Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.
Most people use AI like a stranger they re-introduce themselves to every single time: they open a blank chat, re-explain their service, re-explain their buyer, re-paste their tone — and get generic output because they gave generic context. The freelancers who get 3x leverage do the opposite: they set their service context once, in a place the AI reads on every message, so every draft starts already knowing who they are and who they serve. This is the habit the entire rest of the course rides on — your offer variants, your outreach, your delivery drafts all get better the moment the AI stops guessing. Setting it up takes fifteen minutes today and saves you the re-explaining tax on every task from here forward. Skip it and you'll fight the same blank-slate battle a hundred times.
Pick one assistant and set standing context. In ChatGPT, use 'Customize ChatGPT' (Settings > Personalization > Custom Instructions) or create a Project and put context in the project instructions. In Claude, create a Project and fill 'Project knowledge' / custom instructions. Either way, paste one standing block. Here is a real one to adapt: 'I am a freelance newsletter writer. My service: I write and send one monthly email newsletter for a client, matching their brand voice. My buyer: founders of Shopify skincare/supplement brands doing $20k-$100k/month, no marketing hire, who send late newsletters at 11pm and feel guilty. My tone: warm, plain-spoken, no hype, short sentences, no exclamation marks. When I ask for drafts, always give me three subject-line options and keep newsletters under 300 words unless I say otherwise. Ask me for the month's products/updates before drafting.' Save it. Now test it cold: open a fresh chat and type only 'Draft this month's newsletter — we launched a vitamin C serum and had a 15% off weekend.' Because the context is standing, the output should already be under 300 words, in the plain warm tone, with three subject lines, and no re-explaining. If it isn't, tighten the instructions and test again. That loop — set context, test cold, tighten — is the workbench.