It is easy to measure the wrong thing.
If you want to understand how visible your business is in AI search, you might start by entering a broad query related to your market, seeing whether your business appears, and recording the result.
But what if that query bears little resemblance to the searches that could actually bring you a customer?
A business could have excellent visibility for a broad informational query and gain very little from it. Another business might hardly appear for the broad query at all, but consistently appear once the searcher describes the particular problem that business is especially well suited to solve.
We ran a simple experiment in Google AI Mode to explore this.
The results suggest that AI recommendation visibility only becomes commercially useful when you measure it against queries that represent the customers you actually want to attract.
Start With the Customer, Not the Visibility Score
Imagine a small accountancy firm.
Ask the owner who they work with and the initial answer might be:
“Small businesses.”
But that probably isn’t the whole story.
After thinking about their actual clients, they might realise that most are:
- owner-managed limited companies;
- based within a particular area;
- relatively small;
- looking mainly for reliable compliance and practical tax support;
- using particular accounting software;
- and wanting predictable monthly costs rather than complex advisory services.
That is a much clearer business proposition.
And it changes the type of AI queries that are worth tracking.
“Recommend an accountant” tells us very little.
“Recommend an accountant for a small construction limited company in Bristol that regularly uses subcontractors” describes a much more recognisable prospective customer.
We wanted to see whether adding that kind of commercial detail actually changed the businesses AI recommended.
Our Test
We used three progressively more specific queries.
Query 1
Can you recommend an accountant for a small limited company in Bristol?
We ran the identical query three times.
Across those three responses, nine different accountancy firms were recommended.
Only one firm, Nasa Accountants, appeared in all three responses.
Others appeared once or twice.
Already this demonstrated one important point: a single AI search would have given a very incomplete picture of recommendation visibility.
If an accountancy firm checked once and appeared, it might assume it had strong visibility.
If it repeated the test a minute later and disappeared, the conclusion could look very different.
Then We Added One Piece of Information
Our second query was:
Can you recommend an accountant for a small construction limited company in Bristol?
The only additional information was the industry.
The result changed dramatically.
Across all three runs, Google AI Mode recommended exactly the same four firms:
| Firm | Run 1 | Run 2 | Run 3 |
|---|---|---|---|
| Octane Accountants | Yes | Yes | Yes |
| Streets Steele | Yes | Yes | Yes |
| RSBC Accountants | Yes | Yes | Yes |
| Nasa Accountants | Yes | Yes | Yes |
Octane appeared first in every response.
The broad query had produced nine different recommendations across three searches.
Adding construction produced a far more consistent set.
That does not prove that more specific searches always produce more stable results. Three runs are a small practical test, not a scientific sample.
But it demonstrated something worth investigating.
The customer had told AI more about who they were, and the group of businesses being recommended changed.
Then We Described How the Business Operates
Our third query was:
Can you recommend an accountant for a small construction limited company in Bristol that regularly uses subcontractors?
Again, we ran it three times.
The recommendations changed again.
| Firm | Run 1 | Run 2 | Run 3 | Recommendation visibility |
|---|---|---|---|---|
| Streets Steele | Yes | Yes | Yes | 100% |
| EasyAccounts and Tax | Yes | Yes | Yes | 100% |
| Dunkley’s | Yes | No | Yes | 67% |
| Octane Accountants | No | Yes | Yes | 67% |
| Tax Savers Direct | Yes | Yes | No | 67% |
This time the additional information wasn’t simply another keyword.
It described an actual operational characteristic of the prospective customer’s business.
The customer wasn’t merely saying:
“I work in construction.”
They were saying:
“I work in construction and regularly use subcontractors.”
That distinction matters because it introduces issues such as the Construction Industry Scheme, subcontractor administration and related accounting requirements.
AI now had more information with which to decide which firms appeared relevant.
The Same Business Can Have Very Different AI Visibility
This was perhaps the most useful lesson from the test.
Consider Octane Accountants.
Its recommendation visibility was:
- 33% for the broad small limited company query;
- 100% once construction was specified;
- 67% once regular use of subcontractors was added.
Now consider EasyAccounts and Tax:
- 0% for the broad query;
- 0% for the construction query;
- 100% once we specified regular use of subcontractors.
If EasyAccounts had tested only:
“Recommend an accountant for a small limited company in Bristol”
it could have concluded that its AI visibility was terrible.
Yet when the prospective customer more closely resembled the type of customer its website specifically talks about, it appeared in every test.
That is why an overall AI visibility percentage can be misleading.
The important question isn’t simply:
What is our AI share of voice?
It is:
What is our AI visibility for the customer situations that actually matter to our business?
A Simple Way to Calculate Recommendation Visibility
For a small manual test, the calculation can be kept very simple:
Number of responses in which the business was recommended ÷ total number of runs × 100
So if a business appears in three out of five responses:
3 ÷ 5 × 100 = 60% recommendation visibility
There is an important distinction here.
