A single AI search can hide a surprising amount of complexity.
Ask Google AI Mode something like:
“I’m looking for an accountant in Birmingham for a manufacturing business that uses Xero. Which accountancy firms should I consider, and why?”
To the person asking the question, that is one search.
But answering it well may require an AI search system to investigate several separate subjects.
Which accountancy firms serve Birmingham? Which have manufacturing experience? Which understand Xero? Which have evidence of relevant expertise? What particular accounting requirements might matter to a manufacturer?
This is where query fan-out becomes interesting.
In simple terms, query fan-out is the process by which an AI search system can expand a main question into a number of related searches covering the different subtopics or information requirements needed to construct an answer.
For AI visibility testing, that raises an important question:
Could understanding those potential subtopics help us choose better queries to test?
We decided to explore that using the free Otterly.ai Query Fan-Out Tool and Google AI Mode.
Why Query Fan-Out Matters to a Business
Suppose somebody is looking for an accountant for a Birmingham manufacturing company that uses Xero.
The obvious requirements are:
- Birmingham;
- manufacturing;
- Xero.
But the customer’s eventual decision could involve much more.
They might want to know:
- whether the accountant has genuine manufacturing experience;
- whether they understand Xero well;
- how much they charge;
- whether they have relevant reviews or case studies;
- whether they understand inventory and work-in-progress;
- whether they can help with manufacturing software integrations;
- whether they provide management accounts;
- or whether they can offer more strategic outsourced finance support.
One broad question can therefore contain, or lead towards, many smaller information needs.
An AI search system may explore some of those subjects when assembling its response.
For a business trying to understand its AI visibility, those component topics are potentially interesting because they may reveal different places within the customer journey where the business does or does not appear.
Testing the Otterly.ai Query Fan-Out Tool
We entered our Birmingham manufacturing-accountant scenario into the free Otterly.ai Query Fan-Out Tool.
It returned 52 suggested fan-out queries.
Crucially, we did not treat those as 52 searches that Google had definitely performed.
Nor did we assume that all 52 should automatically become part of an AI visibility monitoring programme.
They were better understood as possible directions for investigation.
The output included straightforward reformulations of the original query, but it also branched into areas such as:
- pricing;
- reviews;
- case studies;
- manufacturing cost accounting;
- job costing;
- inventory management;
- Bills of Materials;
- Xero integrations;
- R&D tax relief;
- outsourced FD services;
- software comparisons;
- and several more specialised manufacturing scenarios.
Some looked highly relevant to the buying journey.
Others seemed much less important to the particular customer we had in mind.
That distinction turned out to be important.
How We Ran the Test
This was an exploratory test, not an attempt to establish statistically stable brand visibility.
Our process was:
- Start with one commercially realistic customer query.
- Generate potential fan-out queries using Otterly.ai.
- Review the 52 suggestions rather than automatically accepting them.
- Select eight that represented different parts of the customer’s decision.
- Run each selected query once in Google AI Mode.
- Record how AI Mode interpreted the query.
- Note whether individual accountancy firms were recommended.
- Look at the type of evidence AI Mode appeared to use when explaining those recommendations.
We therefore ran nine AI Mode searches in total: the original baseline query followed by eight selected fan-out queries.
One run of each query is not enough to establish which firm is consistently most visible.
That wasn’t our purpose.
We wanted to see whether fan-out could help us uncover useful areas for further AI visibility testing.
Establishing a Baseline
Our original query was:
“I’m looking for an accountant in Birmingham for a manufacturing business that uses Xero. Which accountancy firms should I consider, and why?”
AI Mode recommended:
- PKF Smith Cooper;
- Prime Accountants Group;
- Dains Accountants;
- Linford Grey.
This gave us a baseline.
We could now explore narrower areas suggested by the fan-out tool and see whether the same businesses remained visible.
They often didn’t.
The First Fan-Out Exposed a Problem
One Otterly suggestion was:
“Cost of Xero accounting packages for manufacturers in Birmingham”
The intended topic seemed commercially useful: what might accountancy support cost?
But Google AI Mode interpreted “Xero accounting packages” largely as Xero software packages.
The response concentrated on Xero subscription prices and manufacturing software before eventually mentioning some local accountants.
That was one of the most valuable results of the experiment.
The topic behind the fan-out was useful.
The wording was less useful for the test we wanted to conduct.
This suggests an important distinction:
Fan-out discovery and prompt selection are not the same thing.
A generated query may point towards a worthwhile customer concern while still needing refinement before it becomes a good visibility-testing prompt.
A Better Pricing Query Produced a Different Result
Otterly had also suggested:
“Cost of hiring a Xero specialist accountant for a manufacturing SME in Birmingham”
That removed the ambiguity.
