Across 25 broad software searches, Google AI Mode repeatedly produced remarkably small groups of familiar brands.
Four of the five categories tested never moved beyond the same three products. The fifth category retained the same two leading products in every run.
The brands themselves were not especially surprising. In many cases, they resembled the answers an informed person might give if asked to name three prominent products in the category.
What was surprising was the lack of displayed alternatives.
The results suggest that broad AI searches may produce a small group of category-default brands, while more specific customer requirements give other products an opportunity to enter the response.
Why I Ran the Test
An earlier experiment asked Google AI Mode to recommend project-management software for a small UK marketing agency.
The same broad prompt was submitted five times, and every response returned:
- Monday.com
- Asana
- ClickUp
Not only were the same three products selected, but they appeared in the same order every time.
Project-management software is a large market, so I wanted to know whether this was an isolated result or part of a wider pattern.
I therefore repeated the process across four additional software categories:
- Email marketing
- CRM
- Website builders
- Accounting software
You can also read the earlier repeated product-search experiment that led into this investigation.
How the Test Was Conducted
The experiment contained five broad prompts. Each prompt was submitted five times, producing:
- 25 AI Mode responses
- Three recommendations per response
- 75 recommendation places
The project-management searches were conducted on 7 August 2026. The remaining four categories were tested on 8 August 2026.
Each run used the same process:
- Close the existing browser tab.
- Open a new tab.
- Return to Google.
- Paste the exact query into the search bar.
- Open the AI Mode response.
- Record the three-product shortlist before any follow-up interaction.
The searches were conducted on the same mobile device and Google account, with personalisation turned off.
I did not reopen an earlier result, edit an existing AI Mode conversation or ask it to regenerate a previous answer. Each result began with a newly submitted Google search.
This was a practical experiment rather than a scientific study. The searches still shared the same account, location, device and short testing period.
The Five Prompts
The project-management prompt was:
Which three project management software platforms should a small UK marketing agency shortlist? Present only the three names in a table.
The other four prompts followed the same structure:
Which three email marketing software platforms should a small UK business shortlist? Present only the three names in a table.
Which three CRM software platforms should a small UK business shortlist? Present only the three names in a table.
Which three website builder platforms should a small UK business shortlist? Present only the three names in a table.
Which three accounting software platforms should a small UK business shortlist? Present only the three names in a table.
The prompts did not specify a budget, industry, required integrations or specialist features. They were deliberately broad.
What I Counted
I recorded:
- The three products in each requested shortlist
- Which products appeared in all five runs
- Whether the complete shortlist changed
- Whether its order changed
- The total number of different products displayed
I counted only products included in the three-name shortlist. Websites used as supporting sources and products mentioned elsewhere in a response were excluded.
“First listed” should not automatically be interpreted as a formal ranking. The prompts requested shortlists rather than an ordered best-to-worst comparison.
The Complete Results
| Category | Products appearing in every run | Candidate pool | Ordering pattern |
|---|---|---|---|
| Project management | Monday.com, Asana, ClickUp | 3 | Identical in all five |
| Email marketing | Mailchimp, MailerLite, Brevo | 3 | One order change |
| CRM | HubSpot CRM, Zoho CRM | 5 | Third position rotated |
| Website builders | Shopify, Wix, Squarespace | 3 | Two different orders |
| Accounting | Xero, QuickBooks Online, Sage | 3 | One order change |
I treated a product as part of the stable core if it appeared in all five responses for its category.
These stable-core products occupied 70 of the experiment’s 75 recommendation places.
Only the CRM category introduced products that did not appear in every run. HubSpot remained first and Zoho remained second, while the third position rotated between:
- Salesforce Essentials
- Monday.com CRM
- Freshsales
Across the other four categories, not one additional product entered the displayed shortlist.
The Lack of Variation Was the Interesting Result
The experiment did not show that AI Mode always returns an identical order.
The order changed once for email marketing and accounting. Shopify and Squarespace exchanged positions within the website-builder results. CRM rotated its third recommendation.
However, shortlist membership was much more stable than its order.
