Yeah I hit the same wall with Claude, tbh not sure I ever fully fixed it 😅 The only thing that helped a little was giving it actual example sentences in the tone I wanted instead of just telling it to be warmer. Still feels like I'm fighting it more than ChatGPT though. Did giving examples do anything for you, or does yours ignore that too?

@Zayn Rahman good thing you asked, because I owe you a correction. I said "a few bucks for 10k" earlier and that was me going from memory — I actually went and checked. Twilio's line type intelligence is $0.008 per number, so roughly $8 per 1,000 and $80 per 10,000. Not a few bucks, my bad 😅 Still cheap next to burning sends on landlines, but worth knowing before you run a big list. Two things that soften it: duplicates only get looked up once, and you can test on the first 10 numbers before committing to the whole thing. And since I ended up building the little script anyway, just take it: https://gist.github.com/Nakoliss/85ebad6d826ea1e160bfe06e469bc56d — you drop in a CSV, it hands back a mobiles-only one (plus a full results file so nothing disappears quietly), and it prints the cost estimate and asks you to confirm before it spends anything. Free, do whatever you want with it. lmk if it breaks 🙂

@Max Stam You're right that you can't tell from the number itself in the US, there is no mobile prefix like you have in Australia. That's exactly why the lookup API route works there, it checks the carrier and ported number data instead of guessing from the format. I never used PhoneValidator so I can't compare honestly, but it looks like the same idea. One thing I learned building mine, cache the results locally. If you re run overlapping lists you end up paying twice for numbers you already checked.