All articles
tracking6 min read

Prompt testing: systematic measurement of AI visibility

Measuring is steering. Set up a prompt test suite that measures the same thing every week.

Published: January 15, 2026Updated: January 22, 2026

In short

  • Prompt testing = systematically testing how AI talks about your brand
  • Build a set of 20-50 relevant prompts
  • Test monthly in ChatGPT, Perplexity, Gemini
  • Measure mention rate, citation rate, correctness
  • Document everything for trend analysis

You can't improve what you don't measure. That applies to AI visibility too. Prompt testing is the foundation of your measurement strategy.

What is prompt testing?

Prompt testing is systematically testing how AI tools respond to questions relevant to your brand. You ask the same questions regularly and document the answers.

The difference from ad-hoc testing: consistency. Only with the same prompts over time can you see trends.

Building a test set

Step 1: Categorize your prompts

Brand queries (5-10 prompts)

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

Service queries (5-10 prompts)

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

Comparisons (5-10 prompts)

  • "[You] vs [competitor]"
  • "Top 10 [your category]"
  • "Alternatives to [competitor]"

Definitions (5-10 prompts)

  • "What is [core concept]?"
  • "How does [process] work?"

Problem/solution (5-10 prompts)

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

Step 2: Refine your prompts

Bad prompt: "marketing" Good prompt: "Best B2B marketing agency in Belgium for SaaS companies"

Bad prompt: "CMO" Good prompt: "What does a fractional CMO cost for a scale-up?"

Step 3: Prioritize

Start with 20-30 prompts. Expand later. Quality over quantity.

The test protocol

Which tools to test?

Minimum:

  • ChatGPT (with web search on)
  • Perplexity

Extended:

  • ChatGPT
  • Perplexity
  • Gemini
  • Claude

How to document?

Create a spreadsheet with columns:

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

How often to test?

  • Minimum: monthly
  • Optimal: bi-weekly
  • After major content updates: immediately

What do you measure?

Mention rate

Formula: (Prompts with mention / Total prompts) × 100

Example: 8 of 30 prompts = 26.7% mention rate

Citation rate

Formula: (Mentions with link / Total mentions) × 100

Example: 5 of 8 mentions have link = 62.5% citation rate

Correctness score

Rate each mention on correctness: 1 = Completely incorrect 3 = Partially correct 5 = Completely correct

Average over all mentions = correctness score.

Trend over time

Plot your metrics monthly. Rising line = progress.

Avoiding pitfalls

Inconsistent prompts

Don't change your prompts in between. Then you can't compare.

Too little documentation

Screenshot everything. You'll need it later for analysis.

Only looking for positives

Also document where you DON'T appear. Those are your opportunities.

Too quick conclusions

AI models update irregularly. Give it at least 3 months.

From testing to action

After each test round:

  1. Analyze gaps: Which prompts don't yield a mention?
  2. Identify patterns: What do successful prompts have in common?
  3. Prioritize fixes: Focus on high-impact prompts
  4. Implement: Update/create content
  5. Retest: Test again after 4-6 weeks

Example reporting

Monthly Report January 2026

Summary

  • Mention rate: 26.7% (+4% vs December)
  • Citation rate: 62.5% (+12% vs December)
  • Correctness: 4.2/5 (stable)

Wins

  • First citation in Perplexity for "fractional CMO Belgium"
  • Correct mention in Gemini for brand query

Gaps

  • No mentions for comparison prompts
  • Competitor X dominates category queries

Actions next month

  1. Create comparison page X vs Y
  2. Expand FAQ with pricing info
  3. Add case study

FAQ

Can I automate prompt testing?

Partially. Tools are emerging, but manual testing gives more insight.

How much time does it take?

2-3 hours per month for 30 prompts in 2 tools.

What if my results don't improve?

Analyze why. Often: content not citable enough or technical blockers.


Need help? MatthCon sets up complete prompt testing programs. Schedule a call.

prompt testingmetenKPIspraktisch

Request an AI-search audit

Discover where you stand and what you can improve

Start the audit