I'm not finding much in the way of user agent strings for the new "AI" browsers (Perplexity Comet, GPT Atlas, Opera Neon, etc). If there's any I'm missing, feel free to add to my list. I'm trying avoid downloading and (shudder) installing any of these just to figure out the UA string they use.

Essentially, I want to deny these browsers use of a few applications I develop (yes, this includes Tesseract).

However, I want to avoid any false positives, so I'm trying to be as specific as I can in the filter.

For now, I'm not bothering with Edge/Chrome despite them now advertising themselves as "AI Browsers"; I'm just denying the ones made directly by the AI companies.

all 6 comments

sorted by: hot top controversial new old
[–] 4 points 11 months ago (1 child)

Unfortunately, a lot of browsers now completely obfuscate themselves and don't put anything unique in the user agent string. They intentionally don't want to be detected at all beyond "I'm a modern browser".

  • source
  • hideshow 1 child comment
  • [–] [S] 1 point 11 months ago

    Yeah. Perhaps that's why I'm not seeing them on any of the User agent lists, though I assumed it's because they're all new and just haven't been documented yet. (Hoping it's the latter).

    I may have to break down and install them in a sacrificial VM, check, and submit my findings.

  • source
  • parent
  • [–] 0 points 11 months ago*

    I can say that I've used a plugin for WordPress called Dark Visitors which seems to track and report on bot crawlers, including LLMs. Not sure how accurate their info is, but our marketing dept seems to believe it so 🤷‍♂️

    Maybe looking into what Dark Visitors does would give you pointers on how you might detect them.

  • source
  • [–] 1 point 11 months ago (1 child)

    Just install the browsers and test them on your website. That's the best way to be sure it works.

  • source
  • hideshow 1 child comment
  • [–] 0 points 10 months ago

    You should look into something like Anubis: https://github.com/TecharoHQ/anubis

  • source