In an earlier experiment, we found that making a customer’s situation more specific could substantially change the businesses or products recommended by Google AI Mode.
A broad request for an accountant produced one set of firms.
Adding the fact that the customer ran a construction company produced another.
Adding that the company regularly used subcontractors changed the recommendations again.
We then saw a similar pattern when a generic commercial-cleaning query became a search for a cleaner for a dental practice.
And we reproduced it again with CRM software.
That raised a more difficult question.
Does simply adding more detail change AI recommendations, or does the detail need to materially change what the customer actually needs?
We decided to test that.
Starting With a Broad CRM Query
We began with:
Can you recommend CRM software for a small business?
We ran the identical query three times.
The result was extremely stable.
Four products appeared in every run:
| CRM | Run 1 | Run 2 | Run 3 |
|---|---|---|---|
| HubSpot CRM | Yes | Yes | Yes |
| Pipedrive | Yes | Yes | Yes |
| Zoho CRM | Yes | Yes | Yes |
| Capsule CRM | Yes | Yes | Yes |
This was very different from some of our local-business tests, where broad queries produced a large amount of variation.
Here, Google AI Mode repeatedly returned a familiar group of general-purpose CRM products.
That was not particularly surprising.
We had told it almost nothing about the customer other than that they were a small business.
So recommending established general CRM platforms was a reasonable response.
The more interesting part came next.
We Identified the Type of Business
We changed the query to:
Can you recommend CRM software for a small recruitment agency?
Again, we ran it three times.
The recommendation landscape changed dramatically.
Recruit CRM, Giig Hire, Loxo and Zoho Recruit appeared in all three runs.
HubSpot and Pipedrive, which had appeared in every broad CRM search, disappeared entirely.
| Product | Generic small business | Small recruitment agency |
|---|---|---|
| HubSpot | 3/3 | 0/3 |
| Pipedrive | 3/3 | 0/3 |
| Zoho CRM / Zoho Recruit | 3/3 | 3/3 |
| Capsule CRM | 3/3 | 2/3 |
| Recruit CRM | 0/3 | 3/3 |
| Giig Hire | 0/3 | 3/3 |
| Loxo | 0/3 | 3/3 |
The AI had not simply changed the names it recommended.
It changed the criteria it used to judge suitability.
The broad CRM searches focused on things such as:
- sales pipelines;
- email marketing;
- general contact management;
- simplicity;
- price;
- integrations.
Once the customer became a recruitment agency, the answer started focusing on:
- applicant tracking systems;
- candidate management;
- client relationship management;
- CV parsing;
- LinkedIn sourcing;
- job boards;
- email outreach;
- permanent, contract and temporary placements.
The identity of the customer had changed what AI considered important.
We Checked One of the Products Afterwards
Only after running the queries did we inspect Recruit CRM.
That sequence matters.
We did not read Recruit CRM’s website first and then design a query to match it.
The product emerged from the experiment first.
When we subsequently inspected the website, we found that Recruit CRM explicitly describes itself as an ATS and CRM designed for recruitment agencies.
Its proposition includes candidate management, client management, recruitment pipelines, CV parsing, sourcing and recruitment-specific workflows.
So Recruit CRM’s visibility went:
0/3 for generic small-business CRM
to:
3/3 for small recruitment agency CRM
and afterwards we found a website clearly built around that exact type of customer.
That was consistent with patterns we had already seen in our accountant and commercial-cleaning experiments.
But we wanted to go one level deeper.
What Happens When the Business Model Changes?
Not all recruitment agencies operate in the same way.
An agency that mainly makes permanent placements has different operational requirements from one supplying large numbers of temporary workers.
So we changed the query again:
Can you recommend CRM software for a small recruitment agency that mainly places temporary staff?
This time, Google AI Mode started prioritising a different set of functions:
- shift scheduling;
- candidate availability;
- compliance checking;
- digital timesheets;
- mobile worker access;
- payroll;
- pay-and-bill;
- rota management.
And the products changed again.
Across three runs:
| Product | Temp Run 1 | Run 2 | Run 3 |
|---|---|---|---|
| Vincere | Yes | Yes | Yes |
| Firefish | Yes | Yes | No |
| PrimePRO | Yes | Yes | No |
| Recruit CRM | Yes | No | No |
| Voyager Infinity | No | Yes | No |
| Bullhorn | No | No | Yes |
| Tracker | No | No | Yes |
The standout result was Vincere.
