AI search lets people describe what they need in much more detail than a traditional short search query.
Someone no longer has to search simply for:
accountant Bristol
They can ask:
Can you recommend an accountant for a small construction limited company in Bristol that regularly uses subcontractors?
Those two searches are looking for broadly the same service.
But the second tells the AI much more about the customer.
That raises an interesting question:
As a customer describes their circumstances more precisely, does AI start recommending businesses whose websites more closely reflect those particular needs?
We decided to test it.
We started with accountants.
Then, to make sure we weren’t simply seeing something peculiar to accountancy, we repeated the basic experiment with commercial cleaning companies.
The results do not prove how Google AI Mode chooses which businesses to recommend.
But the pattern we found was striking enough that we think it deserves much more investigation.
Experiment 1: Finding an Accountant in Bristol
We began with a fairly broad but commercially realistic query:
Can you recommend an accountant for a small limited company in Bristol?
We ran exactly the same query three times.
Across those three responses, Google AI Mode recommended nine different firms.
Only one appeared every time.
That immediately demonstrated something we have seen repeatedly in AI search testing: a single search is a very weak way of measuring visibility.
A business could appear in one response and disappear in the next.
But this experiment was really about what happened when we started describing the customer more precisely.
We Added One Detail: The Industry
Our second query was:
Can you recommend an accountant for a small construction limited company in Bristol?
The only major change was the word construction.
The results changed dramatically.
Across three runs, the same four firms were recommended every time:
| 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 |
In fact, Octane appeared first in every run.
The broad small-company query had produced nine different firms.
Adding the customer’s industry produced a remarkably stable group.
Three runs is obviously a small sample. We are not suggesting that more specific searches will always behave this way.
But the change was large enough to make the next question obvious.
What happens if we describe not just the industry, but how the customer’s business actually operates?
We Added Another Customer Circumstance
Our third query was:
Can you recommend an accountant for a small construction limited company in Bristol that regularly uses subcontractors?
Again we ran the identical query three times.
The recommendation set changed again:
| Firm | Run 1 | Run 2 | Run 3 |
|---|---|---|---|
| Streets Steele | Yes | Yes | Yes |
| EasyAccounts and Tax | Yes | Yes | Yes |
| Dunkley’s | Yes | No | Yes |
| Octane Accountants | No | Yes | Yes |
| Tax Savers Direct | Yes | Yes | No |
The most interesting changes were not necessarily the firms that disappeared.
They were the firms that suddenly became highly visible.
EasyAccounts was particularly interesting
EasyAccounts appeared:
- 0 out of 3 times for the broad limited-company query;
- 0 out of 3 times for the construction query;
- 3 out of 3 times once we said the construction company regularly used subcontractors.
If EasyAccounts had measured its AI visibility only using:
accountant for a small limited company in Bristol
it might have concluded that its visibility was poor.
But for a much more specific prospective customer, its visibility in our small test was perfect.
That raised another question.
What does the EasyAccounts website actually say it specialises in?
Importantly, we had not looked at these firms’ websites before designing our queries.
The businesses emerged from the test first.
We inspected their websites afterwards.
What We Found on the Accountants’ Websites
The results were fascinating.
Octane Accountants
Octane Accountants appeared only once in our three broad small-company searches.
Once we specified construction, it appeared first in all three runs.
Afterwards, we discovered that Octane has a dedicated page specifically for construction accountants in Bristol.
The page discusses issues including:
- construction companies;
- limited companies;
- subcontractors;
- CIS;
- monthly CIS returns;
- retentions;
- staged payments;
- payroll;
- VAT;
- project cashflow.
We cannot say that this page caused Octane to be recommended.
But the association is difficult to ignore.
We progressively described a construction customer, Octane became substantially more visible, and only afterwards did we discover a substantial part of its website devoted to that exact customer group.
Streets Steele
Streets Steele produced an equally interesting pattern:
- 0/3 for the broad query;
- 3/3 once construction was specified;
- 3/3 when subcontractors were added.
