It is tempting to reduce AI visibility to a number.
You test 50 prompts. Your business appears in 18 answers. Your AI visibility is 36%.
That number may be useful.
But on its own, it tells you surprisingly little.
Which prompts did you appear for? Were they broad research questions or searches from someone close to choosing a supplier? Did you appear consistently when the same prompt was repeated? Which competitors appeared instead? Did the testing expose subjects that matter to customers but are barely covered on your website?
And perhaps most importantly:
What are you going to do with what you learned?
The more AI visibility testing we carry out, the more we think it makes sense to view it as an ongoing process rather than simply a way of producing a visibility score.
A useful version of that process might look something like this:
Map → Select → Test → Observe → Investigate → Improve → Retest
The measurement still matters.
But much of the value comes from what happens around it.
Start by Mapping the Customer Journey
The first challenge is deciding which prompts are worth testing.
There is no realistic way to anticipate every question a potential customer might ask.
Conversational search makes that even less practical. Someone can describe their circumstances, add constraints, ask follow-up questions and express essentially the same underlying need in countless different ways.
The goal should therefore not be exhaustive prompt discovery.
A better starting point is to map the kinds of questions customers might ask as they move from initial exploration towards a possible buying decision.
For an accountancy firm, that journey might begin with:
Do I need an accountant if I use Xero?
It might then progress to:
What accounting support does a manufacturing company need?
Then:
Accountants in Birmingham experienced with manufacturing businesses
And eventually:
Which Birmingham accountant should I consider for a 25-person manufacturing company using Xero that needs payroll, year-end accounts and monthly management accounts for a fixed monthly fee?
These are not simply different phrasings of the same search.
The customer’s situation is becoming clearer.
We have looked at this in more detail in our guide to building an AI visibility query set around real customer situations.
The aim is to select a reasonable number of prompts that represent important customer needs at different stages of the journey.
Not every possible prompt.
Enough useful prompts to learn something meaningful.
More Prompts Do Not Automatically Mean Better Testing
It is easy to assume that monitoring 100 prompts must be better than monitoring 20.
It may be.
But only if those additional prompts add useful coverage.
One hundred searches clustered around slight variations of the same broad question could tell you less than 30 prompts deliberately covering different customer needs, situations and levels of buying intent.
We recently explored this when looking at OtterlyAI’s 100-prompt framework and what it can teach us about choosing better AI visibility prompts.
A large initial query set can be useful for exploration.
But it should not become a permanent list simply because those were the prompts chosen on day one.
Some may turn out to be less commercially interesting than expected.
Others may overlap heavily.
Testing may reveal new customer requirements worth adding.
A previously unknown competitor might cause you to investigate an entirely new group of questions.
Your prompt set should evolve as your understanding improves.
This is also why we don’t think businesses need to discover every possible AI query. A more practical approach is to test representative customer-need themes rather than attempting to predict every phrase somebody might type.
Repetition and Variation Answer Different Questions
There is another important distinction.
Repeating the same prompt and testing different prompts are not the same test.
If you run exactly the same prompt several times, you are looking at consistency.
Does the business appear repeatedly?
Does it appear occasionally?
Do the recommended companies change substantially from one run to another?
Our testing has repeatedly shown why a single AI answer should be treated cautiously.
That does not mean there is one universal number of repetitions every business should use. As we discuss in How Many Times Should You Repeat an AI Search When Testing Brand Visibility?, the appropriate level of repetition depends on what you are trying to establish.
Changing the prompt answers a different question.
Compare:
Accountant in Birmingham
with:
Accountant in Birmingham for a manufacturing company using Xero
and:
Birmingham accountant for a 25-person manufacturer needing payroll and monthly management accounts on a fixed monthly fee
The customer’s requirements are changing.
Testing those prompts helps you understand whether a business remains visible as the search becomes more specific.
So useful testing has at least two dimensions:
Coverage: Are we testing the right kinds of customer needs?
Consistency: How reliably does the business appear for the important ones?
A single visibility percentage can conceal both.
Commercially Meaningful Visibility Matters
Suppose a business appears frequently for:
What does an accountant do?
but rarely for:
Which Birmingham accountant should I use for a manufacturing business running Xero?
Those appearances should not automatically be treated as having equal commercial significance.
Early-stage visibility can still matter. Someone researching a subject today could become a customer later.
But businesses generally care about AI visibility because they ultimately hope it contributes to attracting customers.
That means it makes sense to pay particular attention to queries representing genuine customer problems, requirements and buying situations.
This is the thinking behind our RAMP Framework for measuring the commercial value of AI visibility.
Before asking whether the brand appeared, it is worth asking whether appearing for that particular query actually matters.
