Is there any reliable ai detector that's free to use?

Your take is right. Detectors flop, stacking helps a bit, humanizers evade, and teh human pass still matters.

Try a simpler path, provenance checks. Do this:

  1. Pull three claims with dates or numbers. Verfiy each with a primary source. If 2 of 3 fail, flag it.
  2. Quote search two 8 to 12 word spans. Zero hits plus generic tone, raise risk.
  3. Check for named sources and data tables. Thin sourcing, flag it.
  4. Post a tiny change request within 30 mins, e.g., add a missing figure. No response, treat as suspect. Fast, free, low fuss.