When businesses talk about visibility in AI search, citations tend to get a lot of attention.
If ChatGPT, Google AI Mode or another AI search system links to your website, that certainly looks positive.
Your content has been found.
Your website has apparently contributed something useful to the answer.
And there is at least a possibility that somebody will click the link.
But our recent AI Mode experiments have raised a more difficult question:
Does being cited actually mean your business is being recommended?
The answer appears to be: not necessarily.
And the reverse may also be true.
A business can appear prominently in an AI-generated recommendation even when its own website is not the source being cited alongside that recommendation.
That means we may need to stop treating citations and recommendations as if they are the same measure of AI visibility.
AI Visibility Can Mean Different Things
Suppose somebody asks:
Can you recommend an accountant for a small business?
There are several different ways a business could become visible in the resulting answer.
The AI might:
- recommend the business by name;
- mention the business without recommending it;
- cite an article from the business’s website;
- use its website as evidence for a wider explanation;
- recommend the business while citing somebody else’s website;
- cite the business without mentioning its services at all.
All of these represent some form of visibility.
But commercially, they are not equivalent.
If your aim is to attract potential customers, there is a substantial difference between:
Your article was used to explain CIS tax rules
and:
Your accountancy firm was recommended to a construction company that uses subcontractors.
Both could be valuable.
But they are valuable for different reasons.
We Have Already Seen Recommendations and Citations Separate
This distinction emerged during our earlier Google AI Mode experiments.
We repeatedly ran the same or very similar queries and found that the businesses being recommended could remain relatively consistent while the supporting citations changed.
We also encountered cases where the source linked alongside a recommendation did not appear to be the recommended company’s own website.
One particularly noticeable example involved Nasa Accountants.
Nasa appeared in our accountant recommendation results, but some of the citations associated with those answers pointed to material from another accountancy firm, Acumon.
That immediately created a problem if we tried to answer the apparently simple question:
Was Nasa visible in this AI result?
From a recommendation perspective, clearly yes.
It was presented to the customer as an option.
But from a citation perspective, its own website was not necessarily receiving the same visibility.
Those are two different things.
A Citation May Be Supporting the Answer, Not the Business
This makes more sense when we think about how an AI-generated answer is constructed.
Imagine somebody asks:
Can you recommend an accountant for a construction company that regularly uses subcontractors?
The AI has several tasks to perform.
It may need to understand:
- why construction accounting is different;
- what CIS is;
- why subcontractor verification matters;
- which firms serve construction businesses;
- which firms operate in the required location;
- which firms appear suitable for the customer.
The source that helps explain CIS rules does not necessarily have to be the same source used to identify a suitable accountant.
An authoritative page from one company could provide useful explanatory information.
A different accountant could then be recommended.
The final AI response merges those pieces into a single answer.
From the customer’s perspective, that synthesis may feel seamless.
From the business’s perspective, however, it means:
The source providing the evidence and the business receiving the recommendation can be different entities.
That is an important distinction for AI visibility measurement.
The Reverse Can Happen Too
Now consider the opposite situation.
Suppose an accountancy firm publishes an excellent guide to director’s loan accounts.
AI repeatedly cites that guide when answering questions about overdrawn director’s loan accounts.
That sounds like excellent AI visibility.
And in one sense it is.
But what if the firm is never actually presented as an accountant the searcher could contact?
The business might have:
high citation visibility
but:
low recommendation visibility.
That could still have considerable value.
People may click the citation.
The firm may build authority.
Its content may influence the answer.
But it is not the same outcome as being directly presented to the prospective customer as a suitable provider.
Recommendation Visibility and Citation Visibility Should Be Measured Separately
This suggests that an AI visibility report needs to distinguish between at least two measurements.
Recommendation Visibility
This asks:
How often does AI actually recommend the business?
For example, if we run a commercially meaningful query ten times and the business is recommended six times:
Recommendation visibility: 6/10
or:
60%
That tells us how often the business enters the customer’s potential consideration set.
Citation Visibility
This asks:
How often is the business’s own website cited in the answer?
The same company might be cited only twice across those ten runs.
So we could have:
Recommendation visibility: 60%
Citation visibility: 20%
That would be a very different picture from simply reporting:
AI visibility: 60%.
A Business Could Also Have the Opposite Pattern
Imagine another company.
Its useful educational content is cited in seven of the ten answers.
But the company itself is only recommended once.
Its results could look like:
Citation visibility: 70%
Recommendation visibility: 10%
Which business has better AI visibility?
There isn’t one simple answer.
They have different kinds of visibility.
The first is regularly being placed in front of potential customers.
The second is regularly contributing information to AI answers.
Depending on the company’s objectives, both could matter.
But collapsing them into one number would hide what is actually happening.
Our Representation Experiment Added Another Layer
We recently looked at something else:
When AI recommends a business, does it describe that business accurately?
We compared Google AI Mode’s descriptions with the businesses’ own websites.
The results were mixed.
Some claims were directly supported.
Some appeared to be reasonable interpretations.
Some could not be verified from the company’s own website.
And AI sometimes omitted commercially important information that the company clearly published.
That gives us a third dimension:
Representation Accuracy
When the business appears, how accurately does AI explain what it does and why it is relevant?
A business could therefore perform very differently across three measures.
For example:
| AI visibility measure | Example result |
|---|---|
| Recommendation visibility | 70% |
| Citation visibility | 25% |
| Representation accuracy | 90% |
That tells a much richer story than:
AI visibility score: 70%.
Brand Mentions May Be a Fourth Category
There is another distinction we may eventually need to make.
A business can be mentioned without being recommended.
For example:
“Platforms such as HubSpot, Pipedrive and Zoho are common alternatives, although specialist recruitment products may be more suitable.”
