Technical

Myth:Serving a `.md` version of every page gets you cited more by Claude and custom GPTs.

Evidence:Some assistants (Claude with fetched URLs, Custom GPTs, MCP clients) parse markdown more cleanly than HTML, and .md endpoints reduce token cost when an agent retrieves your page. Direct citation lift is unproven in independent tests.

Myth:Publishing llms.txt is required to be discoverable by AI assistants.

Evidence:No major generative engine has publicly confirmed reading llms.txt for retrieval or citation as of mid-2026. It is a proposed convention (llmstxt.org), not an announced ranking or ingestion signal.

Myth:Publishing an llms.txt file makes your site discoverable to LLMs.

Evidence:No major LLM crawler reads llms.txt. It is a signaling gesture, not a technical requirement.

Myth:Rewriting competitor content is enough.

Evidence:For AI Overviews, mostly true: cited pages were statistically indistinguishable from uncited ranking pages on originality (medians 52 vs 55.5, p=.07) across 793 measured AI citations - AIO citation rides on Google rank, not information gain.

Myth:Aggregated round-ups can rank in AI answers.

Evidence:In most consumer verticals they dominate: 80% of CPG/retail AI answers cite at least one neutral reviewer or aggregator (Foglift 2026 benchmark - 375 buyer-intent responses, 25 verticals, 5 engines). The exception is tech SaaS, where vendor first-party content is cited 92.7% of the time.