A high visibility percentage across weakly relevant prompts may be less interesting than much stronger visibility across a smaller group of commercially significant ones.
Testing Can Teach You About More Than Your Own Visibility
One of the more interesting things about AI visibility testing is that you inevitably observe more than your own brand.
You start with:
Does our business appear?
But every answer also shows you:
Who appears instead?
That can turn testing into a useful form of market observation.
We saw a good example of this when reviewing an OtterlyAI case study about NOLA Marketing’s work with Single Point of Contact.
What interested us was not simply the reported visibility improvement.
The wider process was more revealing.
The work explored queries across stages of the buyer journey, looked at where SPOC appeared and where it did not, examined competitor visibility and used the findings to inform subsequent content work.
The process also reportedly surfaced a competitor that SPOC had not previously been monitoring.
That is not really a visibility metric.
It is market intelligence produced by the testing process.
The Competitors AI Surfaces May Surprise You
The businesses you think of as your main competitors are not necessarily the same businesses that repeatedly appear alongside you in AI-generated answers.
That does not mean the AI system has identified the definitive competitive set.
But it does give you something worth investigating.
If an unfamiliar company repeatedly appears for commercially important searches, you might ask:
- What does its website communicate particularly clearly?
- Which customer problems does it address?
- Does it specialise in something important?
- What evidence does it provide?
- Are particular features repeatedly being associated with it?
- Is the AI system perhaps misunderstanding our business, its business, or both?
Testing may also reveal customer requirements you had not previously regarded as particularly important.
Price transparency, integrations, opening hours, industry experience, qualifications, delivery times, service areas or particular features may repeatedly influence the way businesses are presented.
Again, those discoveries would never appear in a simple visibility percentage.
A Visibility Gap Should Prompt an Investigation
This is where AI visibility testing starts to fit into a wider website and content process.
Suppose you identify an important customer query and your business rarely or never appears.
The temptation might be to conclude:
We need to write a page targeting that prompt.
We think the better question is:
Does our website actually contain useful, clear information addressing the underlying customer need?
Sometimes the answer will be yes.
If the site already covers the subject comprehensively, then the lack of visibility is interesting in itself.
But sometimes the audit reveals something different.
Perhaps the business genuinely provides the relevant service, but the website barely explains it.
Perhaps an important feature is mentioned only once.
Perhaps there is little evidence supporting a claimed specialism.
Perhaps competitors provide much more complete information around the same customer problem.
In that case, the visibility test may have uncovered a genuine weakness in the website.
Not Every Visibility Gap Is a Content Gap
That distinction matters.
A business failing to appear for a query does not prove that the solution is another article.
There may be several different kinds of gap.
Visibility gap
Your business does not appear for an important query.
Content gap
Your website does not adequately address the underlying customer need.
Communication gap
You provide the relevant service, but the website does not explain it clearly enough.
Evidence gap
You claim expertise or capability but provide relatively little supporting information.
Market-awareness gap
Testing reveals competitors, requirements or customer concerns that you had not previously considered.
Those should not be treated as interchangeable.
A visibility gap is therefore a reason to investigate.
It is not automatically an instruction to publish.
Improving Genuine Content Gaps Is Sensible — But It Is Not a Guarantee
This is where claims about AI optimisation need to be treated particularly carefully.
If potential customers are asking an important question, your business genuinely helps with that problem and your website barely covers it, improving the information is a sensible thing to do.
You are making the website more useful.
You are also making your business’s capabilities clearer to any system attempting to understand the site.
What we cannot reasonably say is:
Publish this page and your business will appear in AI answers.
There are too many variables.
Competition alone can make an enormous difference.
A specialist local service operating among a handful of comparable businesses is very different from trying to gain visibility in a market dominated by global brands, established publications and thousands of strong competing pages.
Other factors may include:
- the number of credible competitors;
- the authority and strength of those competitors;
- the amount and quality of competing information;
- how specific the query is;
- how clearly the website communicates the relevant expertise;
- what evidence exists to support its claims;
- how established different brands are;
- the particular AI platform being tested;
- and normal run-to-run variation.
Our own experiments reinforce the need for caution.
In Why Being Relevant Is Not Enough for AI Visibility, we found that a page could be relevant, useful and available to the search system without automatically being selected in the resulting AI answer.
That is why we prefer to talk about creating better conditions for visibility rather than promising a ranking outcome.
Improving the Website Can Still Be Worthwhile if Visibility Does Not Change
There is another benefit to approaching testing this way.
Suppose the process reveals that potential customers care about implementation times.
Your business has a well-developed implementation process, but the website barely discusses it.
Creating a useful resource explaining likely timescales, what affects them, what customers need to provide and where delays can arise could materially improve the site.
