The biggest mistake new AI users make is starting a new chat every time something is not right. Here is what to do instead. Imagine hiring a consultant. They deliver a first draft. It is good, not great. You would not fire them and hire someone new. You would give them feedback and ask for a second round. That is exactly how Claude works, and almost no one uses it that way. When Claude sends you a response, treat it as a first draft. Reply in the same conversation with one of the following: Make this shorter. Change the tone to something more direct. Keep the structure but rewrite the opening. I like the second point. Expand on that and remove the first one. This is for a board audience, not a team one. Adjust. Give me two alternative versions. Each reply refines the output. Claude remembers the full conversation, so you do not need to repeat context. Three habits to build: First, name what is off. Not just "make it better." Tell Claude what specifically is not working. Too long. Too corporate. Too generic. Missing the point about pricing. The more precise your feedback, the faster you converge on something useful. Second, ask for alternatives. If a draft is almost right but you cannot name what is wrong, ask for three different versions in different styles. Pick the one closest to what you want, then refine from there. Third, know when to start fresh. If you are four or five exchanges in and the conversation has drifted badly, it is sometimes faster to start a new chat with a sharper initial prompt. You have learned what works. Use it. People who iterate well get five times the value out of AI compared to people who do not. Same tool, and same prompts, but completely different results!

There is a simple structure that turns an average prompt into a great one. Once you see it, you cannot unsee it. Most prompting advice online is overcomplicated. Here is the version that actually works. Use these five parts, in roughly this order. ✅ One. Role. Tell Claude who to be. Example: You are an experienced HR director reviewing a job description. This is not a trick. When you give Claude a role, it draws on the part of its training that reflects that perspective. A marketing consultant, a tax accountant, a plain-spoken editor, a sceptical investor. Pick the lens you need. ✅ Two. Context. Give Claude the background. Who is involved, what has happened, what the stakes are. The more specific, the better. If you are working on something confidential, say so and Claude will adjust its tone accordingly. ✅ Three. Task. State clearly what you want. Write, review, summarise, compare, brainstorm, restructure. One verb is usually enough. ✅ Four. Constraints. Length, tone, format, audience, what to avoid. These are where people leave value on the table. Without constraints, Claude picks reasonable defaults. With them, you get exactly what you need. ✅ Five. Examples or format. If you have a sample of good work, paste it in. If you want the output in a specific structure, say so. Bullet points. Table. Email. Three options to choose from. Here is the formula applied to a real task: You are a senior events operations manager. I am preparing for a one-to-one with a junior team member who missed a key deadline last week. This is the first time it has happened. I want to address it directly without damaging trust. Write me three opening lines I could use. Keep each one under 30 words. Tone should be warm but serious. Compare that to: How do I talk to my employee about missing a deadline. Same topic. Different results. Use the formula! This community is designed for knowledge sharing; share an effective prompt you've developed in the comments below.

Most people open Claude, type 'hi,' and get nothing useful back. Here is how to get something you can actually use in under five minutes. Open claude.ai in your browser or the Claude app on your phone. Sign in. You will see a single text box. That is the whole interface. Now pick a real task. Something you were going to do today anyway. For this walkthrough, we will use a common one: writing a polite follow-up email to a client who has gone quiet. Step one. Do not type the task. Type the situation. Bad prompt: Write a follow-up email. Good prompt: I sent a proposal to a potential client two weeks ago. They replied once saying they were interested and would come back to me. I have not heard from them since. I want to write a warm, non-pushy follow-up that gently asks where things stand. Keep it under 120 words. Notice the difference. The second version gives Claude the context, the goal, the tone, and the constraint. Step two. Read what Claude sends back. It will almost certainly be close but not perfect. Step three. Do not start a new chat. Reply in the same conversation with what you want changed. For example: Good, but make the opening less formal and remove the line about next steps. Claude will revise. You can keep going until the draft is right. This back-and-forth is where most of the value sits. The first draft is a starting point, not a finished product. Step four. Copy the final version into your email client. Send. You have now done in five minutes what might have taken you twenty, and the email is arguably better than one you would have written cold. The pattern is always the same. Context, request, iterate. Context, request, iterate. Every task we cover for the next month uses some version of this loop.

Claude, ChatGPT, and Gemini are all built on something called a large language model, or LLM. The name is off-putting. The idea is simple. An LLM is a system that has read an enormous amount of text and learned the patterns in how humans write. When you type a question or instruction, it predicts what a thoughtful, well-written response would look like, one word at a time. That is worth rereading. It predicts. It does not retrieve facts from a database. It does not look things up unless you give it that ability. It produces the most likely useful response based on the patterns it has learned. Three implications follow from this. First, LLMs are extraordinary at tasks that involve language. Writing, summarising, explaining, translating, structuring ideas, drafting, editing. If your task is mostly words, Claude will help. Second, LLMs can make mistakes on facts. They can state something incorrect with total confidence. The word for this is hallucination. It does not mean the tool is broken. It means you should verify anything specific, such as dates, figures, names, or citations. Third, the quality of what you get out depends heavily on what you put in. A vague question produces a vague answer. A clear, specific request produces a clear, specific response. This is the single most important skill we will teach you. One useful mental model: treat Claude like a brilliant new hire on their first day. They are smart, fast, and willing. They know a lot about the world. But they do not know your company, your clients, your preferences, or your context unless you tell them. The more context you provide, the better they perform. In the next post, we'll walk through your first real conversation with Claude, step by step. What questions do you have about LLMs after reading this post? Let us know in the comments and we'll answer all of them!

If you can type an email, you can use AI well. The hard part is knowing where to start. Welcome to Vernal. We built this community for one reason: AI is the most useful working tool released in our lifetime, and almost no one was taught how to use it. That gap is what we close. Over the next thirty days, we will walk through the skills that make AI genuinely useful on a Tuesday afternoon when you have six tabs open and a deadline at four. We'll showcase practical skills you can apply to your next email, your next meeting, your next report. We teach using Claude, made by Anthropic. We use it because it writes well, reasons carefully, and handles nuance better than most tools on the market. If you are already using ChatGPT, Gemini, or something else, most of what we cover will transfer. Where Claude has a specific strength or feature, we will point it out. Here is what you can expect across the month: Week one covers the foundations. What AI is, what it is not, and how to have your first useful conversation with it. Week two covers writing and communication. Emails, reports, meeting notes, and tone. Week three covers thinking and problem solving. Using AI as a research partner, a sounding board, and a second pair of eyes. Week four covers workflows. How to build small systems that save hours a week. There are two rules for this community. First, ask questions. Every question is a good one. If something is unclear, someone else has the same question and is quieter about it. Second, try things. Reading about AI teaches you nothing. Using it for twenty minutes teaches you more than any course. What are you excited to use AI for in your life? Is there a particular issue you'd like to use AI to solve? Let us know in the comments, and we'll share any tips we think might help!