HubSpot has been mentioned.
But it wasn’t necessarily recommended for the specific customer.
So we could ultimately distinguish:
Brand mention visibility
How often does the company appear anywhere in relevant answers?
Recommendation visibility
How often is it actually presented as a suitable choice?
Citation visibility
How often does its own website appear as a source?
Representation accuracy
How accurately is the company described?
These are related.
But they are not interchangeable.
Why This Matters for Share of Voice
This also affects how we think about AI share of voice.
Imagine an AI monitoring tool tells you:
Your company has 28% AI share of voice.
That sounds precise.
But the first question should be:
28% of what?
Does that mean:
- mentioned in 28% of answers?
- recommended in 28%?
- cited in 28%?
- appeared in 28% of the citations?
- appeared anywhere in the generated text?
Those measurements can lead to very different commercial interpretations.
This is why we think AI visibility metrics should always explain exactly what is being counted.
A percentage without a clearly defined underlying event can create more confidence than understanding.
Commercial Intent Makes the Difference Even More Important
The distinction becomes particularly important when the search query has genuine commercial intent.
Consider these two searches:
What is CIS?
and:
Can you recommend an accountant for a small construction company in Bristol that regularly uses subcontractors?
An accountant’s website being cited in the first answer could be valuable.
But the second search involves someone actively asking for a provider.
Being recommended there has a much more direct relationship with a possible customer enquiry.
This links back to another conclusion from our experiments:
AI visibility should be measured against customer situations that actually matter to the business.
If the objective is lead generation, recommendation visibility on high-intent searches may matter more commercially than citation visibility on broad informational queries.
That does not make citations unimportant.
It means they answer a different question.
Citations May Still Be Extremely Valuable
None of this should be interpreted as saying citations don’t matter.
They can matter for several reasons.
A citation can:
- create a direct click opportunity;
- expose a brand to somebody who had never heard of it;
- establish the website as a useful source;
- support brand familiarity;
- contribute to an AI-generated answer even when the company is not recommended.
There could also be situations where citation visibility eventually contributes indirectly to recommendation visibility.
We have not established whether that happens, and we should not assume it does.
But citation visibility is clearly something worth measuring.
It just shouldn’t automatically be treated as the same thing as being recommended.
The Citation Itself Also Needs Checking
There is another complication.
Seeing your domain cited is not enough.
We should also ask:
What is the citation actually supporting?
Suppose AI says:
“Construction businesses regularly using subcontractors need to file monthly CIS returns.”
and cites an accountant’s guide.
That citation supports general information.
It does not necessarily mean AI considers that accountant the best provider for the customer.
Likewise, an AI recommendation could be followed by several citations, but those citations may support different sentences or different parts of the synthesis.
So a proper citation audit should probably record:
- which website was cited;
- what claim the citation appears to support;
- whether the cited page actually supports that claim;
- whether the cited business is also being recommended;
- whether a competitor’s content is supporting your recommendation or vice versa.
That last situation could be particularly interesting.
AI Answers May Create a Network of Visibility
Traditional search results made the relationship relatively simple.
A page appeared.
The user clicked the page.
The website receiving visibility and the website receiving the click were normally the same thing.
AI-generated answers can separate those roles.
One business may supply the information.
Another may be recommended.
A third-party review site may provide evidence.
A directory may establish a location.
The AI then synthesises all of those sources into one response.
That means AI visibility may increasingly operate as a network rather than a simple ranking list.
Different websites can play different roles within the same answer.
This is one reason why simply asking:
What position did we rank?
may no longer tell us enough.
A More Useful AI Visibility Dashboard
If we were building a simple AI visibility report for a business, we would now be tempted to show several separate measurements.
For example:
| Metric | What it tells us |
|---|---|
| Recommendation rate | How often AI presents the business as a suitable solution |
| Citation rate | How often the company’s own website is used as a source |
| Mention rate | How often the brand appears anywhere in the response |
| Representation accuracy | Whether AI’s description matches the business |
| Query relevance | Whether those appearances happen on commercially meaningful searches |
That would tell the business far more than a single headline visibility score.
We Have Not Yet Quantified the Relationship
There is an important limitation to this discussion.
We have observed enough examples to show that citations and recommendations can separate.
But we have not yet taken a large query set and systematically calculated:
What percentage of recommended businesses have their own websites cited?
Nor have we measured the reverse:
What percentage of cited businesses are actually recommended?
That would be a worthwhile future experiment.
We could take a set of repeated commercial recommendation queries and record, for every answer:
- businesses recommended;
- businesses mentioned;
- domains cited;
- whether each citation belongs to the recommended business;
- which claim each citation supports.
That would allow us to quantify the relationship rather than simply observe it.
But the Main Lesson Is Already Clear
A citation and a recommendation are not the same event.
A business can apparently receive one without necessarily receiving the other.
And that means businesses measuring AI search visibility need to be precise about what they are actually tracking.
Instead of asking:
Are we visible in AI?
better questions may be:
Are we being recommended?
Is our website being cited?
Are we merely being mentioned?
And when we do appear, are we being described accurately?
Those questions describe different parts of the customer journey.
Final Thought
AI search visibility is beginning to look less like one metric and more like a collection of different signals.
A business may be highly visible as a source but rarely recommended.
Another may be regularly recommended while its own website is rarely cited.
A third may appear frequently but be described inaccurately.
So rather than searching for one universal AI visibility score, it may be more useful to understand what kind of visibility a business actually has.
Because ultimately:
Being used as a source, being mentioned and being recommended are three different outcomes.
And if the goal is to understand whether AI search can bring customers to a business, knowing the difference matters.