Mistakes
Find out why Claude gives generic, agreeable or wrong answers, and how to fix the prompt and check the result.
Prompting mistakes
Fix generic answers, replies that miss your question, easy agreement, and prompts that bundle unrelated jobs.
4 techniques →-
Why Does Claude Answer the Story Instead of Your Question? Usually because the story is specific and the question isn't. Level 1: Basic prompts
- 1Vendor missed two datesfact
- 2Launch is on March 3fact
- 3Legal hasn't approvedfact
- 4Three-step plan to decidethe ask: change vendors or not, by Friday
-
How Do You Stop Claude From Just Agreeing With You? Don't show which answer you're hoping for. Level 2: ContextShows your preferenceHides your preference
- "Isn't this right?"
- "I lean A, thoughts?"
- Options A and B
- Fresh chat, criteria first
-
Should You Split a Big Claude Prompt Into Steps? Only when you need to check the result between steps, or when the steps must run in a fixed order. Level 1: Basic promptscost of a wrong first step →Separate chatscheck each on its ownNumbered list, then a stopcheck before the rest runsNumbered lista missing part is easy to spotOne promptClaude handles most steps internallysteps depend on each other →
-
Why Does Claude Give Generic Answers? Usually because the prompt left things for Claude to guess. Level 1: Basic prompts
- "Fix the login bug"no symptom, no place, no proof
- Add the symptomfails after session timeout
- Add where to looktoken refresh
- Add the proofa failing test, then the fix
Verification mistakes
Check summaries against the source and code by running it, ignore confident wording, and keep hard decisions for yourself.
4 techniques →-
How Do You Check a Claude Summary Against the Source? Have Claude tie each claim to a word-for-word quote, then look up the claims you'll rely on in the original. Level 2: Context
- 1Pull exact quotes
- 2Write from the quotes
- 3Match each claim to a quote
- 4Remove what has no quote
-
Which Decisions Shouldn't You Leave to Claude? Anything that's hard to reverse, affects other people, or depends on context Claude can't see. Level 2: Context
- Easy to undo, just youClaude drafts, you skim
- Affects other peopleClaude drafts, you decide
- Hard to reverseyou decide, and limit what tools can do
- High-risk areasa qualified person must review and can change it
-
How Do You Check Code Claude Wrote Before Shipping It? Run it, and ask for proof. Level 2: Context
- 1In the promptClaude decides when it's satisfied
- 2/goala separate model checks each turn
- 3Stop hookClaude Code enforces it
- 4CIchecks outside the session
-
Why Does Claude Sound Sure When It's Wrong? Claude writes in the same fluent tone whether a detail is right or wrong, so tone tells you nothing. Level 2: Context↑ cost if wrongFind a second sourcecostly, hard to checkAlways checkcostly, easy to checkTreat as a leadcheap, hard to checkQuick lookcheap, easy to checkhow easy to check →
Are you making these Claude mistakes?
Six common habits and what to do instead, from vague endings to trusting "all tests pass." Answers appear instantly, with no sign-up.
Two kinds of failure
Most problems with Claude happen either before the answer or after it. Before, the prompt leaves something out, and Claude fills the gap with a safe guess. After, a good-looking answer gets used without a check. This module covers both. Prompting mistakes shows how to spot and rewrite a weak prompt, and Verification shows what to check and how.
Why old advice can mislead
Some common tips no longer match Anthropic's docs. The docs recommend putting your question after a long document, not before it. They also say models handle most multistep reasoning internally, so splitting every prompt into steps isn't always worth it. Where these cards disagree with what you've read elsewhere, check the sources listed on each card.
Why does Claude give wrong answers?
Anthropic says even the most advanced models can sometimes generate incorrect text, and that this can be reduced but not eliminated. Common causes include missing context, facts after the knowledge cutoff, and no chance to admit uncertainty.
Does Claude make things up?
It can. Anthropic calls this hallucination and suggests letting Claude say it doesn't know, asking for quotes and citations, and limiting it to documents you provide. It also says to validate critical information.
Can I rely on Claude for legal or medical decisions?
Not by itself. Anthropic's Usage Policy says a qualified person must meaningfully review high-risk advice and decisions, including legal, medical and financial ones, before they're delivered, and people affected must be told AI was used.
Why does Claude ignore part of my prompt?
Parts can get lost in long, unstructured requests, and Anthropic says agents may stop after one part at lower effort. Use a numbered list when completeness matters, and ask for a check against it at the end.