This time AI Mode understood that we wanted the cost of hiring an accountant.
But it did something else that was interesting.
It gave estimated accountancy fee ranges and discussed the factors affecting cost, but did not recommend any Birmingham accountancy firms.
The query was commercially relevant.
A prospective customer could genuinely want the answer.
But it produced very little information about individual brand visibility.
That gives us another useful distinction:
A commercially important query is not necessarily a brand-discovery query.
Both types can belong to the customer journey, but they may need to be measured differently.
Asking About Reviews Changed the Firms
We next tested:
“Birmingham accounting firms with Xero manufacturing expertise reviews”
AI Mode recommended:
- CASS;
- Avonmead Accountants;
- Prime Accountants Group;
- Integrity Accountancy Services.
Only Prime remained from the four firms in our original baseline response.
The underlying customer had not changed dramatically.
They still wanted a Birmingham accountant with manufacturing and Xero expertise.
But introducing reviews and evidence of reputation changed the businesses surfaced.
We then tried another Otterly suggestion:
“Best Birmingham accountants for a manufacturing company using Xero (reviews and case studies)”
This time AI Mode returned:
- Prime Accountants;
- Dains;
- Inform Accounting;
- Spotlight Accounting.
Again, the set of firms changed.
The answer also appeared to lean more heavily on specific evidence of manufacturing experience, including case-study-style material.
This suggested that different fan-out branches may cause AI search to consider different forms of evidence about the same potential supplier.
Moving Into More Specific Manufacturing Requirements
We then moved further down the customer journey.
Rather than asking generally who might be suitable, we tested specific capabilities.
One query was:
“Birmingham accountants specialising in manufacturing cost accounting and job costing using Xero”
AI Mode recommended:
- Onyx Accountants;
- Prime Accountants;
- CASS.
We then tested:
“Birmingham accountants offering Xero-based inventory management and bill of materials (BOM) support for manufacturers”
This produced:
- Inform Accounting;
- CASS;
- Perpetual Accountancy;
- Dains.
And the closely related:
“Accountants in Birmingham who implement Xero add-ons for manufacturing”
returned:
- Inform Accounting;
- CASS;
- Perpetual Accountancy;
- eCloud Experts.
This was particularly interesting.
Three of the four businesses remained the same across the two closely related technical queries.
That was much more stability than we had seen when moving between broader questions about general suitability, reviews and case studies.
A Strategic Requirement Changed the Picture Again
Our final selected fan-out moved away from technical implementation and towards strategic support:
“Which Birmingham accountants provide outsourced FD services for Xero-based manufacturers?”
AI Mode recommended:
- CASS;
- Linford Grey;
- Avonmead Accountants;
- Spotlight Accounting.
Again, the underlying type of customer was similar.
But the requirement had moved from accounting systems and manufacturing processes towards strategic financial support.
The firms surfaced changed accordingly.
A Business Is Not Simply “Visible” or “Invisible”
This was probably the most important lesson from the experiment.
It would be easy to test one query such as:
“Which Birmingham accountant should a manufacturing company using Xero consider?”
and then describe the businesses appearing in that answer as the visible firms.
But our fan-out tests showed a much more complicated picture.
A business might appear for:
- the broad recommendation;
- reviews;
- case studies;
- job costing;
- inventory;
- systems implementation;
- or outsourced FD support.
Another business may appear strongly for a different combination of those needs.
So rather than asking only:
“Are we visible for manufacturing accountants in Birmingham?”
a potentially more useful question is:
“Where within this customer’s journey are we visible?”
That is a much richer visibility problem.
The Same Customer Can Generate Different Visibility Tests
This also helps explain why simply building a very large list of prompts may not be enough.
The objective isn’t to accumulate prompts.
It is to represent meaningful customer needs.
Query fan-out can help expose some of those needs.
Our original search was centred on three obvious elements:
Birmingham + manufacturing + Xero.
But the fan-out tool encouraged us to investigate:
- price;
- reputation;
- proof of expertise;
- job costing;
- inventory;
- BOM;
- implementation;
- strategic finance support.
Some proved useful visibility tests.
One was ambiguous.
Another was commercially valuable but didn’t generate brand recommendations.
Others revealed businesses we had not seen in the original query.
That is precisely why generated fan-out queries need human judgement.
Fan-Out Is a Discovery Mechanism, Not a Ready-Made Prompt List
This is probably the clearest way to describe what we learned.
Query fan-out is not a ready-made prompt list. It is a discovery mechanism.
The 52 queries generated by Otterly were not 52 prompts we suddenly had to monitor.
Nor were they a definitive record of 52 searches Google had secretly run.
Instead, they gave us possible subtopics and search directions to consider.
