Faro AI Signals·August 8, 2026

YouTube Citations Split Wildly by Platform; llms.txt Delivers Nothing

New data confirms that AI citation behavior varies dramatically across platforms — Perplexity cites YouTube at 38.7% while Gemini sits at 0.2% — and a multi-study consensus now puts the citation lift from llms.txt at zero. Treating AI search as one channel, or llms.txt as a shortcut, are both strategic errors.

ShareShare

This Week's Signals

1

YouTube AI Citations: Perplexity at 38.7%, Gemini at 0.2% — One Channel, Five Different Realities

Source: OtterlyAI ↗

Why it matters for your score

If you are creating video content expecting AI platforms to surface it, the platform you optimize for changes everything. Perplexity and Google AI Overviews drive the vast majority of YouTube citations while Gemini and Microsoft Copilot barely cite YouTube at all. Long-form videos account for 94% of all AI citations, meaning short clips and playlists are nearly invisible to AI systems regardless of platform.

2

llms.txt Adoption at 51.8% Among Tech-Forward Hosts, But Two Independent Studies Confirm Zero Citation Lift

Source: Digital Applied ↗

Why it matters for your score

More than half of tech-forward hosts now have the file, but that adoption is running ahead of any evidence it works. Two separate studies — one across roughly 300,000 domains, one from actual server log analysis across roughly 900 hosts — both report no measurable citation lift, and no verified requests from a frontier AI lab were confirmed in the log analysis. With no W3C, IETF, or schema.org recognition as of mid-2026, and Google on the record saying it does nothing for Search, llms.txt should not be treated as a visibility investment.

3

Google Engineers Detail Why Stateless MCP Was Unavoidable at Cloud Scale

Source: Google Developers Blog ↗

Why it matters for your score

Google hit a hard wall deploying MCP across cloud-native infrastructure because the original protocol required persistent state, handshakes, and session pinning. The 2026-07-28 spec drops session state for a stateless core, and all four Tier 1 SDKs already speak the new version. For any business building or buying agent-connected services, this spec change is now the baseline — older MCP integrations may not be compatible with current clients or servers.

What to do this week

  1. 1

    Audit your video content strategy by platform rather than treating AI search as one channel — the OtterlyAI data shows Perplexity at 38.7% and Gemini at 0.2% for YouTube citations, so prioritize long-form video content and check where your target audience's AI tools actually surface video results.

  2. 2

    Run your site through the Faro AI Readiness Scan (/tools/ai-readiness-scan) to confirm your visibility signals are based on content structure and schema rather than relying on llms.txt, which two independent studies now show produces no measurable citation lift.

  3. 3

    If you have an MCP-connected service or are evaluating one, verify whether your integration targets the 2026-07-28 spec — the legacy HTTP+SSE transport is officially deprecated with a one-year offramp, and servers on the new revision may not work with older clients.

Sources

← All editionsScan your site now →