For a general small recruitment agency:
0/3
For a small recruitment agency mainly placing temporary staff:
3/3
That looked very similar to the visibility shifts we had seen elsewhere.
Then We Inspected Vincere
Again, we looked at the website only after the recommendation pattern had emerged.
Vincere separates its proposition by the way recruitment businesses operate.
It includes dedicated solutions for:
- temporary recruitment;
- contract recruitment;
- permanent recruitment;
- executive search.
It also has dedicated material for small and startup recruitment agencies.
Most importantly for our test, its temporary-recruitment proposition specifically discusses capabilities such as:
- shift scheduling;
- candidate availability;
- digital timesheets;
- compliance;
- payroll;
- billing;
- mobile access;
- temporary-worker workflows.
Those were very close to the requirements Google AI Mode began emphasising after we added the phrase:
mainly places temporary staff
Once again, that does not prove that the website caused the recommendation.
But the pattern was striking.
The product went from invisible in our general recruitment-agency sample to appearing in every temporary-staffing run, and its published proposition closely matched the operational needs introduced by that extra piece of information.
But Was This Simply Because the Query Was Longer?
That was the next question.
Perhaps adding almost any detail would make Google AI Mode reconsider the recommendations.
If so, the finding would be much less interesting.
We needed some kind of comparison.
So we created another query:
Can you recommend CRM software for a small recruitment agency that has been trading for five years?
This still made the customer more specific.
But being five years old does not obviously require a fundamentally different CRM.
We ran that query three times.
The Result Was Very Different
Across the three five-year queries, the recommendation set stayed broadly within the same group we had seen for a normal small recruitment agency.
| Product | General recruitment | Five years trading |
|---|---|---|
| Recruit CRM | 3/3 | 3/3 |
| Recruiterflow | 2/3 | 3/3 |
| Loxo | 3/3 | 2/3 |
| Firefish | 1/3 | 3/3 |
| Vincere | 0/3 | 0/3 |
| PrimePRO | 0/3 | 0/3 |
The extra information was not ignored.
Google AI Mode interpreted a five-year-old agency as potentially having:
- an established database;
- historic candidate and client records;
- migration needs;
- more emphasis on automation;
- business-development requirements;
- a need to scale beyond spreadsheets or simpler systems.
So the explanation changed.
But the fundamental type of product being recommended did not change nearly as much.
That contrast is important.
“Temporary Staffing” Changed What the Software Needed to Do
Compare the two extra pieces of information.
Detail 1
The agency has been trading for five years.
That provides useful context.
It may imply that the business has more data, more established processes or greater migration needs.
But it does not completely redefine the job the CRM has to perform.
Detail 2
The agency mainly places temporary staff.
That changes the operational requirements substantially.
Now the software may need to handle:
- shifts;
- rotas;
- worker availability;
- timesheets;
- compliance;
- temporary-worker records;
- payroll;
- invoicing;
- pay-and-bill.
And that is where we saw the much larger recommendation shift.
Our Hypothesis Has Changed
When we began these experiments, the emerging idea was fairly simple:
More specific AI queries may produce more specific recommendations.
That still appears true in many of our tests.
But it now seems incomplete.
Our latest experiment suggests something more useful:
The kind of specificity may matter more than the amount of specificity.
Adding descriptive information does not necessarily change which solution is appropriate.
Adding information that materially changes the customer’s requirements can.
That is an important distinction.
Why This Matters for AI Visibility Testing
Suppose a software company wants to understand its AI visibility.
It could track hundreds of prompts containing slightly different wording and customer details.
But if those details do not alter the buying decision, the additional measurements may add very little value.
The more commercially useful questions may be based on the factors that actually determine suitability.
For recruitment software, those might include:
- permanent versus temporary recruitment;
- executive search;
- high-volume staffing;
- candidate sourcing;
- contractor management;
- timesheets;
- payroll;
- compliance;
- agency size;
- integrations.
For an accountant, they might include:
- industry;
- business structure;
- use of subcontractors;
- software;
- payroll requirements;
- complexity of tax affairs.
For a cleaner:
- office versus dental practice;
- healthcare setting;
- out-of-hours access;
- infection-control requirements;
- frequency;
- type of premises.
