Skip to content
Back to course

8. Troubleshooting and improving prompts

Welcome to the final lesson of this course! You have already learned how to write clear, role-based, example-rich prompts. But even experienced prompt writers sometimes get an answer that misses the mark. In this lesson you will learn a simple, repeatable method for diagnosing what went wrong and rewriting your prompt until the AI gives you exactly the answer you need.

Getting a bad answer from an AI does not mean the AI is broken — it almost always means the prompt was unclear, incomplete, or asked for the wrong thing. Understanding the most common reasons a prompt fails makes it much easier to fix.

Here are the four most common causes of a bad AI answer:

1. Too vague — the prompt does not give enough detail. The AI has to guess, and its guess may not match what you wanted.

Example: 'Help me with my business.' — What kind of business? What kind of help?

2. Too much at once — the prompt asks for several different tasks in one go. The AI may handle some well and ignore others.

Example: 'Summarise this text, translate it, write a new version, and give me feedback.' — That is four separate tasks.

3. No role or context — the AI does not know who it is speaking to or why, so it gives a very general answer.

Example: 'What is a budget?' — for a primary-school child? for a small-business owner in Addis Ababa? for an accountant?

4. Wrong format — the output is in a form you cannot use. You needed a table but got a paragraph. You needed three bullet points but got a 400-word essay.

Once you know which of these four problems is causing your bad answer, fixing the prompt becomes straightforward.

promptAI
Every prompt goes through an improvement loop: write → review the output → diagnose the problem → rewrite. Most prompts only need one or two revisions to get a great result.

When the AI gives you an answer you do not want, stop before you retype the whole prompt from scratch. Instead, ask yourself these four diagnostic questions in order:

Question 1 — Was the task clear?

Read your prompt again. Would a person who knows nothing about the topic understand exactly what you wanted? If not, add more detail or break the request into smaller pieces.

Question 2 — Did I give enough context?

Did the AI know who you are, what you need the answer for, and who the audience is? If not, add a sentence of background before the main task.

Question 3 — Did I ask for the right format?

Look at the output. Is the problem the content, or just the shape? If the AI gave the right information but in the wrong layout, the fix is a format instruction — not a complete rewrite.

Question 4 — Was the prompt one task or many?

Count the instructions you gave. If there are more than two main tasks, split the prompt. Ask one task at a time and build the answer step by step.

Working through these four questions usually points to the single change that will fix the problem. You do not need to guess — you diagnose.

The rewrite rule: change ONE thing at a time. If you rewrite the whole prompt at once, you will not know which change fixed the problem. Make one targeted edit, send the new prompt, and check whether the answer improved. Repeat until you are satisfied.

Let us look at three real-world before-and-after examples to see the method in action.

EXAMPLE 1 — Abebe owns a small electronics repair shop in Mercato, Addis Ababa.

Before (vague prompt): 'Write a message for my customers.'

Problem: No context, no purpose, no audience detail, no format.

Diagnosis: Fails questions 1, 2, and 3.

After (improved prompt): 'You are a friendly small-business owner in Mercato, Addis Ababa. Write a short WhatsApp message (under 40 words) to your regular customers, letting them know your phone-repair shop will be closed this Saturday for a public holiday. Warm, polite tone.'

Result: A ready-to-send WhatsApp message in the right length and tone.

EXAMPLE 2 — Sara is studying for a school exam in Hawassa.

Before (too many tasks at once): 'Explain photosynthesis, translate it to Amharic, make a quiz, and draw a diagram.'

Problem: Four tasks in one prompt — the AI may skip some or do all of them poorly.

Diagnosis: Fails question 4.

After (split into separate prompts):

Prompt A: 'Explain photosynthesis in simple English for a Grade 8 student. Use three short paragraphs.'

Prompt B: 'Translate this explanation into Amharic.' (paste Prompt A's result)

Prompt C: 'Write five multiple-choice quiz questions based on this explanation.'

Result: Each task is done well because each prompt is focused.

EXAMPLE 3 — Almaz manages a small grocery shop in Jimma and asked for a price list.

Before (wrong format): 'Give me a price list for my ten grocery items.'

Problem: The AI wrote a paragraph naming items and prices in flowing prose — hard to read quickly at the counter.

Diagnosis: Fails question 3 (format).

After (format added): 'Give me a price list for my ten grocery items. Present it as a table with two columns: Item Name and Price (in Birr). Each row is one item.'

Result: A clean, printable table Almaz can stick on her wall.

To help you remember the full prompt-improvement process, use the CLEAR checklist. Before you send any prompt, run through these five checks:

C — Concrete: Is the task concrete and specific? Have you said exactly what you need?

L — Length and format: Have you told the AI how long the answer should be and in what format (bullet, table, paragraph)?

E — Examples: Have you given at least one example of a good answer when the task is new or complex?

A — Audience: Have you told the AI who the answer is for — a beginner, a child, an expert, a customer?

R — Role: Have you given the AI a role that fits the task — teacher, business writer, friendly assistant?

If any of these five checks fails, fix it before you send the prompt. This single habit will cut your revision rounds in half.

Here is a complete CLEAR prompt Almaz can use for her grocery shop:

'You are a friendly neighbourhood shop assistant in Jimma, Ethiopia. [ROLE]

I need a one-week grocery promotion plan for Ramadan, targeting families. [CONCRETE + AUDIENCE]

For each day Monday to Friday, suggest one discounted item and one short sale message. [CONCRETE]

Here is an example for Monday: "Teff flour — 10% off — Stock up for injera this week!" [EXAMPLE]

Present the plan as a table with three columns: Day, Discounted Item, Sale Message. [LENGTH AND FORMAT]'

Avoid over-engineering your prompts. Beginners sometimes try to write a perfect 10-line prompt for a simple question and end up confusing the AI with too many competing instructions. Start simple: if a short prompt gives a good answer, use it. Only add complexity when the simple version falls short.

Scenario

Tigist is a pharmacy assistant in Dire Dawa. She typed this prompt: 'Tell me about medicines.' The AI gave a very long, general answer about hundreds of medicines. Which single change will most improve her next prompt?

Course recap — Troubleshooting and improving prompts: • Bad AI answers almost always point to a problem in the prompt, not the AI. • Diagnose before you rewrite: ask the four questions (clear task? context? format? one task at a time?). • Change ONE thing per revision so you know what worked. • Use the CLEAR checklist — Concrete, Length/format, Examples, Audience, Role — before you send any important prompt. • Iteration is normal and part of good prompt engineering. Congratulations — you have completed the Prompt Engineering and AI Assistants course! You now know how to write precise prompts, use roles and examples, structure step-by-step instructions, control output format, and troubleshoot when things go wrong. These skills will help you get more value from every AI tool you use.

Check your understanding

1/7 · 82 XP

When an AI gives you a bad answer, the most likely cause is that the AI is broken or unintelligent.