Its website has a dedicated construction accounting section discussing builders, construction companies, tradespeople, contractors, subcontractors, CIS, VAT, cashflow and other sector-specific issues.
It also has separate information covering payroll and CIS.
Again, we discovered this after the firm had repeatedly emerged in the more specific searches.
EasyAccounts and Tax
EasyAccounts and Tax gave us perhaps the clearest example.
Its website specifically identifies CIS contractors and subcontractors as a customer group.
Its dedicated CIS accountants page discusses:
- contractors and subcontractors;
- subcontractor verification;
- CIS deductions;
- monthly CIS returns;
- payroll;
- construction businesses;
- Bristol;
- cloud accounting;
- fixed fees.
And this was the business whose visibility went:
0/3 → 0/3 → 3/3
as soon as we introduced regular use of subcontractors.
We still cannot conclude that Google’s recommendation was caused by that page.
But it gave us a hypothesis worth testing in a completely unrelated industry.
Experiment 2: Commercial Cleaning in Leeds
We started again with a broad commercial query:
Can you recommend a commercial cleaning company for a small business in Leeds?
We ran it three times.
The results were even more variable than the accountancy test.
Across the three runs, 10 different cleaning companies appeared.
Not one appeared in all three responses.
Again, a single visibility check would have told a business very little.
Google AI Mode also repeatedly asked us what type of business premises we had.
So we answered that question by making only one substantial change to our prompt.
Small Business Became Small Dental Practice
Our new query was:
Can you recommend a commercial cleaning company for a small dental practice in Leeds?
Nothing else materially changed.
But the recommendations changed enormously.
Across all three runs, four companies appeared every time:
| Company | Run 1 | Run 2 | Run 3 |
|---|---|---|---|
| Ashworth Commercial Cleaning | Yes | Yes | Yes |
| It’s Clean | Yes | Yes | Yes |
| Top Notch Cleans | Yes | Yes | Yes |
| Spark Cleaning Services | Yes | Yes | Yes |
Compare that with our broad commercial-cleaning test:
- 10 businesses across three runs
- no business appearing every time
Then:
- one extra piece of customer information;
- four businesses appearing 3/3.
Again, this does not prove that specificity automatically creates more stable AI recommendations.
But we had now seen a strikingly similar pattern in two completely different industries.
So we repeated the second part of our methodology.
We inspected some of the websites after the recommendations had emerged.
Ashworth Commercial Cleaning
Ashworth Commercial Cleaning did not appear once in our three broad commercial-cleaning searches.
It appeared in all three dental-practice searches.
When we subsequently inspected its website, we found a dedicated page for medical and dental practice cleaning.
That page explicitly discusses:
- dental practices;
- Leeds and West Yorkshire;
- medical and healthcare premises;
- infection-control procedures;
- colour-coded cleaning systems;
- prevention of cross-contamination;
- COSHH procedures;
- treatment areas;
- documented cleaning reports;
- audit trails.
In other words, once the customer became a dental practice, a business whose website had a substantial proposition aimed specifically at dental and medical practices suddenly became consistently visible.
Again:
query first → recommendation second → website inspection third.
We did not design the query around Ashworth’s website.
It’s Clean
It’s Clean showed the same pattern:
Broad commercial cleaning: 0/3
Dental practice: 3/3
Afterwards we found a dedicated dental-practice cleaning page.
The site explicitly states that it cleans dental practices in Leeds and the surrounding area and discusses:
- dental treatment rooms;
- CQC requirements;
- healthcare cleaning;
- out-of-hours cleaning;
- more than 20 years’ dental-practice cleaning experience.
That out-of-hours point was particularly interesting because Google AI Mode itself repeatedly highlighted flexible and out-of-hours cleaning when describing the firm.
The business had actually published that information.
Top Notch Cleans Also Taught Us Something Else
Top Notch Cleans appeared in all three dental runs.
While checking the business, we initially looked at a similarly named company in London.
It was the wrong business.