The broad searches produced:
- A completely fixed three-product project-management group
- A completely fixed three-product email-marketing group
- A fixed two-product CRM core
- A completely fixed three-product website-builder group
- A completely fixed three-product accounting group
The project-management result therefore did not appear to be an isolated anomaly.
Across these five software markets, Google AI Mode displayed a strong and narrow output consensus.
Were These Simply the Market Leaders?
That is a plausible explanation, but this experiment did not measure sales, subscriber numbers or actual market share.
It is safer to describe the repeatedly selected products as category-default brands.
They may benefit from:
- Strong brand recognition
- Extensive documentation
- Frequent inclusion in comparison articles
- Large numbers of reviews
- Broad suitability
- Clear association with the category
- Regular comparison with one another
Consider the accounting result.
If an accountant were asked to name three packages commonly used by UK businesses, Xero, QuickBooks and Sage would be a conventional answer.
Similarly, someone asked to name three website builders might immediately think of Shopify, Wix and Squarespace.
In hindsight, three familiar recommendations may be more understandable than three less prominent alternatives.
The surprising part is not necessarily which brands appeared. It is how decisively AI Mode returned them across repeated searches.
An Important Limitation: I Asked for Only Three Names
The experiment measured displayed recommendations, not the size of AI Mode’s hidden candidate pool.
Google explains that AI Mode may use query fan-out to conduct several related searches across different subtopics and sources before synthesising its response. It could therefore encounter numerous products and supporting pages while displaying only the three names requested. Google Search Central
The results do not prove that AI Mode considered only three products.
They establish something narrower:
When instructed to display three products for a broad software query, AI Mode repeatedly compressed the answer into a small group of familiar brands.
Asking for five or ten recommendations may reveal a less stable secondary group. That would be a useful separate experiment.
Broad Prompts Leave AI Mode to Supply the Criteria
A broad question such as:
Which three accounting software platforms should a small UK business shortlist?
does not explain:
- The business’s industry
- Its size
- Its budget
- Whether it employs staff
- Whether payroll is needed
- Whether it carries stock
- Whether it requires project accounting
- Whether an external accountant needs access
AI Mode must therefore supply its own assumptions about what makes a generally suitable product.
With no distinguishing requirements, recommending three established and broadly capable platforms is a defensible response.
The broad prompt may therefore measure category prominence more than suitability for a particular customer.
Specific Requirements Produced Different Candidate Groups
Our earlier project-management testing showed what happened when the customer’s circumstances became more detailed.
| Prompt emphasis | Products that became prominent |
|---|---|
| Broad project management | Monday.com, Asana, ClickUp |
| Accurate billable-time tracking | Productive, Scoro, Harvest |
| Essential Xero integration | Scoro, Teamwork, Productive |
| Budget, team size, client access, time tracking and Xero | Avaza, ClickUp, Paymo and others |
The specific requirements did not merely rearrange the original broad shortlist. They caused different products to enter the response.
This suggests:
Broad searches may produce the category defaults. Specific searches give AI Mode a reason to move beyond them.
Broad Visibility and Commercial Visibility Are Different
Appearing for a broad query can still be valuable. It may place a brand into the customer’s initial consideration set or reinforce its familiarity.
However, it does not necessarily provide a reason to visit the company’s website.
A broad AI response might already answer the early question by supplying three familiar names. Commercial value may become stronger as the searcher adds the requirements that influence an actual decision.
| Visibility type | What it may indicate |
|---|---|
| Category visibility | The brand enters the initial conversation |
| Requirement visibility | The product fits a stated need |
| Decision visibility | The product survives budget and suitability checks |
| Action visibility | The response creates a reason to visit, enquire or buy |
This suggests another useful measurement: journey persistence.
Does the brand remain in consideration as the searcher moves from broad exploration towards action?
A product could dominate the broad category response but disappear once the customer specifies an integration, budget or specialist requirement.
Another product might be absent initially but enter when its particular strengths become relevant.
The second product may have lower broad visibility but a stronger commercial opportunity.
