini retrieved the correct JSON-LD data.
• Google AI Mode hallucinated schema types that were not implemented, including a “Service” schema that did not exist on the page.
• Increase detected on Google AI Mode and AI Overviews: after 3 months, SERP features increased by +377% and AI Overviews by +1,500%. Schema still delivers meaningful SEO results.
• Impact on other AI Search platforms: For AI Search platforms beyond Google’s own ecosystem, however, its direct impact appears to be much more limited.
• AI Search Engines don’t need schema markup the way traditional crawlers do. Schema was invented for algorithms that couldn’t handle unstructured text. AI Search is built to do exactly that.
• Use AI Merchant programs for products. ChatGPT’s Merchant Program & Perplexity’s Buy with Pro are better alternatives than schema for AI platforms. The future is AI buying products directly.

Why Studying Schema Markup Matters for GEO

AI Search visibility is increasingly determined by citation selection rather than page ranking. When ChatGPT, Perplexity, or Google AI Overviews answer a query, they retrieve and cite sources, and the logic that determines which sources they pull from is fundamentally different from the logic that governs traditional SEO.

Schema markup sits at the intersection of both worlds, which makes it worth testing carefully. It was designed to annotate structured meaning on top of raw content, to help parsers that couldn’t read unstructured text understand what a page was about. If AI platforms are now doing that parsing natively, as large language models are designed to do, then schema’s role in GEO may be very different from what it is in traditional SEO.

That’s the question this experiment was designed to answer with real data.

Scope of Study

This GEO Experiment was conducted from 7th of December 2025 to 7th of March 2026 across seven AI platforms: ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Gemini, and Claude.

The following Schema Markup types were implemented and tested:

• FAQ Schema
• Organization Schema
• SoftwareApplication Schema
• Review Snippets Schema
• Article / BlogPosting Schema

On Dec 7th. Schema Markup was placed on the entire Otterly.ai website. For the purpose of this GEO experiment, we’ll limit the scope to discussing the impact on the following URLs:

• otterly.ai/agencies, a high-traffic product landing page
• otterly.ai/blog/otterlyai-ranks-10-on-g2s-2026-best-software-awards/, a recent news and award announcement post
• https://otterly.ai/blog/ai-brand-monitoring-tools-the-ultimate-guide-to-protecting-and-growing-your-brand-in-2025/ an old blog that’s been performing well for over 1 year.
• https://otterly.ai/about

The study has 3 distinct tests:

1. A direct capability test: Each AI platform was prompted to fetch and return the raw schema markup code from both URLs, testing whether AI systems can access and interpret structured data at all.
2. Impact on AI brand visibility test: OtterlyAI’s Brand Coverage was tracked across 319 prompts in the US market using OtterlyAI’s AI Search Monitoring, from November 1, 2025 through March 2026, with schema implemented on December 7, 2025 as the intervention point. Competitor brand coverage was tracked simultaneously to control for algorithmic shifts.
3. Unique information test: A piece of information was added to a specific page via an FAQ Schema that could not exist anywhere else on the web.

Methodology

Step 1: Select Schema Types That Represent Real GEO Use Cases

Three schema types were selected to cover the most common structured data patterns relevant to a SaaS and content brand:

• FAQ Schema: Annotates question-and-answer content, directly relevant to how AI platforms decompose informational queries
• Organization / SoftwareApplication Schema: Establishes entity-level brand signals and product classification
• Article / BlogPosting Schema: Provides authorship, publication date, and content-type signals for editorial content

Step 2: Implement All Schema in JSON-LD

All schema was implemented using JSON-LD format, the approach recommended by schema.org and Google’s developer guidelines, placed inside