This is not traditional market share.
Several businesses can appear in the same AI answer, so the percentages do not need to add up to 100%.
If four businesses appear in all five test runs, all four could have 100% recommendation visibility for that query.
The percentage simply tells you how consistently a business appeared within the sample you tested.
Then We Looked at the Websites
The next question was obvious.
Were the businesses being recommended actually publishing information that corresponded with our increasingly specific searches?
We inspected three of the firms that appeared repeatedly.
Octane Accountants
Octane Accountants describes itself as a Bristol chartered accountancy firm working with SMEs and says it primarily works with small limited companies, sole traders and company directors.
More interestingly, it has an entire page dedicated to Construction Accountants in Bristol.
That page explicitly discusses construction companies, limited companies, subcontractor payments, CIS compliance, monthly CIS returns, payroll for employees and subcontractors, VAT, retentions and project cashflow.
Remember what happened in our test:
Broad small limited company query: 1/3
Construction query: 3/3
We cannot say the construction page caused Octane to be recommended.
Google’s retrieval and recommendation systems involve factors we cannot see.
But we can make a much narrower observation:
Once our prospective customer described themselves as a construction company, Octane became consistently visible — and Octane’s website contains a substantial dedicated section clearly explaining its relevance to exactly that type of customer.
That is useful evidence for a business thinking about how clearly its own proposition is communicated online.
EasyAccounts and Tax
EasyAccounts gave us an even more interesting example.
Its visibility went:
Broad: 0/3
Construction: 0/3
Construction + regularly uses subcontractors: 3/3
When we inspected EasyAccounts and Tax, its main navigation included a specific “CIS Contractors & Subcontractors” section alongside its limited company and other customer-type pages.
Its dedicated CIS accountants page goes much further.
It explicitly discusses contractor and subcontractor registration, verifying subcontractors before payment, monthly CIS returns, CIS deductions, payroll for subcontractors, Xero and FreeAgent, construction cashflow and fixed monthly pricing.
Again, this does not tell us why Google selected EasyAccounts.
But the association is striking.
EasyAccounts did not appear when we merely said “construction”.
It became consistently visible when we introduced subcontractors — an area its website describes in considerable detail.
Streets Steele
Streets Steele produced another useful pattern:
Broad query: 0/3
Construction: 3/3
Construction + subcontractors: 3/3
The Streets Steele website describes a Bristol accountancy firm working with SMEs and limited companies, with services including CIS and Xero-based cloud accounting. It also states that fixed-fee packages are available.
Its dedicated construction accounting page discusses construction businesses, builders, tradespeople, contractors, subcontractors, CIS, VAT, bookkeeping, cashflow and limited companies.
There is also a separate Payroll and CIS service page explicitly aimed at those working in construction as contractors or subcontractors, covering HMRC submissions and CIS statements.
Streets Steele therefore gives us a particularly clear example of a business whose website contains detailed information covering both the industry and the specific operating circumstances contained in our queries.
This Is Not Evidence of an AI Ranking Hack
There is an important limit to what this experiment proves.
We did not demonstrate:
Create a construction page and Google AI Mode will recommend your accounting firm.
Nor did we establish that particular words, page structures or content formats caused any recommendation.
What we found is simpler.
As our prospective customer’s requirements became more specific, the recommended businesses changed.
And when we subsequently inspected several businesses that became consistently visible, their websites clearly contained information relevant to those increasingly specific requirements.
That is association, not proof of causation.
But from the perspective of a small business, there is also a common-sense question:
If your website does not clearly say that you provide a particular service, work with a particular type of customer or solve a particular problem, what evidence are you giving either a human visitor or an AI system that you are a good match?
Your Website May Have an Information Problem, Not an AI Problem
This may become one of the more important implications of AI search.
Many business websites are extremely vague.
They tell visitors that the company provides:
“Tailored solutions designed around the needs of our clients.”
But they don’t necessarily explain:
- who those clients actually are;
- what size businesses they usually work with;
- which industries they understand;
- where they operate;
- whether services can be delivered remotely;
- which software they support;
- typical costs;
- turnaround times;
- what is included;
- what problems they regularly solve;
- or what happens after someone makes contact.
Those facts may already be obvious inside the business.
They may simply never have been clearly published.
AI search could make that weakness more important because a searcher can now describe a much fuller situation.
Instead of searching:
accountant Bristol
they can ask:
I run a small family-owned construction limited company in Bristol, use Xero, pay around ten subcontractors each month and mainly want someone to handle our accounts, payroll and CIS for a predictable monthly fee. Which accountants look suitable?
That gives the AI system a lot of criteria to work with.
A business website that clearly addresses those criteria gives the system information it can potentially use.
