Updated
Updated · KDnuggets · Sep 17
SerpApi Launches Markdown Output, Cutting Search-Result Tokens by Up to 74%
Updated
Updated · KDnuggets · Sep 17

SerpApi Launches Markdown Output, Cutting Search-Result Tokens by Up to 74%

1 articles · Updated · KDnuggets · Sep 17

Summary

  • 24,723 tokens for a sample "coffee" search fell to 6,435 with SerpApi's new Markdown output, and to 1,298 with further field restriction.
  • Markdown trims token-heavy JSON structure by removing internal tracking links, image URLs and duplicate metadata, while flattening results into YAML, tables and inline links better suited to LLM reading.
  • The feature is available across SerpApi's 100-plus APIs at no extra cost through output=md, the /search.md route or an Accept: text/markdown header.
  • JSON still fits pipelines that need typed fields such as prices, coordinates or ratings, but SerpApi argues Markdown can lower AI-agent costs and free context-window space where models only need readable summaries.

Insights

Could switching your AI search data to Markdown actually increase costs by disrupting prompt caching?
Does stripping JSON metadata for token efficiency secretly blind AI agents to crucial context clues?