One of the strangest things about AI search is that you can ask the same question twice and get a different answer.
A business might be recommended in Google AI Mode, ChatGPT or another AI search tool on one search, then disappear when you repeat the query later.
That can make AI visibility feel random.
But “random” is not quite the right way to think about it.
A better mental model is a box full of business cards.
Imagine AI Search Has a Box of Business Cards
Suppose somebody asks:
“What are the best accountants in Bristol for a small construction company?”
Imagine that AI search effectively creates a box containing the business cards of accountants it considers relevant to that particular question.
But the box is not evenly balanced.
Some businesses appear to have much stronger evidence connecting them with the customer’s need than others.
So, in our analogy, one accountant might have the equivalent of 20 cards in the box.
Another might have 10.
Another might have only two.
Some firms may not make it into the box at all.
Now imagine the AI draws five different businesses to recommend.
Those become the businesses shown in the answer.
Put the cards back, mix the box and draw again.
You might get a different combination.
That is a useful way to think about the probabilistic nature of AI search.
One Search Shows You the Draw, Not the Odds
Suppose Accountant A has the equivalent of 20 cards in the box and Accountant B has only two.
Accountant A is much more likely to appear.
But that does not necessarily mean Accountant A will be selected every single time.
Likewise, Accountant B might occasionally appear despite being a much weaker candidate overall.
This gives us an important distinction:
One AI search tells you which cards were drawn. Repeated testing starts to tell you how many cards you might have in the box.
If a business appears in 18 out of 20 searches, that tells a very different story from appearing once.
Equally, failing to appear in one search does not prove that a business has no AI visibility.
It may simply not have been drawn that time.
First, You Have to Get Into the Box
Before worrying about how many imaginary cards your business has, there is a more basic question:
Does your business belong in the box at all?
Suppose you run an accountancy practice specialising in construction businesses.
Your website says:
“We provide accountancy and tax services to businesses across Bristol.”
Now compare that with another practice whose website clearly explains that it helps construction companies with:
- CIS returns
- VAT
- payroll
- Xero
- subcontractor payments
- job profitability
- cash-flow forecasting
- year-end accounts
Which website gives an AI system more evidence from which to understand exactly who the business helps and what problems it solves?
Probably the second.
That does not mean adding a few phrases to your website automatically makes the AI recommend you.
The point is simpler.
AI systems need evidence from which to connect a business with a customer need.
If your offer is vague, poorly described or missing important detail, the AI has less information from which to make that connection.
In our analogy, the first job is to give the AI a good reason to put your business card into the box.
Then You Want Stronger Representation
Once your business is a plausible candidate, stronger and more consistent evidence may increase the likelihood that it is surfaced.
That could include:
- clear service pages
- detailed explanations of problems you solve
- location information
- industry specialisms
- useful supporting content
- reviews
- third-party mentions
- authoritative references
None of these things should be thought of as mechanically adding one more card to the box.
The analogy is simply a way of thinking about probability.
The more convincing the overall evidence that your business is genuinely relevant to a particular customer need, the more strongly represented it may become.
Or, in Business Card Box terms:
You are trying to earn more cards.
Every Question Creates a Different Box
There is not one permanent box.
Change the question and the candidate set can change too.
Consider:
“Best accountant in Bristol for a construction company.”
Now compare it with:
“Accountant in Bristol who can help me move from Sage to Xero.”
Some firms might be relevant to both.
Others may have strong evidence around construction accounting but very little around software migration.
Change the customer need again:
“Accountant in Bristol experienced in helping technology startups raise investment.”
Now the box may look very different.
This is one reason AI visibility cannot sensibly be reduced to a single ranking.
A business may have strong visibility for one customer need and almost no visibility for another.
Sometimes You Should Not Be in the Box
This is also an important point.
The objective is not to appear for everything.
If you specialise in small construction companies, there may be no reason for an AI system to recommend you to somebody looking for specialist accounting support for a biotech startup.
That is not an AI visibility failure.
It may actually mean the system has understood your business correctly.
The aim is not to get as many cards as possible into every box.
It is to make sure your business is strongly represented in the boxes that correspond to customers you can genuinely help.
That is a much more useful objective than simply chasing mentions.
This Is Why Repeated AI Visibility Testing Matters
We cannot open Google AI Mode or ChatGPT and literally inspect the box.
We cannot count the cards.
And AI systems are obviously far more complicated than someone drawing business cards from a container.
The Business Card Box is only an analogy.
But repeated testing gives us something useful.
If we run the same or closely related prompts multiple times, we can observe how frequently businesses appear.
That starts to reveal patterns.
A business that appears once in 20 tests probably has relatively weak visibility for that need.
A business that appears 17 times appears to have much stronger visibility.
We still do not know exactly why the AI selected it.
But we have moved from looking at a single outcome to measuring a pattern.
AI Visibility Is About Probability, Not Just Position
Traditional search trained businesses to think in terms of fixed positions:
Number 1. Number 3. Page one. Page two.
AI search requires a slightly different mindset.
Think in terms of:
- likelihood rather than a fixed ranking
- customer needs rather than isolated keywords
- repeated observations rather than one search
- relevance rather than visibility for its own sake
And when that starts to feel complicated, come back to the Business Card Box.
For every customer question, imagine the AI assembling a box of potentially relevant businesses.
Some businesses appear to have stronger representation than others.
Each search is another draw.
Your job is not to control the draw.
Your job is to give the AI stronger reasons to understand when your business is relevant, make sure you belong in the right boxes, and improve the likelihood that your card is drawn when the right customer asks the right question.
One AI search tells you which cards were drawn. Repeated testing starts to tell you how many cards you might have in the box.