Stylometry tools may be useful if you already have a small candidate pool of suspected aliases. They produce too many false positives to be useful for blind cross-linking of accounts. Once or twice somebody has done stylometric analysis of HN accounts and I've looked at the results for my accounts. Even though I don't try to obscure style across accounts, stylometry didn't match my actual accounts with each other. My top matches were for accounts controlled by other people.

I specifically write with different perspectives, tones, and opinions on different sites in a probably vain attempt to mitigate this.

For example, on YouTube I use twitch slang, and on Reddit I use TikTok slang, and on TikTok I use reddit slang. On hackernews a use a slightly whimsical pedantically-infused undergrad tone.

You really care about this and use most privacy invasive platforms at same time? Sounds like interesting acrobatics to me.

Using stats this is called stylometry and I agree this will probably be easier at scale now. You can also match posting windows, pull additional features from database dumps/hacks.

Fun post applying it to HN, not sure if the site is still live: https://news.ycombinator.com/item?id=33755016

Then people will start using browser extensions that automatically "fuzz" your writing style randomly. That is, if chasing anonymity is someone's true goal.