These are not arbitrary prompt variations.
They are decision-relevant customer characteristics.
This Could Change How Businesses Think About Website Content
There is also an important implication for websites.
Our earlier conclusion was that businesses should communicate more clearly who they help and what they do.
That still stands.
But our latest test suggests a more refined question:
Which facts about our customers actually determine whether we are the right solution?
A software company does not necessarily need pages aimed at:
recruitment agencies that are exactly five years old.
That is a specific description, but it may not meaningfully change what product the customer requires.
A page explaining:
temporary recruitment software with shift scheduling, compliance and digital timesheets
is different.
Those facts directly address the operating model that determines which software is suitable.
That distinction may be extremely important.
Decision-Relevant Information May Matter Most
This suggests that businesses should think particularly carefully about publishing information such as:
- who the product or service is for;
- what situations it handles;
- which problems it solves;
- which operating models it supports;
- which industries it understands;
- which integrations are available;
- what limitations exist;
- which services are included;
- pricing where appropriate;
- turnaround times;
- geographic restrictions;
- eligibility requirements.
These are the facts that help somebody decide:
Does this business actually fit my situation?
And they may also be the facts that give an AI system evidence with which to make the same distinction.
We cannot yet say how strongly any individual page or fact influences a recommendation.
But the principle itself is useful even without AI.
It makes websites better at helping prospective customers understand whether they have found the right provider.
More Information Is Not Necessarily Better Information
There is a temptation, whenever AI search is discussed, to conclude that businesses simply need to publish more content.
Our experiments point towards a different idea.
A website could contain enormous amounts of material and still fail to explain the things that actually determine whether the business is suitable.
A shorter site could potentially communicate the proposition much more effectively if it clearly explains:
We help this kind of customer, in these circumstances, with these problems, using these capabilities.
That may be much more valuable than producing another 100 generic articles.
The Testing Method Is Becoming Clearer Too
From these experiments, we can now create a more structured way of testing AI recommendation visibility.
Stage 1: Broad need
Start with the general product or service.
CRM for a small business
Stage 2: Customer type
Specify who the buyer actually is.
CRM for a small recruitment agency
Stage 3: Operational requirement
Add a characteristic that materially changes what the solution needs to do.
CRM for a small recruitment agency mainly placing temporary staff
Stage 4: Lower-relevance comparison
Add a true customer detail that provides context without fundamentally changing the solution.
CRM for a small recruitment agency that has been trading for five years
Then compare:
- which businesses or products appear;
- how stable the recommendations are;
- which capabilities AI begins prioritising;
- which businesses gain or lose visibility;
- and whether the businesses gaining visibility clearly publish propositions matching those requirements.
This is a much more useful test than simply asking the same broad question once.
What We Are Not Claiming
It is worth being very clear about the limits of the experiment.
We have not demonstrated that Google AI Mode uses a specific website page or phrase as a direct ranking factor.
We have not proved that businesses can cause themselves to appear simply by publishing highly specific service pages.
And the number of searches in these experiments is small.
What we have observed is a repeated pattern:
Customer information that materially changes the required solution can substantially change the AI recommendation set.
We have also repeatedly found, after the recommendations were generated, that businesses gaining visibility publish propositions closely aligned with those more specific customer requirements.
That is enough to justify further investigation.
It is not enough to claim causation.
The Bigger Question Has Become More Interesting
Our first question was:
Does making an AI search more specific change the recommendations?
The more useful question may now be:
Which details cause AI to reconsider what a suitable solution looks like?
That is a much better question for businesses.
Because customers are not merely collections of keywords.
They have circumstances.
They operate in particular ways.
They have constraints.
They have specific problems.
And those things determine which providers are genuinely appropriate.
Final Thought
AI search may make one distinction increasingly important:
descriptive detail versus decision-relevant detail.
A customer saying they have traded for five years tells us something about them.
A customer saying they run a temporary staffing agency tells us something much more important about what their software must actually do.
In our experiment, Google AI Mode treated those two kinds of information very differently.
That does not prove how its recommendation system works.
But it suggests that businesses thinking about AI visibility may need to stop asking:
How can we add more information to our website?
and start asking:
What information would actually help someone understand when our business is the right solution?
That may be the more important question.