AI Visibility Testing is an independent practical resource for understanding how brands appear in AI search. We explain how to test visibility manually, what those results actually mean, and when monitoring software becomes useful as the number of prompts, platforms and competitors grows.
AI visibility can be surprisingly difficult to measure.
A business might be recommended in one AI-generated answer and disappear from the next. Its website might be cited without the business itself being recommended. A broad search might favour familiar market leaders, while a more detailed customer situation produces a completely different shortlist.
That is why AI Visibility Testing takes two connected approaches.
If you want to test the AI visibility of your own business, start with the practical guides.
If you want to understand the experiments and evidence behind that advice, explore the testing section.
The two belong together.
The experiments help us understand what happens.
The practical guides turn those observations into methods businesses can use for their own testing.
I Want to Test My Business
If your main question is:
How do I actually test whether my business is visible in AI search?
start here.
1. Choose Searches That Actually Matter
How to Choose AI Search Prompts That Actually Matter to Your Business
The prompts you choose determine what your visibility results mean.
A business can look highly visible across broad searches while being almost invisible when a potential customer describes a more specific requirement.
This guide explains why AI visibility testing should begin with realistic customer situations rather than simply creating a long list of loosely relevant prompts.
2. Run the Test Consistently
How to Run a Repeatable Manual AI Visibility Test
Finding your business once does not establish consistent visibility.
This guide explains how to create a simple manual testing process using the same prompt, consistent conditions and repeated searches.
It covers what to record, why repetition matters and how an initial test can become a baseline for future monitoring.
3. Understand What Kind of Visibility You Have
Mentioned, Cited or Recommended: What Kind of AI Visibility Do You Have?
An appearance in an AI-generated answer does not always mean the same thing.
A business might be:
- directly recommended;
- mentioned elsewhere in the answer;
- used as a supporting source;
- cited without being recommended at all.
This guide explains why those different appearances should not automatically be combined into one generic AI visibility figure.
4. Calculate Recommendation Visibility
How to Calculate AI Recommendation Share of Voice Manually
Once you have repeated search results, you can begin turning them into useful measurements.
This guide demonstrates simple calculations for:
- recommendation coverage;
- share of recommendation positions;
- first-position rate;
- average recommendation position.
The calculations themselves are straightforward. The important part is being clear about exactly what has been counted.
5. Measure Visibility Across Different Customer Requirements
How to Measure AI Share of Voice Manually: What 25 Searches Revealed
Repeating one query tells you how consistently a business appears for that particular situation.
It does not tell you whether the same business remains visible when the customer’s requirements change.
This worked example uses five different customer requirements and 25 Google AI Mode searches to demonstrate how the apparent AI visibility leader can change depending on what the customer actually needs.
6. Check Whether AI Understands Your Business Correctly
How to Test If AI Understands Your Brand: A Simple Sense Check
Visibility is only useful if the AI-generated description gives people the right impression of your business.
This simple test compares what you believe your website communicates with the description an AI system generates.
It can help identify:
- incorrect services;
- misunderstood audiences;
- invented features;
- confused business identities;
- important information that has been omitted.
A Simple Starting Process
If you are completely new to AI visibility testing, you do not need to test hundreds of prompts.
Start small.
Choose several realistic customer situations.
Think about who your customers are, the problems they need solved and the details that could genuinely affect which provider is suitable.
Run each important query several times.
Do not treat one favourable or unfavourable answer as a permanent result.
Record different forms of visibility separately.
Was the business recommended?
Was it merely mentioned?
Was its own website cited?
What was said about it?
Compare competitors.
Which businesses appear repeatedly for the same customer situations?
Repeat the test later.
That turns a snapshot into monitoring.
The objective is not to find one impressive AI result.
It is to understand the pattern.
I Want to See the Experiments Behind the Advice
The practical guidance on this site has developed from hands-on testing.
If you want to see what actually happened when identical searches were repeated, customer requirements were changed or AI descriptions were checked, the experiments below provide the underlying evidence.
What Happens When You Repeat Exactly the Same AI Search?
I Ran the Same Google AI Mode Product Search 10 Times — Here’s How the Recommendations Changed
This was one of the foundational experiments.
Exactly the same product-selection query was submitted to Google AI Mode ten times.
Six different products appeared.
Four different products occupied first place.
Only one product appeared in every response.
The experiment showed why one AI search cannot be treated as a fixed ranking.
What Happens If You Keep Repeating It?
