Results from a Controlled Bot Experiment

GEO advice is everywhere: add schema markup, restructure for "chunkability," publish an llms.txt, compress your content. What's missing from nearly all of it is evidence. Vendor blogs cite correlations; nobody controls for anything.

This session presents a controlled experiment run on a live production site. Pages were randomized into test arms at the URL level, variants were served to AI crawlers via edge middleware, and behavior was measured on both sides of the pipeline: what GPTBot, ClaudeBot, PerplexityBot, and their retrieval-time counterparts fetched, down to the byte, and what actually surfaced downstream in AI-generated answers, tracked through weekly citation sampling against the major answer engines.

We'll cover:

• The experiment design: randomization, baselines, and why per-request bot splits fail
• The interventions tested, from structural HTML changes to extractable-fact density
• What moved, what didn't, and where the null results are just as instructive
• A replicable methodology you can run on your own site, whatever stack you're on

Whether you own content strategy, SEO, engineering, or analytics, you'll leave with actual data in a space currently running on folklore.
Free, virtual, one hour including Q&A. Recording available to registrants.