Why your first answers were generic, and the five moves that fix it

You tried it. You opened Claude, wrote a real prompt with the three-part shape, and the result was — fine. Not bad. But a little generic, a little too long, with a confident claim about something you are not sure is true. You are not imagining things. This is where almost every new Claude user gets stuck: the difference between "it worked" and "it was genuinely useful" is usually a small adjustment you do not yet know to make.

This lesson is the adjustment kit. It covers the five failure modes you will run into most often, a framework for thinking about your collaboration with Claude, and a simple way to test Claude on the tasks you actually care about.

The five most common problems — and the move that fixes each

Every response Claude gives you is a data point. When the output misses, diagnose why before you try again.

| Problem | What is happening | The move | |---|---|---| | Too generic | Your prompt did not include enough context about your specific situation. | Add details about audience, role, and constraints. Instead of "Write an email about the project delay," try "Write an email to our enterprise client explaining the integration will be delayed by two weeks. They have been patient but this is the second delay. Keep it professional but apologetic." | | Wrong length | Claude is guessing at the right length. | Be explicit. "Give me a two-paragraph summary." "Keep this under 100 words." "I want a comprehensive analysis — length is not a concern." | | Wrong format | Claude understood what you want but not how you want it presented. | Show, do not just tell. Provide an example, or describe the structure. "Use bullet points with bold headers for each section." | | Confident-sounding but wrong | Claude occasionally generates plausible information that turns out to be incorrect, especially on niche topics or very specific facts. | For high-stakes work, verify key facts independently. Ask Claude to cite sources or indicate confidence level. Turn on web search to ground the answer in current information. | | Wrong tone | Claude defaulted to helpful and professional, which may not be what you need. | Describe the tone in plain language. "Make this more conversational." "This should sound authoritative and formal." Or provide an example of writing in the style you want. |

Notice that all five moves are additive. You are not fighting Claude — you are giving it the missing piece of the brief. That is the whole troubleshooting mindset: when the output is off, assume the prompt was the gap, not the model.

The iteration mindset

The deeper shift that separates great Claude users from frustrated ones is how they relate to the first response.

Treat first drafts as starting points. Read what Claude produced, note what is working and what is not, and then refine. The second message is where the value usually lives.

Give specific feedback. "Make it shorter" is fine. "Cut the first two paragraphs and make the conclusion more action-oriented" is much better. Specificity is the difference between a guided revision and a lottery.

Know when to start fresh. If a conversation has drifted off-course, sometimes the fastest path is a new chat with a cleaner prompt. Do not grind on a thread that has accumulated confusion — Claude remembers everything in the current chat, including your early mistakes.

The four competencies that make AI collaboration work

When you zoom out from individual prompts, working well with AI comes down to four core skills. Anthropic did not invent this — it comes from academic research on AI fluency developed by Professor Rick Dakan (Ringling College of Art and Design) and Professor Joseph Feller (University College Cork). Their framework identifies four competencies, often called the 4 Ds:

Delegation. Deciding what work should be done by humans, what should be done by AI, and how to split the task between them. Delegation starts with understanding your goals and Claude's capabilities well enough to make a deliberate choice — not a default.

Description. Effectively communicating with Claude: clearly defining outputs, guiding the process, and specifying behaviours you want. Every skill you picked up in the last lesson — the three-part prompt, file uploads, style presets — is Description in practice.

Discernment. Thoughtfully evaluating Claude's outputs, processes, and interactions. Is the quality good enough? Is the reasoning sound? Is the answer accurate? Where does it need improvement? Discernment is what turns you from a passive recipient into an active editor.

Diligence. Using Claude responsibly. Making thoughtful choices about when to rely on AI, maintaining transparency about its role in your work, and taking accountability for the final result.

You have been doing these already. The three-part prompt from the last lesson is Description. The five troubleshooting fixes above are Description and Discernment. Choosing whether to use Claude at all for a given task is Delegation. Staying honest about AI's role when you send the work on is Diligence.

The four move together. Get good at all of them and you stop asking "is Claude any good at this?" and start asking "given what I know about Claude and this task, how do I set this up?"

Run a simple eval on the work you actually do

Here is the question that matters more than any other: "Is Claude actually good at this particular task — for me, for my work?"

You will never know until you test it systematically. Not with synthetic examples from a prompting blog, but with your real work.

Formal AI evaluations can get complicated. You do not need any of that. You need a lightweight eval loop you can run in an hour.

Step 1: Gather examples. Collect 5–10 examples of a task you do regularly. Emails you have written. Reports you have produced. Analyses you have done. The point is that you already know what good looks like, because you have the finished version.

Step 2: Write test prompts. Using the three-part shape, write prompts that would ask Claude to generate the same kind of output. Include the context you would naturally have when doing this work.

Step 3: Compare outputs. Run each prompt and compare Claude's response to your example. Ask yourself: Does Claude capture the key information? Is the tone right? What is missing? Where did Claude add value you did not expect?

Step 4: Refine your approach. Based on what you learn, adjust your prompts, add examples to show Claude what good looks like, or flag tasks where human review is essential. Keep the prompts that worked — those are templates you can reuse.

You are not trying to prove Claude is good or bad. You are trying to build calibrated confidence in how Claude performs on your specific work, so you know when to lean in and when to stay close. That is Discernment in action.

Before you move on

Pick one problem from the table at the top of this lesson that you have already run into. Write down which move you will try next time — not in theory, but the actual phrasing you will add to your next prompt.