Use Claude like an AI Expert

Ruben teaches Claude to rooms of senior executives, and he says the first thing that trips them up is choosing between models and effort levels. He doesn’t think anyone should have to make that choice. His own setup is short, with memory so Claude knows his work and connectors so it can draft email replies from his meeting notes. Kieran called connecting email one of the biggest time savers.

Try this: Pick the one task you repeat most, and build your setup around that first.

Ruben says many people feel behind because they aren’t “using agents,” when Claude and ChatGPT already complete tasks for them, including ones that run on a schedule. His view is that the technology should fade into the background, and choices like which model to use should eventually disappear on their own. In his words, “Technology must become invisible.”

Try this: Schedule one recurring task, and see how far that single step takes you.

Heavy style files and stacks of skills can lock Claude into one predictable voice. Ruben says writing should keep evolving with your experience, and rigid files hold it in place. He also notes that today’s models handle tasks well with less context than they needed earlier this year, so a setup built for older models can now weigh you down.

Try this: Keep any saved context about yourself short, and add to it only when Claude gets something about you wrong.

Ruben writes every word of his newsletter and teaches his team to write from their stream of thoughts. His view is that style is valuable because it’s distinctive, and content starts to sound alike once everyone drafts with the same tools. He also points to research suggesting that language common in AI output is showing up in how people speak, which led Kieran to wonder whether the AI was learning his style or he was picking up the AI’s. It’s a question worth asking about your own writing.

Try this: Compare something you wrote before using AI with something you wrote recently, and notice which sounds more like you.

Kieran tells people early in their careers that writing is thinking, and Ruben agrees. You can often tell when someone wrote about something they never lived through, because the thinking behind the words is the real work. That makes your own point of view the part Claude can’t supply for you.

Try this: Before you ask Claude for a draft, write three sentences on what you believe about the topic.

Human Taste And Connection Still Lead
Kipp’s takeaway was that the companies that win will pair invisible technology with strong human taste and judgment. Ruben makes the same point about work like customer success, where clients want to win with someone they enjoy working with. Great marketing is social too, and it depends on people who show up and connect.

Try this: Use the time Claude saves for the part that needs a person, such as a conversation with a customer, a colleague, or a partner.

Use this once when you start, or any time your role or priorities change.

Remember the following about me. I'm a [job title] at [company], working on [what you work on]. My priorities right now are [priorities]. When I give you a big task and my request is ambiguous, ask me a clarifying question first. Don't invent data or sources, and tell me when you're unsure. Keep answers concise, and flag anything that needs my judgment before you finalize it.

Why it works: It’s short on purpose. A few lines on your role and how you like to work are enough to start from, and you can add to it only when Claude gets something about you wrong. The working rules target two common frustrations, which are guesses presented as facts and answers to questions you didn’t quite ask.

Use this when you need to reply to someone and the history matters. If your email isn’t connected, paste the thread into the prompt.

My goal: draft a reply to [name] about [topic] that [what the reply should accomplish].

Context: [paste the email thread, or ask Claude to find the most recent thread with [name] about [topic] if your email is connected]. Here are my notes from [meeting or conversation]: [paste your notes, or name the meeting if your notes tool is connected]. Also take into account [any earlier agreements, deadlines, or constraints].

Before you write, ask me about anything that's missing or unclear, and wait for my answers. If nothing is missing, go straight to the draft. Write the reply in my voice, under [number] words, and flag anything you're unsure about. Save it as a draft and don't send it. If you can't save drafts, show it to me here.

Why it works: Starting with the goal and the context means the draft reflects the history behind the email. The clarifying question line is where missing details, like an agreed price or date, come up before they show up as guesses. Saving it as a draft keeps you as the person who decides what gets sent.

Use this when you have ad results and want to know what to test next.

Here are the ads I ran and their results: [paste ad copy, headlines, and numbers like spend, clicks, and conversions]. My goal for these ads was [goal].

First, tell me whether this is enough data to draw conclusions, and what's missing if not. Then identify the patterns that separate my stronger ads from my weaker ones, and give me three specific things to test next, each with a reason tied to the data I gave you. Don't make claims the numbers don't support.

Why it works: It asks Claude to judge whether the data is enough before analyzing anything, which keeps small samples from producing confident sounding conclusions. Requiring each test idea to point to your numbers keeps the output grounded in your results and gives you something to challenge.

Use this to see where your market sounds the same and where you could stand apart.

I'm positioning [product] for [audience]. Here is how [three to five competitors] describe themselves: [paste their homepage headlines and key messaging, or name them if you have web search turned on].

Map the claims they all make, the claims only one of them makes, and what none of them say. Then give me three possible positioning directions that would set me apart, each with the evidence from the messaging that supports it. Mark anything you couldn't verify.

Why it works: Sorting claims into “everyone says,” “only one says,” and “no one says” shows you quickly where the market has converged. Asking for evidence behind each direction lets you check Claude’s reasoning against what competitors actually wrote, and the last line stops it from filling gaps with guesses.

Use this when you need fresh angles instead of the first obvious answer. Start in an incognito chat, or turn memory off, so Claude works from your brief and not from what it already knows about you.

I need campaign ideas for [product or offer], aimed at [audience], with a goal of [goal] and a constraint of [budget, timeline, or channel].

Before you suggest anything, ask me up to three questions that would change your ideas. After I answer, give me five campaign angles that are different from one another. For each, write one sentence on the idea, one on why it could work for this audience, and one on the biggest risk.

Why it works: Asking questions first prevents generic ideas built on assumptions. Five different angles, each with a risk attached, are easier to compare and give you something to push back on.

Use this when you want a regular update without asking for it each time. Scheduled tasks are available on paid Claude plans, and you’ll find them under “Scheduled” in the left sidebar.

Create a scheduled task that runs every [Monday at 8 a.m.]. Each time, check [three competitors] for changes in their messaging, new launches, or announcements. Give me a short summary of what changed and why it might matter for how I position [product]. If nothing changed, tell me so.

Why it works: It’s one task with a clear scope, and a weekly cadence gives you something useful without becoming noise. The last line keeps Claude from inventing updates on a quiet week.