For each one, we could ask:
Does this actually matter to our potential customer?
If it does:
Is the wording appropriate for the test?
Then:
What happens when we search it?
And finally:
Is it useful enough to continue monitoring?
Some queries will survive that process.
Others won’t.
That is exactly what we would expect from an exploratory testing programme.
What Evidence Is AI Using?
Another issue became increasingly noticeable as our queries became more specific.
AI Mode sometimes appeared to combine several related pieces of information in order to recommend a firm.
For example, it might find evidence that a firm:
- works with manufacturing businesses;
- is experienced with Xero;
- provides cloud accounting;
- and offers management reporting.
It may then conclude that the firm is suitable for a more specific requirement.
That could be a perfectly reasonable synthesis.
But there is an important difference between:
a business explicitly saying it provides a particular service
and:
an AI system inferring from several pieces of evidence that the business is probably suitable.
When analysing AI visibility, we therefore shouldn’t simply record which brands appeared.
We should also ask:
Why did the AI appear to think this business was relevant?
And where important claims are involved, the underlying sources should be checked.
This is particularly important when the requirement becomes highly specific.
Does Your Website Address the Relevant Subtopics?
For businesses, query fan-out also raises a useful content question.
Suppose a fan-out reveals that customers choosing your type of business may care about:
- pricing;
- a particular service;
- industry expertise;
- implementation;
- case studies;
- support after purchase.
If that genuinely matters to your customers, does your website clearly address it?
This does not mean creating pages for every query produced by a fan-out tool.
Nor can we say that adding a page about a particular subject will cause an AI system to recommend the business.
AI visibility depends on far more than that.
But if a subject is genuinely important to customers and genuinely part of what the business provides, making that information clear and accessible is sensible regardless.
The fan-out process can therefore help reveal potential information gaps, as well as additional prompts worth testing.
The Scale Problem Appears Very Quickly
There is another practical lesson from this experiment.
One starting query produced 52 possible fan-out directions.
We heavily filtered them and tested just eight.
Now imagine beginning with 30 or 50 commercially important customer queries.
Then select several useful fan-outs from each.
Then test them across Google AI Mode, ChatGPT, Perplexity and other AI systems.
Then repeat the tests over time because AI responses vary.
Then record:
- whether your business appeared;
- which competitors appeared;
- how prominently each business was presented;
- the sources cited;
- what was said about the business;
- and how those results changed.
The workload quickly becomes difficult to manage manually.
Manual testing remains extremely useful.
It is particularly valuable for exploratory work like this, where we want to inspect individual responses, notice unexpected interpretations and decide which prompts deserve further investigation.
But sustained AI visibility monitoring at scale is another matter.
That is one reason I have separately reviewed Otterly.ai as an AI visibility monitoring tool I can recommend.
My view there is consistent with what we found in this experiment: software can automate much of the repetitive collection and monitoring, but it doesn’t remove the need for human judgement about which customer questions are actually worth tracking.
AI Visibility Testing Is a Dynamic Process
There is unlikely to be one perfect prompt list that you create and then monitor unchanged forever.
Customer behaviour changes.
Your business changes.
Competitors change.
The questions worth testing change.
And AI responses themselves can vary.
Query fan-out adds another layer to that process because it can expose new subtopics or information requirements worth investigating.
Some will prove useful.
Some won’t.
Some may initially appear promising but later turn out to have little commercial value.
Others may lead to entirely new testing ideas.
That is why AI visibility testing is better viewed as a dynamic, evolving process rather than a one-time keyword exercise.
So, Can Query Fan-Out Help You Choose Better AI Visibility Tests?
Based on this exploratory experiment, we think it can.
Not because every generated fan-out should become a tracked query.
Not because a fan-out tool can tell us exactly which searches Google used to construct a particular answer.
And not because testing eight related queries tells us which Birmingham accountant is permanently the most visible.
The value is more practical.
Query fan-out can help expose the different information needs hidden inside a larger customer question.
Those subtopics can then be evaluated as potential visibility tests.
In our Birmingham manufacturing-accountant experiment, exploring different branches of the original question produced noticeably different results.
The businesses appearing changed depending on whether we asked about:
- general suitability;
- pricing;
- reviews;
- case studies;
- manufacturing costing;
- inventory;
- software implementation;
- or outsourced financial direction.
That reinforces something increasingly important in our AI visibility testing:
The aim is not simply to discover whether a business appears for one broad query. It is to understand how that business is represented across the different questions and requirements that make up a real customer journey.
Query fan-out can help uncover those questions.
The Otterly.ai free fan-out tool gave us useful places to start.
But, like most things in AI visibility testing, the output was not the answer.
It gave us suggestions about where to investigate next.
Human judgement still had to decide which directions were worth following.