Writing Effective Follow-Up Prompts

A follow-up prompt is not a complaint — it is a correction signal that steers the model toward a better output.

Effective follow-up prompts are specific about what was wrong and what you want instead. Vague follow-ups like "Try again" or "That's not quite right" give the model nothing useful to work with.

Learning a small set of course-correction patterns lets you fix almost any type of output problem quickly.

When a response is too long, verbose, or padded, use length-reduction patterns:

• "Make it shorter — aim for 100 words."
• "Remove the introductory paragraph and any filler sentences."
• "Summarize the above in three sentences."
• "Cut this by half without losing the key points."

Specificity helps: telling the model a target word count or a specific thing to remove is more effective than just saying "shorter."

When a response is too general or high-level, use specificity-injection patterns:

• "Be more specific about the pricing model."
• "Give concrete numbers, not percentages."
• "Focus on the case where the user is a small business owner, not enterprises."
• "The advice about security is too vague — name the specific vulnerability and how to fix it."

These patterns narrow the model's focus to the dimension where you need more depth.

When a response includes content you do not want, use removal patterns:

• "Remove the part about historical context — I only need the current situation."
• "Delete the last two bullet points, they are off-topic."
• "Rewrite without the disclaimer paragraph at the end."
• "Cut everything after the third point and end there."

Reference the specific section to remove. Saying "remove the irrelevant parts" is ambiguous — the model may not agree on what is irrelevant.

When a response is correct but abstract, use example-addition patterns:

• "Add a concrete example after each bullet point."
• "Illustrate the third point with a real-world scenario."
• "Give one before/after code snippet for each best practice."
• "Add a counterexample showing what NOT to do for each recommendation."

Examples make abstract advice actionable. This pattern is especially useful for instructional content and how-to guides.

When a response has the right content but the wrong structure, use reformatting patterns:

• "Rewrite this as a bulleted list."
• "Convert this into a numbered step-by-step guide."
• "Break this into sections with H2 headers."
• "Format this as a comparison table: Feature | Option A | Option B."

Reformatting follow-ups preserve the content and only change the presentation. They are low-risk — the substance is already good.

You can stack multiple correction patterns in a single follow-up prompt:

"Rewrite the above response as bullet points, remove the historical background section, and add a code example for each point. Keep it under 300 words."

This is more efficient than sending four separate follow-up messages. However, keep the combined follow-up clear and readable — if it becomes complex, split it into two follow-ups rather than one confusing one.

You can automate follow-up correction in a Python pipeline by appending correction instructions to the conversation history:

The most effective follow-up prompts anchor to specific parts of the previous response. This removes ambiguity:

• Weak: "Can you improve the second part?"
• Strong: "In the section titled 'Security Considerations', the advice is too general. Rewrite that section with three specific firewall rules."

Anchoring by quoting or naming a specific section, sentence, or point forces the model to focus on exactly what you mean.

Sometimes a follow-up pattern does not produce the improvement you want. Signs of a stuck conversation:

• The model reverts to the same structure after two corrections
• It acknowledges your correction but keeps making the same error
• Each correction creates a new problem while fixing the old one

At this point, starting fresh with a rewritten original prompt is often faster than continuing to correct. We will cover that decision in a later lesson.

Because correction patterns are reusable, it is worth keeping a small personal library of follow-up phrases you reach for repeatedly:

• "Make it shorter — target X words."
• "Add a concrete example for each point."
• "Rewrite as bullet points with a one-line summary at the top."
• "Remove the disclaimer / historical background / introductory fluff."
• "Be more specific about [X] — give actual numbers / steps / code."

Storing these in a text file or note-taking app saves time on every future session.

You asked the model to write a product description for a project management app. It returned a 600-word essay covering the app's history, philosophy, core features, pricing, and competitor comparison. You only needed a 100-word description of the core features for a website hero section.

Which follow-up prompt is most effective?

Follow-up prompts are correction signals, not complaints. The five core patterns — shorten, specify, remove, add examples, reformat — cover the vast majority of output problems you will encounter.

The key principles are: be specific about what was wrong, anchor to the relevant section, and combine patterns when needed. When follow-ups stop working, consider starting fresh.

In the next lesson, you will practice multi-turn conversations that build on previous responses rather than correcting them.