A/B testing hooks against 100 personas

The post does not test that. It reports near-real-time A/B testing of TikTok and Instagram hooks against 100 rich personas, replacing a panel that cost $1 to $5 per run, with no methodology and no comparison against how those hooks performed on real audiences.

Undocumented. The post calls them rich personas and says nothing about how they were written, sourced, or validated. If they are hand-authored character sketches, agreement between them may just reflect the author's assumptions rather than distinct audience segments.

An estimate from the author. He says it now costs just a few cents and is real time, but gives no token counts, no latency figures, and no per-run breakdown, so the comparison with the old $1 to $5 panel is directional at best.

Ranking is the defensible use. The model returns a scored decision over candidate hooks against a persona set, which can cheaply reorder your shortlist. The post offers no evidence that the winning hook survives contact with a real feed, so treat it as a filter rather than a predictor.

Vybhav Ramachandran, who says he is co-founder of an affiliate agency and is building what he calls the world's best affiliate agency. The persona panel is his own experiment on top of Jev rather than a product feature, and no repository or tool is linked from the post.

The post does not say. He describes only a panel that cost one to five dollars per run and was slow at this scale, without stating whether it was human respondents, another model, or a paid service. That omission matters because the whole comparison is between Jev and whatever that baseline actually was.

He calls it a new ChatGPT moment and a brand new world, meaning the cost change reframes what is economic to do. The argument is about economics, not accuracy: a persona panel that was ROI negative at one to five dollars a run becomes cheap enough to run routinely. No accuracy claim about the personas is made.