Why AI-Powered Outreach Emails Often Feel Inauthentic and How Top Reps Use AI Differently

Ozan D. · 2026-08-05T01:58:09.340Z

anyone else noticing that 'AI-powered outreach' has basically become the new 'sent from my iPhone' — the recipient can immediately tell? the reps who are actually crushing it with AI aren't using it for the writing part at all. they're using it for research — finding the right person, understanding their stack, timing the outreach to a trigger event. the writing itself is still human. feels like most sales AI tools optimized for the wrong step.

7 comments

· Sorted by Oldest

jonatan
·
·
largely agree, though i'd put the reason somewhere else. the models don't write badly, they write fine. the tell isn't the prose, it's that the message has no cost behind it. a human written note implied someone spent twenty minutes on you, and that implication was most of what a cold email was ever buying. once the words are free the same words mean something different. the research versus writing split is the right diagnosis but i think the boundary sits elsewhere. the two steps usually run separately and never share a view of what's worth mentioning. the model surfaces twelve facts, someone has to pick one, and the picking is the judgement nobody has automated, mostly because finding is measurable and choosing isn't. the tool i'd actually want is one willing to come back with nothing when there's no reason to reach out. everything on the market is built to always produce a message, which guarantees a weak one every time the trigger isn't really there.
Sam
·
·
as someone looking to build near this space, these are some great insights, thanks for sharing, great stuff!
Mike B.
·
·
Not going to lie. One of the things that has actually helped me cut through the noise is being wildly casual and making my messaging so obviously human it hurts (I spell stuff wrong on purpose sometimes). Being the one message in someone's inbox that is well-researched AND human has turned into a competitive advantage. Now my girl Claudia is trained to talk like me too to save more time.
Michael M.
·
(edited)·
anyone have good workflows or prompts they are willing to share that work well for the research bit? I've been wrestling with Claude and ChatGPT all day and am only getting extremely surface-level results and tenuous connections to transition to my value prop.
Shreyash C.
·
·
Agree with all of this. the tell isn't even the writing quality, it's that the message could have gone to 500 people. The research half is where the leverage is and it's also the boring half nobody wants to build. Finding who actually has a reason to hear from you this week, then writing to that reason. We do the research layer at bitscale so i'm biased, but the tools that optimized for drafting solved the easy problem.
✅1
Ozan D.
·
·
the workflow that actually works: feed the prospect's last 3 linkedin posts + company news into claude, ask it to find the one thing they care about this quarter that connects to what you sell. then write the outreach around that specific thing. the research step takes 2 min but it's what makes the message unsendable to anyone else
🙏1
🙌🏻1
Michael M.
·
·
Thank you!!

jonatan
·
·
largely agree, though i'd put the reason somewhere else. the models don't write badly, they write fine. the tell isn't the prose, it's that the message has no cost behind it. a human written note implied someone spent twenty minutes on you, and that implication was most of what a cold email was ever buying. once the words are free the same words mean something different. the research versus writing split is the right diagnosis but i think the boundary sits elsewhere. the two steps usually run separately and never share a view of what's worth mentioning. the model surfaces twelve facts, someone has to pick one, and the picking is the judgement nobody has automated, mostly because finding is measurable and choosing isn't. the tool i'd actually want is one willing to come back with nothing when there's no reason to reach out. everything on the market is built to always produce a message, which guarantees a weak one every time the trigger isn't really there.
Sam
·
·
as someone looking to build near this space, these are some great insights, thanks for sharing, great stuff!
Mike B.
·
·
Not going to lie. One of the things that has actually helped me cut through the noise is being wildly casual and making my messaging so obviously human it hurts (I spell stuff wrong on purpose sometimes). Being the one message in someone's inbox that is well-researched AND human has turned into a competitive advantage. Now my girl Claudia is trained to talk like me too to save more time.
Michael M.
·
(edited)·
anyone have good workflows or prompts they are willing to share that work well for the research bit? I've been wrestling with Claude and ChatGPT all day and am only getting extremely surface-level results and tenuous connections to transition to my value prop.
Shreyash C.
·
·
Agree with all of this. the tell isn't even the writing quality, it's that the message could have gone to 500 people. The research half is where the leverage is and it's also the boring half nobody wants to build. Finding who actually has a reason to hear from you this week, then writing to that reason. We do the research layer at bitscale so i'm biased, but the tools that optimized for drafting solved the easy problem.
✅1
Ozan D.
·
·
the workflow that actually works: feed the prospect's last 3 linkedin posts + company news into claude, ask it to find the one thing they care about this quarter that connects to what you sell. then write the outreach around that specific thing. the research step takes 2 min but it's what makes the message unsendable to anyone else
🙏1
🙌🏻1
Michael M.
·
·
Thank you!!