← Daily AI recipesStudy✓ Verified Jul 18, 2026

Get excellent AI results in 2-3 iterations instead of the first response

A request reformulation technique that transforms a mediocre result into accurate and useful output

  1. 1.Open Claude.ai in your browser and log in via Google or email.

    ⚠️ Third party: Anthropic processes your content according to their privacy policy. Do not share company confidential data, medical information, or trade secrets.

    Claude.ai ↗
  2. 2.Enter your first prompt with a description of the task. For example, ask the AI to write text, analyze data, or create a plan.

    ⚠️ Personal data: do not paste real names, phone numbers, addresses, or confidential information into your prompt unless it is anonymized.

    Claude.ai ↗
    Prompt
    You are an expert in [TASK AREA]. Help me [SPECIFIC TASK]. The result should be [REQUIRED FORMAT AND CRITERIA].
    What this prompt doesReplace [TASK AREA] with the field where you need help (marketing, programming, copywriting, etc.), [SPECIFIC TASK] with what you need done, [REQUIRED FORMAT] with your desired answer format (list, article, code, etc.). This prompt sets the context and role for the AI.
  3. 3.Read the first response. Note what doesn't work: too generic, lacks detail, wrong tone, or missing aspects.

    Do not accept the first response as final - it is a baseline, not a solution.

    Claude.ai ↗
  4. 4.Write a second prompt with specific revisions. Use this formula: state what you liked, what needs to change, and add new criteria.

    ⚠️ Confidentiality: if you add examples or context, ensure they do not contain real company data or personal information.

    Claude.ai ↗
    Prompt
    The previous answer was good in [WHAT WORKED], but I need you to change [WHAT TO FIX]. Also add [ADDITIONAL CRITERIA]. Make it [DESCRIPTION OF DESIRED IMPROVEMENT].
    What this prompt doesReplace [WHAT WORKED] with the strengths of the previous response, [WHAT TO FIX] with specific shortcomings, [ADDITIONAL CRITERIA] with new requirements, [DESCRIPTION OF DESIRED IMPROVEMENT] with the desired result (more detailed, shorter, simpler, with examples). This speeds up corrections.
  5. 5.Evaluate the second response. If the result is close to ideal - use it. If final tweaks are needed - write a third prompt with the most specific, concrete corrections.

    Usually 2-3 iterations are enough to get a working result. If after the third attempt the answer still doesn't satisfy you - rephrase the task from scratch.

    Claude.ai ↗
  6. 6.Copy the final result and use it - in document text, code, email, or plan. Save the conversation in Claude for history (bookmark function or screenshot).

    The conversation is stored in your Claude account for 24 hours; if you need a longer history - copy the result to notes or a document.

    Claude.ai ↗
💰 What it costs to followprices as of 2026-07
  • freeClaude.ai (free plan)No limits on the number of requests for users in most countries; Anthropic provides free access to all features in the web version.
Try it and be the first to check in

Why today

Anthropic Officialweb
Claude Documentationweb

AI services change fast - interfaces and free limits may differ from what's described.