React Tax[01]
React Tax
The part of a page that's framework rather than content — class strings, hydration data, scripts, chrome. People never see it. Agents read every byte.
Measure it
[02]This docs site is a Next.js app, and it serves every page twice: as HTML, and as Markdown at the same path plus .md. Pick one:
Measuring…
Why agents pay it
[03]Because
TokensThe model reads the whole payload, markup included, before it reaches a sentence
RoomMarkup fills the context window that the answer needed
AccuracyThe less of the input is content, the easier it is to miss the one line that mattered
What Onto removes
[04]What happens
Only the contentThe first
<main> — or <article>, or the whole body. Everything outside it goes.RemovedScripts (and the hydration data inside them), styles, noscript, iframes, SVG, nav, header, footer, meta, link
Dropped in conversionClass names and other attributes — they don't survive the move to Markdown
KeptHeadings, paragraphs, links, lists, tables, and your JSON-LD as a block at the end
Your own numbers
[05]Every /v1/read response measures the page it read, in stats: raw_html_size_kb, markdown_size_kb and reduction_percent. Log them and you have the real tax across your own traffic — sizes are counted in characters, and the Markdown includes Onto's short header.