AI-search testing and optimization: prompt testing and iteration

What is prompt testing?

Prompt testing is systematically testing how AI tools respond to questions relevant to your brand. By testing the same prompts regularly, you measure progress and identify optimization opportunities.

Building a test set

Categories of prompts

1. Brand queries

• "What does [your company] do?"
• "Is [your company] reliable?"
• "Reviews about [your company]"

2. Service/product queries

• "Best [your service] agency in Belgium"
• "Who offers [your service]?"
• "What does [your service] cost?"

3. Comparisons

• "[Your company] vs [competitor]"
• "Alternatives to [competitor]"
• "Top 10 [your category] agencies"

4. Definition/education

• "What is [core concept from your industry]?"
• "How does [relevant process] work?"
• "What are the benefits of [your approach]?"

5. Problem/solution

• "How do I solve [problem]?"
• "Help with [challenge]"
• "Solve [symptom]"

Optimal test set size

• Minimum: 20 prompts
• Optimal: 30-50 prompts
• Enterprise: 100+ prompts

Testing protocol

Per test session:

1. Use the same prompts Consistency is crucial for comparability.

2. Test in multiple tools

   • ChatGPT 4o (with web search)
   • Perplexity
   • Gemini
   • Claude

3. Document systematically Per prompt note:

   • AI tool used
   • Date
   • Was your brand mentioned? (yes/no)
   • Was information correct? (yes/partly/no)
   • Which sources were cited?
   • Full answer (for analysis)

4. Screenshot evidence Save screenshots for later comparison.

Analysis framework

Mention rate

Percentage of prompts where your brand is mentioned.

Formula: (Number of mentions / Total prompts) × 100

Benchmark:

• 0-10%: Low, urgent action needed
• 10-30%: Average, improvement possible
• 30-50%: Good, fine-tune optimization
• 50%+: Excellent, maintain position

Citation rate

Percentage where your website is cited as source.

Correctness score

Percentage of correctly displayed information.

Sentiment

Positive, neutral or negative about your brand.

Iterative optimization

After each test round:

1. Identify gaps Which prompts don't yield a mention?

2. Analyze why

• Is content present but not cited?
• Is content completely missing?
• Is content not well structured?

3. Prioritize fixes Focus on high-impact prompts first.

4. Implement improvements

• Create new content
• Restructure existing content
• Add structured data
• Improve entity consistency

5. Retest After implementation, test the same prompts again.

Common findings and fixes

Testing tools and templates

Documentation spreadsheet

Columns:

• Prompt ID
• Prompt text
• Tool
• Date
• Mention (Y/N)
• Citation (Y/N)
• Correctness (1-5)
• Notes
• Screenshot link

Frequency

• Minimum: Monthly
• Optimal: Bi-weekly
• After major changes: Immediate retest

What can MatthCon do for you?

MatthCon sets up complete prompt testing programs for your brand. We build test sets, perform systematic tests and deliver actionable insights. Start with an audit or read more about measuring impact.