That exposed another issue businesses and agencies need to consider when tracking AI visibility.
A company name may not be enough to identify an entity reliably.
If two businesses have very similar names, a visibility-monitoring system needs to make sure it is recording the correct company and domain.
Otherwise the measurement itself could be wrong.
For serious AI visibility monitoring, business identity matters.
What Do These Two Experiments Actually Show?
We need to be careful here.
We have not proved that creating highly specific service pages causes Google AI Mode to recommend a business.
We have not reverse-engineered Google’s ranking algorithm.
And our sample sizes are deliberately small.
What we observed was this:
When we progressively described the circumstances of a prospective customer more precisely, the businesses recommended by Google AI Mode changed substantially.
Then:
When we subsequently inspected several businesses whose visibility became much stronger for those more specific searches, we repeatedly found websites containing substantial information closely aligned with those customer circumstances.
We observed this with:
- construction accountants;
- CIS/subcontractor accountants;
- dental-practice cleaners.
And we observed it in two unrelated industries.
That is not proof of causation.
But it is enough to justify taking the idea seriously and testing it further.
The Bigger Idea: Customers Can Now Describe Themselves
This may be one of the biggest differences between traditional search and conversational AI search.
A traditional search journey might begin:
commercial cleaner Leeds
or:
accountant Bristol
The user then visits websites and performs much of the filtering themselves.
AI allows the customer to move some of that filtering into the initial search.
They can say:
I run a small construction limited company in Bristol, regularly pay subcontractors and need an accountant familiar with CIS.
Or:
I run a small dental practice in Leeds and need a cleaning company experienced with clinical environments.
Those queries contain far more information about who the customer is.
That potentially gives the AI system more information with which to distinguish between providers.
And that changes the question businesses may need to ask about their websites.
Does Your Website Explain When You Are the Right Business?
Many websites communicate at an extremely generic level.
An accountant might say:
We help businesses with accounts, tax, payroll and bookkeeping.
A cleaner might say:
We provide professional commercial cleaning solutions tailored to your needs.
Neither statement tells us very much.
Compare that with a website that clearly explains:
We work with small construction companies in Bristol, including companies using subcontractors, and handle CIS returns, verification, payroll and construction-specific bookkeeping.
Or:
We clean dental practices across Leeds, work around practice opening hours and provide documented cleaning procedures designed for healthcare environments.
Those descriptions help a prospective customer understand whether the company is relevant.
And, logically, they also provide much richer information for a search system trying to answer a detailed customer query.
That still does not guarantee visibility.
But if important information about your business never appears on your website, there is less explicit information available from which either a person or an AI system can understand that you meet those requirements.
Your Business May Be More Specialised Than Your Website Suggests
This could expose a common problem.
Ask an accountant who their customers are and they might initially say:
everyone.
But after looking at their actual client base, perhaps they mainly work with:
- family-owned limited companies;
- within 20 miles of their office;
- with fewer than 20 staff;
- using Xero;
- needing straightforward compliance;
- wanting occasional practical tax help;
- and preferring affordable fixed monthly fees.
That is the actual proposition.
If none of that is clearly communicated on the website, the online version of the business may look far more generic than the real business.
The same could apply to thousands of companies.
A web designer may primarily work with Shopify retailers.
A cleaner may have substantial experience with medical premises.
A solicitor may predominantly help owner-managed businesses with employment disputes.
A software company may be particularly well suited to five-person recruitment agencies.
The expertise exists.
But does the website make it obvious?
This Is Not About Creating Pages for Every Possible AI Prompt
There is an obvious danger in taking this idea too far.
The answer is not to generate hundreds of thin pages designed around every possible variation of a query.
That would miss the point.
The underlying task is much more fundamental:
Understand your actual customer proposition and communicate it properly.
Ask:
- Who do we genuinely help?
- Which customers are particularly well suited to us?
- What problems bring those customers to us?
- What circumstances make our service relevant?
- Which industries do we actually understand?
- Where do we operate?
