Google AI Mode and other AI-powered search tools can mention businesses, recommend brands and cite websites when answering questions.
This creates an important new question for website owners:
When, where and how often does your brand appear in AI-generated answers?
Answering that question is not as simple as searching once.
Ask the same question more than once and you may receive different answers, sources and recommendations. Change the wording slightly and the results may change again.
AI visibility is therefore not a fixed ranking position. It is a pattern that needs to be tested and measured.
AI Visibility Testing explores practical ways to run those tests, record the results and understand what they can—and cannot—tell you.
One Search Is Not a Visibility Test
Finding your website cited in an AI-generated answer can be encouraging. Seeing a competitor recommended instead can be concerning.
But neither result means very much on its own.
AI-generated answers can vary between repeated searches. A brand that appears in one response may be missing from the next. A website may be cited for one version of a question but not another.
A meaningful test needs to look beyond a single result.
It should consider:
- The exact question that was asked
- Which AI search platform was tested
- Whether the test was repeated
- Whether the brand was mentioned
- Whether the brand’s own website was cited
- Which competing brands appeared
- How the results changed when the question was reworded
- Whether the pattern changed when the test was repeated later
The objective is not to find one favourable result. It is to build a more reliable picture from a collection of results.
What Can Be Measured?
AI visibility can take several forms, and they should not all be treated as the same thing.
Brand mentions
An AI-generated answer may name or recommend a business without linking to its website.
Website citations
The answer may cite or link to a page from the business’s own website as one of its sources.
Competitor appearances
A test can record which competing brands are mentioned or cited in response to the same questions.
Share of voice
Share of voice measures how frequently a brand appears across a defined group of tests, either as a proportion of all the responses or in comparison with selected competitors.
For example, imagine testing ten customer questions three times. That produces 30 AI-generated responses.
If your brand appears in eight of those responses, while a competitor appears in 17, that provides a more useful comparison than searching for each business once.
It is not a permanent visibility score. It does not predict what every user will see. It is a measurement based on a clearly defined set of tests.
For any share-of-voice figure to be meaningful, the testing method must explain what was counted, which prompts were used, where they were tested and when the tests took place.
Starting With Manual Testing
You do not need specialist software to begin investigating your AI visibility.
A basic manual test can start with a small number of realistic questions that potential customers might ask.
Each question can be repeated several times while recording:
- The complete prompt
- The AI platform used
- The date of the test
- Brands mentioned
- Websites cited
- Competing businesses appearing
- Differences between repeated responses
The wording of a prompt can then be changed carefully to see how factors such as location, service type or customer circumstances affect the answer.
Manual testing is useful because it makes the process visible. You can see exactly what is being tested, what is being counted and why the results vary.
It also helps prevent a single response from being presented as evidence of consistent AI visibility.
When Manual Testing Becomes Impractical
Manual testing works well when you only have a handful of prompts to check occasionally.
The workload grows quickly when you want to monitor more prompts, platforms or competitors.
Ten prompts repeated three times across four AI platforms would produce 120 individual responses. Repeating the exercise every week would mean checking and recording 480 responses each month.
The process becomes even more demanding for an agency monitoring several clients or a business tracking a large number of products, services and customer questions.
At that point, an automated AI visibility-monitoring service may be more practical than continuing with a spreadsheet.
Automation does not make a brand more visible. Its purpose is to perform repeated checks, collect the results and show patterns without requiring every response to be recorded manually.
AI Visibility Testing will examine both approaches: how to conduct useful manual tests and how to recognise when the scale of the task may justify automated monitoring.
What This Site Is About
AI Visibility Testing is intended for:
- Business and website owners
- Marketing teams
- Content teams
- SEO professionals
- Consultants and agencies
- Anyone responsible for monitoring how a brand appears online
The site will explore manual testing methods, repeated-prompt experiments, brand mentions, website citations, competitor comparisons and AI share of voice.
It will also look honestly at the limitations of this kind of measurement.
AI search systems can produce variable results. No individual test can guarantee what another user will see, and no monitoring platform can provide a permanent or universal score.
This site does not promise to make a website appear in AI-generated answers. It does not claim to secure citations or optimise a brand for AI search.
Its purpose is more focused:
To help you test AI search systematically, understand what appears and measure the patterns that emerge.
Begin With a Small Experiment
Think of three to five questions that a genuine customer might ask when looking for a business like yours.
Run each question several times. Record the brands mentioned, the websites cited and the differences between the responses.
That small experiment will not tell you everything about your AI visibility.
But it will tell you considerably more than searching once.