Perhaps AI visibility later improves.
Perhaps it does not.
The content can still help somebody who reaches the website through traditional search, a referral, social media, advertising or a direct recommendation.
That gives us a useful principle:
Don’t create content solely because you hope an AI system will reward it. Create it because testing has revealed a genuine customer need that your website could serve better.
Improved AI visibility can then be one possible outcome rather than the only justification for the work.
Retesting Closes the Loop
Once a genuine weakness has been identified and the website improved, visibility testing becomes useful again.
Run the relevant prompts later.
Has anything changed?
Are you appearing more frequently?
Has the way the business is described changed?
Are different sources being cited?
Are competitors still dominating?
Has nothing changed at all?
Any of those results can be informative.
We have already experimented with part of this process ourselves. In We Built a Page for an AI Search Query — Then Tested What It Took for AI Mode to Cite It, publishing a relevant page did not simply produce an immediate citation for the original query.
That experiment was another reminder that relevance does not equal guaranteed visibility.
It also illustrates why repeated testing is more useful than assuming a single before-and-after result establishes cause and effect.
If visibility changes after a website improvement, that does not automatically prove the improvement caused the change.
But observing results over time can still help build a much better picture.
The Process Becomes a Feedback Loop
Put all of this together and AI visibility testing starts to look less like a reporting exercise and more like a feedback loop.
1. Map the customer journey
Think about the questions customers may ask as they move from understanding a problem towards choosing a solution.
2. Select representative prompts
Choose enough useful searches to cover important customer situations without attempting to predict every possible wording.
3. Test and repeat
See which businesses appear and establish whether important results are reasonably consistent or highly variable.
4. Observe the wider answer
Look beyond your own brand. Record competitors, descriptions, citations, recurring requirements and unexpected findings.
5. Investigate meaningful gaps
Pay particular attention to places where the business is absent from commercially relevant searches.
6. Audit the website
Ask whether your site genuinely provides the information a potential customer would need at that point in the journey.
7. Improve genuine weaknesses
Strengthen useful content, communication or evidence where a real deficiency exists.
8. Retest
Observe what changes, what does not and what the next round of testing should investigate.
Then repeat the process.
Your Prompt Set Should Change as You Learn
One consequence of treating visibility testing as a process is that the original prompt list should not be regarded as fixed.
Your first prompt set is really a hypothesis.
You are saying:
We think these are representative questions our potential customers might ask.
Testing helps improve that hypothesis.
Some queries may prove too broad to be particularly useful.
Some may overlap.
Others may consistently reveal something important.
You might discover a new competitor.
A new customer requirement may emerge.
You may realise that searches much closer to the buying decision deserve significantly more attention.
That new information should influence what you test next.
AI visibility testing therefore becomes iterative.
You do not need the perfect query list on day one.
You need a sensible starting point and a process for improving it.
A Visibility Score Can Still Be Useful
None of this means visibility metrics are pointless.
Mention rates, citation rates, share of voice, platform comparisons and trends over time can all be useful.
The problem comes when the score becomes the destination.
Imagine two businesses complete a testing exercise.
The first concludes:
We have 42% AI visibility.
The second concludes:
We are reasonably visible during early research but become much less visible once customers specify a particular requirement. Two competitors dominate those buyer-ready searches. One of them was previously unknown to us. Our website barely explains a service those searches repeatedly require, so we are going to investigate whether that information needs improving and then continue testing.
The percentage may be useful.
But the second business has learned considerably more.
The Ultimate Goal Is Not the Highest Possible Visibility Percentage
Businesses are unlikely to invest time or money into AI visibility testing because they want a larger number on a dashboard.
Ultimately, they are interested because AI search may influence customer discovery and buying decisions.
That should remain the overarching focus.
The aim is to understand whether your business is being surfaced for the kinds of customer needs that could realistically matter commercially.
And the testing process can tell you much more besides.
It can reveal where visibility is strong and weak.
It can expose competitors.
It can uncover customer requirements.
It can highlight potential weaknesses in your website.
It can help you decide where further investigation is worthwhile.
And it gives you a structured way to return later and observe whether anything has changed.
AI visibility testing cannot tell you exactly what will make an AI system recommend your business.
It cannot promise that publishing a particular page will improve your visibility.
And no prompt list can represent every possible customer search.
But it can help you learn systematically from how your business and its market are represented across realistic customer journeys.
That is why we increasingly think of AI visibility testing as a process:
Map. Select. Test. Observe. Investigate. Improve. Retest.
The score still has value.
But the real goal is to use what you learn to make better decisions about your website, your content and the market in which you compete — ultimately improving your chances of reaching the customers your business is there to serve.