Testing my own AI tool against competitors taught me more than building it did

Building letsflw (all-in-one AI writing tools), I wanted real evidence our stuff actually works, not just claims. So I wrote original test passages never used anywhere else, no scraped content and ran them through our paraphraser and humanizer alongside Quillbot, StealthWriter, and a couple others. Same input, unedited output, no cherry-picking.

What I didn't expect: how bad some of the competitor failures were.

One flipped a sentence's actual meaning turned "would hurt a funded company" into "will actually harm" it, changing a hypothetical into a certainty.

Full test here: https://letsflw.com/blog/quillbot-alternative-paraphrase-test?utm_source=indiehackers&utm_medium=community&utm_campaign=quillbot_comparison

Another one, on a contract-clause test, invented a legal condition that
wasn't in the original text at all the kind of error that would give
someone genuinely wrong information about their own liability exposure.

Full test here: https://letsflw.com/blog/ai-humanizer-comparison-test?utm_source=indiehackers&utm_medium=community&utm_campaign=humanizer_comparison

Our tool held up across all the rounds, with one small cosmetic slip I
included anyway — because if I only show the other guy's mistakes, it's
not a real comparison, it's an ad.

Curious if anyone else here has run a real head-to-head test on their own
product vs competitors. Feels like something more founders should do
publicly instead of just claiming "we're better."