Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.
The cold email is where all your upstream work either pays off or evaporates, and the bar is brutal: average cold email reply rates sit in the low single digits, while short, genuinely researched emails to well-chosen targets can pull 8-15% - a difference of 5-10x that comes almost entirely from relevance, not eloquence. The structure that earns replies is boringly simple: an observation that proves you did the work, a reason to talk that connects it to something you can credibly help with, and an ask so small it is easy to say yes to. Everything else - the pitch, the feature list, the calendar link, the 'hope this finds you well' - is weight that sinks it. AI enters here as your critic, not your ghostwriter: an AI-drafted email sounds like every other AI-drafted email in the inbox, but an AI run against your draft with a strict checklist catches the bloat and the generic phrases you can no longer see. This task is only the draft; you send in the next one - separating writing from sending is deliberate, because fear of sending corrupts drafting.
Here is the actual four-sentence email for the running example, written out in full. Subject: 'onboarding 3 new support hires'. Body: 'Hi Priya - saw HelpDeskCo has three support roles open against a team of about ten, which usually means a brutal quarter of repeating the same onboarding explanations live. Your post about cutting repeat tickets suggests you're already thinking about answering things once instead of many times - Loom does exactly that for internal explanations, and support teams like Intercom's use it heavily for exactly this. Worth a look for your onboarding quarter? If useful, I'll send a 3-minute example of a support-onboarding Loom - no call needed.' Sentence one: the observation, with the checkable fact. Sentence two: the bridge from her stated problem to the product, with a credibility anchor. Sentence three: the tiny ask. Sentence four: an even smaller escape hatch that costs her nothing. Roughly 85 words, zero adjectives about the product, no calendar link, no 'I'd love to pick your brain.'
Now the AI critique pass. Paste your draft into Claude or ChatGPT with: 'Critique this cold email against this checklist, scoring each item pass/fail with a one-line reason: (1) The first sentence contains a fact specific to this exact person or company. (2) Under 100 words total. (3) The ask can be answered in one line without opening a calendar. (4) Zero generic phrases - flag anything like: hope this finds you well, quick question, touch base, I'd love to, circle back, revolutionary, seamless. (5) It reads like one specific human wrote it - flag any sentence that could appear in a thousand other emails. Do NOT rewrite it; only critique.' The 'do not rewrite' clause matters: accept the AI's diagnosis, then fix it in your own words. An email that fails item 5 after an AI rewrite is worse than your clumsy original, because clumsy-specific beats polished-generic in a 2026 inbox every single time.