Improving Clay Models' Generate Prompt Accuracy and Efficiency

Hello, Building up on Jordan C.'s for Clay Models https://clayrunhq.slack.com/archives/C025XGWSYTX/p1755019706015319 I tested out the generate prompting feature and it automatically recommends clay models which work great. But I noticed that when I input my prompt for example: "Take this /job title, and normalize it so it could be used in an email. Cut out unnecessary elements so it doesn’t sound like it’s been pulled from a data source, but rather that you know the company. Cut out elements like corp or other things that sound too formal Remove any punctuations, special characters, and ensure proper spacing is done for the formatting and capitalize the first letter of each word." Just as an example as that's a templated run. I hit "generate prompt" and it adds random columns in that I didn't ask it to put. This is fine when it works, but when there are errors or it involves a column I dont want to include, the result turns out farther from what I expect. This was a frequent issue I had with my workbook yesterday and it still included columns despite my sample prompt including "use only /linkedinprofile and /company domain/ as references. It consumes a lot of time having to fix and read through it.