ClaudePrompting mistakesLevel 1: Basic prompts

Why Does Claude Answer the Story Instead of Your Question?

Usually because the story is specific and the question isn't. If your message is five detailed sentences ending in "thoughts?", the story is the clearest thing Claude has to respond to. Keep the useful context, cut the rest, and end with a request that names what you want back.

Prompt to copy 
Situation:
- [fact that changes the answer]
- [fact that changes the answer]
- [deadline or limit]

What I've already tried: [one line, or "nothing yet"]

What I need from you: [a named deliverable, such as "a three-step plan" or "a yes or no with your reason"], for [who will use it].

Don't restate the situation. Start with the [deliverable].
  1. 1Vendor missed two datesfact
  2. 2Launch is on March 3fact
  3. 3Legal hasn't approvedfact
  4. 4Three-step plan to decidethe ask: change vendors or not, by Friday
Rule of thumb, not an Anthropic chart, shown with this page's example: the facts that matter, then one clear request.

Try it now

  1. 1Open your last long message to Claude and underline the sentence that says what you wanted back. If there's no such sentence, write one.
  2. 2Delete any line that wouldn't change the answer, and turn the rest into the short list in the prompt above.
  3. 3Send it in a new chat. You'll know it worked when the reply opens with your deliverable, with no summary of your situation.

Use it when

When an answer discusses your situation well but gives you nothing to do, or when a draft message opens with "So, some background." It's also worth checking when your message has more about how you feel about the problem than about what you need.

Skip it when

When you do want to think out loud. Saying "I'm not asking for a fix, help me work out what the real problem is" is a clear request. Short questions need no restructuring at all.

Why it works

The position of the question isn't really the problem. Anthropic's docs recommend the query at the end for long inputs, and its Fable 5 guide offers a template in which the context comes first and the request comes last. What matters is that a clear request exists. Anthropic's docs ask you to be specific about the output, and they suggest a test: would a colleague with minimal context know what to do? A message ending in "anyway, any ideas?" fails that test, since it could call for sympathy, a diagnosis or a plan. The Fable 5 guide even offers a sample instruction for this situation, telling the model that when someone is describing a problem or thinking out loud, the deliverable is its assessment. In a chat, you can settle that yourself by naming what you want.

Example

A message describes a stalled project, a frustrated team and a deadline, then ends with "anyway, thoughts?" Nothing says what to produce, so Claude may offer sympathy and general advice. Rewrite it as three facts: the vendor missed two dates, the launch is on March 3, and legal hasn't approved the contract. Then ask: "Give me a three-step plan to decide by Friday whether to change vendors." Now Claude has a job to do.

Going further

Earlier advice often said to put the question at the very top. Anthropic's docs recommend the reverse for large inputs of 20,000 tokens or more: documents first, query last. In shorter prompts, either order can work as long as the ask is clear. Rule of thumb, not a vendor figure: for a long paste, mention the goal in one line at the top too, and put the full request at the end. If you can't write the ask, you may not have decided what you want. Ask Claude to interview you instead, a method Anthropic's Claude Code guide recommends for unclear work.

Common mistakes

  • Ending with "thoughts?" or "any ideas?", which leaves the deliverable for Claude to guess.
  • Counting on emotional weight to show what matters. State the priority directly, such as "the deadline matters more than the cost."
  • Stacking caveats until the request sounds optional. Say what you need and then list real limits.
  • Moving your question to the top of a very long paste because you read old advice. Anthropic's docs recommend the bottom.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Mistake Why Does Claude Give Generic Answers? Technique Where Do Instructions Go in a Long Claude Prompt? Best practice How Do You Tell Claude the Result You Want? Technique How Do You Get Claude to Cut a Draft to Length?

Beyond the course

Glossary Hallucination Pricing guide Claude pricing, plans and models