More Detailed Searches May Create Better Reasons to Click
Google describes AI Mode as particularly useful for nuanced questions, exploration and complex comparisons, with links allowing users to investigate supporting websites. Google Search Central
A click may become more likely when the searcher needs to confirm something that should be checked at source, such as:
- Current pricing
- Exact plan limits
- Availability
- Integration documentation
- Implementation requirements
- A demonstration
- A free trial
- A quotation or consultation
Google reports that clicks from its AI search experiences are increasingly “quality clicks,” by which it means visitors are less likely to return immediately to the results. This is Google’s own measurement, but it supports the possibility that users who click after a detailed search may arrive with stronger intent. Google on AI Search and website clicks
A broad brand mention and a decision-stage recommendation should therefore not automatically receive the same commercial value.
Choosing the Right Queries Matters
A visibility report can look impressive while measuring prompts with little connection to customers or revenue.
Compare:
What is CRM software?
with:
Which CRM supports five UK users, integrates with Xero and Outlook, and costs less than £100 per month?
The first query may create general awareness. The second describes someone actively evaluating practical suitability.
Before adding a prompt to a monitoring list, a business should ask:
- Does this query describe someone we can genuinely help?
- Does it concern a problem our product or service solves?
- Does it contain meaningful buying requirements?
- Could our website provide evidence supporting the answer?
- Is there a logical next step such as visiting, enquiring, booking or buying?
This is why choosing AI search prompts that actually matter is central to meaningful visibility testing.
Automating hundreds of irrelevant prompts would only measure the wrong things more efficiently.
A Related Hypothesis: Decision-Ready Information
This experiment did not test how websites can improve their AI visibility.
However, it suggests a related hypothesis worth investigating.
If AI Mode must decide whether a business satisfies a detailed customer query, the relevant facts logically need to be available for crawling and interpretation.
I think of this as decision-ready information.
| Business type | Decision-ready facts |
|---|---|
| Local service | Hours, staffed times, location, service area, price, turnaround and restrictions |
| Software | Pricing, plan limits, integrations, client access and implementation requirements |
| Professional service | Industries covered, eligibility, costs, timescales and enquiry process |
General claims such as “competitive prices,” “fast service” or “integrates with leading platforms” may not provide enough detail to resolve a specific query.
For example, if someone requires software for eight employees, under £100 per month, with client access and a direct Xero integration, AI Mode may need to establish:
- Per-user or workspace pricing
- Whether annual billing is required
- Which plan includes time tracking
- Whether client accounts cost extra
- Whether the Xero connection is direct or uses another service
- Whether all eight employees need paid licences
A website clearly providing those facts gives AI Mode more evidence with which to assess the product.
Information Does Not Guarantee Recommendation
Publishing complete information cannot guarantee inclusion in an AI response.
Other factors may still matter:
- The page may not have been indexed.
- AI Mode may use another source.
- Better-known competitors may dominate the initial candidate group.
- Sources may contradict one another.
- The requirements may be interpreted differently.
- Only three recommendations may be displayed.
- AI Mode may retrieve the information but still choose another product.
The cautious hypothesis is:
Publishing clear, complete information cannot guarantee an AI recommendation, but failing to publish it makes confident recommendation more difficult.
That is not an optimisation formula. It is a sensible question for future testing.
What the Experiment Demonstrated
This limited experiment covered five business-software categories, one Google account, one mobile device and two testing days.
Within those conditions:
- Broad prompts produced small default recommendation groups.
- Familiar category brands dominated the displayed answers.
- Four categories never moved beyond the same three products.
- CRM retained a completely stable leading pair.
- Order changed more often than shortlist membership.
- More detailed customer requirements produced different candidate groups.
- Broad visibility and commercially useful visibility were not the same measurement.
Final Thoughts
Across five software markets, broad Google AI Mode searches repeatedly produced small groups of familiar category-default brands.
The brands themselves were unsurprising. The lack of displayed alternatives was not.
More specific searches produced different candidates because the customer’s circumstances gave AI Mode a reason to move beyond the defaults. This means broad visibility and commercially relevant visibility should be measured separately.
The important question is not simply whether a brand appears.
It is whether that brand appears accurately when a potential customer describes a problem the business can genuinely solve—and whether it remains relevant as the search moves towards action.