A website that says little more than “accounts, tax and business advice” does not.
There is still no guarantee which business will be recommended.
But providing accurate and useful information about your proposition seems a far more sensible starting point than searching for supposed AI visibility tricks.
How a Small Business Can Run This Test
You do not need specialist software to begin understanding this.
Start with perhaps five to ten commercially important customer situations.
For each one, ask:
If someone entered this query and my business appeared, would there be a realistic reason for that person to visit my website or contact me?
If the answer is no, question whether the query is worth tracking.
Then build the query progressively.
For example:
Broad
Recommend an accountant for a small limited company in Bristol.
Industry
Recommend an accountant for a small construction limited company in Bristol.
Operating situation
Recommend an accountant for a small construction limited company in Bristol that regularly uses subcontractors.
You could then add other genuine purchasing requirements where relevant:
- uses Xero;
- needs payroll;
- wants fixed monthly pricing;
- requires remote meetings;
- has a particular budget;
- needs a fast turnaround;
- operates in a regulated industry;
- needs a particular technical capability.
The exact criteria will depend entirely on your business.
That is the point.
Run Every Query More Than Once
Our broad query produced nine different businesses from just three identical searches.
Testing once would therefore have produced a poor representation of the recommendation landscape.
For a simple manual test, running the same query three to five times gives you a much better starting point.
Record:
| Query | Runs | Your business appearances | Recommendation visibility | Other businesses appearing |
|---|---|---|---|---|
| Query A | 5 | 2 | 40% | A, B, C |
| Query B | 5 | 4 | 80% | B, D |
| Query C | 5 | 0 | 0% | E, F, G |
Do not obsess over the percentage as though it were a precise scientific measurement.
The value is in the pattern.
Which situations make you more visible?
Which make you disappear?
Which competitors repeatedly appear instead?
And most importantly: what distinguishes those situations?
Then Audit the Information on Your Website
Take a query where you believe your business should be highly relevant.
Break it into requirements.
For example:
construction accountant in Bristol using Xero, experienced with subcontractors and offering fixed monthly fees
Then ask whether your website clearly establishes each point.
| Customer requirement | Clearly stated on website? |
|---|---|
| Construction expertise | Yes / No |
| Bristol | Yes / No |
| Xero | Yes / No |
| Subcontractor/CIS experience | Yes / No |
| Fixed monthly fee | Yes / No |
This is not about stuffing a page with keywords.
It is about making sure important facts about your business actually exist in a form people can find and understand.
If you genuinely offer fixed pricing but your website never says so, consider whether that information would help customers.
If you specialise in a particular industry but mention it only once in an old blog post, perhaps the proposition could be communicated more clearly.
If you only work within 20 miles of your office but never explain your service area, that is useful information missing from the site.
If turnaround time is critical to customers, tell them how the service works.
These improvements are useful regardless of whether an AI system rewards them.
They make the website more helpful to prospective customers too.
Good AI Visibility May Mean Being the Right Answer Less Often
There is another counterintuitive lesson here.
A specialist business may not need to appear for every broad query.
EasyAccounts was absent from all three of our broad searches.
But once our hypothetical customer regularly used subcontractors, it appeared every time.
If CIS contractors and subcontractors represent an important part of that firm’s target market, that may be much more commercially valuable than appearing occasionally for a generic “accountant for a small company” query.
This suggests businesses should be wary of AI visibility reports built mainly from broad, high-volume prompts.
A large visibility number can look impressive.
But if those prompts do not represent people who might realistically become customers, the number may have little commercial meaning.
The Question to Ask Is “When Should We Be the Answer?”
Traditional visibility thinking can encourage a business to ask:
How do we appear everywhere?
AI search may make a different question more useful:
Under what circumstances should our business be a particularly good answer?
That requires the business to understand its own proposition.
Who are your best customers?
What do they have in common?
What problems bring them to you?
What requirements matter when they choose a provider?
What information would somebody need before deciding that your business fits their situation?
Those questions are not really AI questions.
They are business questions.
But answering them could make AI visibility testing considerably more useful.
Final Thought
Our experiment started as a simple attempt to calculate AI recommendation share of voice.
It ended up highlighting something more important.
AI visibility should not be measured independently of customer intent.
A business does not necessarily need to be visible for every broad query in its market.
It needs to understand whether it appears when the searcher’s circumstances closely match the customers it is genuinely equipped and interested in helping.
That means choosing commercially meaningful queries, testing them repeatedly, examining who appears, and then checking whether your own website clearly communicates the information relevant to those situations.
There is no guarantee that publishing better information will make an AI system recommend your business.
But before looking for complicated AI optimisation techniques, there is a much simpler question worth asking:
Does our website actually give people — and the systems helping them search — enough information to understand when we are the right business for the job?
That may be one of the most useful places to start.