Why One AI Search Isn’t Enough: What 20 Repeated Google AI Mode Tests Revealed
The original ten-run experiment was extended to 20 identical searches.
The answers continued to change, but something else happened:
a recognisable core group of products became visible.
The experiment suggested an important distinction:
Repeated testing does not necessarily make AI answers stop changing. It can reveal the pattern underneath that variation.
What Happens When the Customer’s Requirements Change?
How to Choose AI Search Prompts That Actually Matter to Your Business
This experiment compared progressively more detailed versions of the same underlying product search.
The recommendation set changed as more meaningful information about the customer and their requirements was introduced.
That led to an important practical conclusion:
AI visibility should be tested against the situations faced by genuine potential customers, not just broad category searches.
What Does AI Share of Voice Look Like Across Several Customer Situations?
How to Measure AI Share of Voice Manually: What 25 Searches Revealed
Five different project-management software requirements were each tested five times.
That created 25 responses and 75 recommendation positions.
Different products became strongest for different requirements.
The experiment showed why AI share of voice is not simply a permanent characteristic of a brand.
It depends on the prompts included in the measurement.
Does AI Default to Familiar Market Leaders?
Does Google AI Mode Default to Market Leaders? What 25 Broad Searches Revealed
This experiment looked specifically at broad recommendation searches.
The results help explore whether general questions tend to produce familiar, established providers — and why more specific customer circumstances may be needed before different businesses enter the recommendation set.
This matters when interpreting broad-query visibility.
Being highly visible for a general search does not necessarily mean a business will remain visible when the prospective customer describes what they actually need.
Can AI Mention a Website but Misunderstand What It Does?
Google AI Mode Mentioned My Website — But Invented a Tool That Does Not Exist
This experiment produced what initially looked like a successful brand appearance.
AI Mode found AIVisibilityTesting.com and described it confidently.
There was one problem.
It described the site as an automated software platform with functionality that did not exist.
The experiment demonstrated why counting a brand appearance without reading what was actually said can produce a misleading visibility result.
Can a Business Be Recommended Without Its Website Being the Main Source?
Mentioned, Cited or Recommended: What Kind of AI Visibility Do You Have?
One Google AI Mode answer contained only three recommended products but many more linked sources.
Some businesses were recommendations.
Some websites provided supporting information.
Some appeared in both roles.
This experiment helped establish why recommendation visibility and citation visibility should be examined separately.
What the Experiments Are Teaching Us
The site is still developing, and the findings should not be treated as universal rules about how every AI system selects recommendations.
But several recurring principles are becoming useful for practical testing.
One AI search is not enough
AI-generated answers can change even when the query does not.
Repeated testing provides a better picture than one isolated result.
The query determines what you are measuring
A broad category search and a detailed customer situation may produce very different recommendation sets.
A visibility percentage therefore only has meaning when you know which queries produced it.
More prompts do not automatically mean better testing
Fifty superficial variations of the same underlying question may tell you less than ten genuinely different customer situations.
The important question is whether each prompt represents something that could affect which provider is appropriate.
Recommendation, mention and citation visibility are different
A business can be named without being recommended.
A website can be cited without the business being presented as a solution.
A business can also be recommended while other websites provide much of the supporting evidence.
These appearances should be recorded separately.
Visibility does not guarantee accurate representation
AI can find a business and still misunderstand what it provides.
Testing should therefore look not only at whether the brand appeared, but also at what the potential customer was told about it.
Practical Advice and Experiments Should Stay Connected
The practical guides on this site are not intended to be unsupported theories about how AI search works.
Where possible, the advice develops from experiments.
The process is:
Test something → record what happened → identify a pattern → test the pattern again → turn the useful parts into practical guidance.
Equally, an individual experiment should not automatically become a universal rule.
AI search systems change.
Different prompts behave differently.
Different industries may produce different patterns.
That is why both sides of the site matter.
If you simply want to test your business, use the practical guides.
If you want to understand why those testing methods are being suggested, read the experiments behind them.
And if a future experiment challenges one of our current assumptions, the practical guidance should change with the evidence.
Where Should You Start?
If you want to begin testing your own business today, I suggest this sequence:
- Choose AI search prompts that actually matter to your business.
- Run a repeatable manual AI visibility test.
- Understand whether you were mentioned, cited or recommended.
- Calculate recommendation visibility from your repeated results.
- Check whether AI understands and represents your business correctly.
You do not need specialist software to learn something useful.
Start with a small number of commercially meaningful questions, test them consistently and record what actually happens.
That is the foundation of useful AI visibility testing.