- What software or systems do we support?
- What is included?
- What does it cost, where appropriate?
- How quickly can we deliver?
- What don’t we provide?
- How does someone start working with us?
Those are useful questions even if AI search disappeared tomorrow.
They make a website more useful to people.
AI Visibility May Be About “When”, Not Just “Whether”
This changes the way we think about AI visibility.
Instead of asking:
Does AI recommend our business?
perhaps the more useful question is:
When does AI recommend our business?
EasyAccounts was invisible for our generic accountancy query.
But it became consistently visible once our customer regularly used subcontractors.
Ashworth was invisible in our generic commercial-cleaning sample.
But it became consistently visible once the customer became a dental practice.
Those businesses might not need maximum visibility for every broad query.
What matters commercially is whether they appear when the person searching resembles the customer they actually want.
This Also Changes How Share of Voice Should Be Measured
Suppose an AI visibility platform reports:
Your recommendation share of voice is 12%.
That sounds precise.
But 12% of what?
If the underlying prompts have little commercial relevance to the business, the percentage may not tell you much.
A specialist company could have weak visibility across broad industry prompts but excellent visibility for the much narrower situations where it genuinely has an advantage.
That might be a much better commercial outcome.
So the quality of the query set is not just a technical detail.
It determines whether an AI visibility metric means anything.
Recommendation Visibility Is Not the Only Thing to Monitor
Our testing also uncovered another issue.
AI sometimes described recommended businesses using claims we could not clearly verify on the businesses’ own websites.
For example, some dental-cleaning answers mentioned HTM 01-05 awareness even where we could not find equivalent wording on the provider’s website.
That suggests businesses may eventually need to monitor at least three separate things:
1. Recommendation visibility
Does AI recommend the business?
2. Citation visibility
Does AI use the business’s website as a source?
3. Representation accuracy
When AI discusses the business, is what it says actually accurate?
A business could be highly visible and still have a problem if the AI misrepresents what it offers.
What Small Businesses Can Do Now
You do not need to chase an “AI optimisation hack”.
Start with something much more practical.
Define a handful of customer situations where you genuinely believe your business should be relevant.
Move from broad to specific.
For example:
commercial cleaner in Leeds
then:
commercial cleaner for a dental practice in Leeds
then perhaps:
dental-practice cleaner in Leeds that can clean outside opening hours
Then test each query several times.
Record who appears.
Look at which businesses become more or less visible as the customer circumstances change.
Only afterwards inspect the websites.
Ask:
What information do the consistently recommended businesses make available that helps explain who they serve and the circumstances in which they are relevant?
Then turn the same question back onto your own site.
Not:
What keywords are we missing?
But:
What useful information about our actual proposition are we failing to communicate?
That is a very different exercise.
Where We Go From Here
Two experiments are not enough to establish a general rule.
So we will continue testing this idea across unrelated industries.
If we repeatedly see the same pattern — businesses becoming more visible as a searcher’s circumstances increasingly match propositions clearly communicated on their websites — the finding becomes much more significant.
We also want to explore where it fails.
There will almost certainly be searches where highly relevant specialist businesses do not appear.
Understanding those failures may teach us as much as the successful matches.
For now, however, our results suggest an interesting possibility.
Final Thought
AI search may be changing online discovery from:
What service do you need?
towards:
Who are you, what is your situation, and what exactly do you need?
If that continues, businesses may need to become much clearer about the other side of that equation:
Who exactly do we help, and under what circumstances are we particularly useful?
Our experiments have not proved that clearly communicating those things will make Google AI Mode recommend a business.
But twice now, across two unrelated industries, we have made the prospective customer more specific and watched specialist businesses emerge.
When we then inspected those businesses, we repeatedly found websites that clearly described the very customer situations our queries had introduced.
That is not proof.
But it is a finding worth taking seriously.
And it may point towards a surprisingly simple principle for AI-era business websites:
Don’t just tell the internet what your business does. Make it clear when your business is the right fit.