We rewrote Stela's default outreach prompt to drop the AI tells (corporate jargon, em-dashes, summary subject lines, fake markdown links) and added context tooltips on every draft's engagement stats so you know what "5% engagement" actually means for that platform. Two changes to how outreach drafts read and how you read them: Default drafts that sound like a DM, not a pitch — If you haven't customized your outreach prompt, your drafts now use Stela's rewritten default — a DM-not-pitch voice anchor with explicit bans on em-dashes, corporate jargon ("thrilled," "reach out," "synergy," etc.), markdown links bleeding into plain-text email, and summary-of-email subject lines. The prompt also tells the model it's OK to skip the brand-fit pitch when the connection would be a stretch — better a short authentic note than a forced one. Drafts that fail the lint regenerate once, automatically — When a draft comes back with the patterns we banned (an em-dash, a banned word, a summary subject), Stela regenerates it once with a hint about what to fix. If the second pass still misses, we save what came back rather than retry forever — cost-bounded, but most issues catch on the first retry. Engagement tooltips on every draft card — Every draft card shows the creator's followers and engagement rate. Hover the info icon next to those stats and you'll see a tooltip explaining the formula (Instagram is follower-based; YouTube and TikTok are view-based), the platform benchmark, and where this creator falls relative to it. The same explanation used to live in small grey text below the card — easy to miss. Now it's right where you're looking.