In one of our audits, we tracked a specialist infrastructure software company's share of voice across 239 prompts mapped to how buyers actually evaluate this kind of product. Overall, their share of voice was 27%. On its own, that looked like a solid number. AI share of voice is becoming a common marketing metric for measuring how often a brand appears relative to competitors in AI answers. But an overall score can hide as much as it reveals.
Then we cut the same data by the kind of question being asked.
On prompts about their core technical capability, the thing they're most clearly documented for, their share of voice was 23.2%. AI was finding them exactly where you'd expect.
On prompts about deciding which GPU-based Kubernetes services to build and roll out, and how to position, price, and package them once built, their share of voice was 0.0%.
These are the questions a company works through before it starts comparing vendors: what to offer and how to bring it to market. If a product isn't part of that conversation, it never reaches the stage where build-vs-buy or vendor comparisons happen at all.
A 27% overall share of voice looks like a reasonable outcome on its own. But it hides where that visibility is coming from, and where it disappears.
That is the problem with tracking AI brand mentions through one visibility number. Share of voice can tell you how often a brand appears. It cannot tell you which buyer questions are driving that visibility, where the brand disappears, or who is seeing it.
What AI Share of Voice Actually Measures
AI share of voice is often the first marketing metric a team looks at when measuring brand presence in AI search. It's simple: how often does AI mention your brand, relative to competitors, across a set of prompts.
But it's one number covering a lot of different ground. It doesn't distinguish between a technical question and a commercial one, or between a buyer who's just starting to look and one who's about to decide. All of that gets folded into a single average, which is exactly what happened in the audit above: 27% overall, while entire categories of buyer questions sat at zero.
Once you have that number, the more useful question is what sits underneath it.
Where AI Share of Voice Started to Break Down
The zero from the opening wasn't an isolated result. When we cut the same audit across all of the buyer needs mapped to this company's core solution, share of voice ranged from 23.2% down to zero, with meaningful differences across the buyer needs in between. We term these as buying considerations.
On tenant isolation, a core piece of what the product does, share of voice was 23.2%. On choosing the right technical architecture, 12.8%. On build vs. buy, 9.5%. On competing in the marketplace, 9.0%. On choosing a credible partner, 7.1%. On maximizing revenue from the underlying asset, 4.0%.
Three buyer needs sat at exactly zero: deciding which services to prioritize building, positioning and pricing them once built, and operationalizing them after deployment.
This wasn't one isolated weak spot. The visibility dropped steadily across different buyer needs. Buyers asking details about the technology were finding this company. Buyers asking what to build with it, and how to take it to market, were not.
How AI Share of Voice Changes by Buyer Persona
Share of voice also depends on who's asking, not just what's being asked.
We cut the same audit by the three personas involved in this kind of purchase. The technical champion, evaluating whether the architecture actually works, saw a 13.1% share of voice. The economic buyer, who owns the final decision to buy, saw 7.8%. The person responsible for operationalizing the product after purchase saw 4.7%.
Same company, same underlying product. Three different pictures, depending on which buyer's questions you're looking at.
The persona breakdown adds another layer to the buying consideration data. The two do not always move together.
Two of the three buying considerations that came back at zero, deciding what to build and how to bring it to market, do belong to the persona with the lowest overall share of voice. But the third, day-two operationalization, belongs to the persona with the highest overall score. A buyer can be well represented on average and still find nothing on the one need they actually care about.
What to Track Beyond AI Share of Voice
Once you have both cuts, the buying consideration breakdown and the persona breakdown, the question changes. It is no longer just, "Are we visible?" It becomes, "How does AI actually understand us?"
Some of what's worth asking directly:
- Where does share of voice drop to near zero, and for which buying consideration, which persona, which stage of the purchase
- When the brand appears, what category is AI placing it in, its own category, or a broader approximation of it
- Which companies show up alongside it, the ones it actually competes against in real deals, or the largest adjacent platforms
- Whether AI's description actually matches the company's real positioning, not just whether it's mentioned at all
That source question matters more than it may seem. In a separate audit we ran for a law firm, AI had pulled its description almost word for word from a directory profile written by a researcher for a completely different purpose. The firm never wrote those words but a buyer heard them anyway.
None of these come with a number to hit. There's no target percentage for "AI cites the right sources" or "AI puts you in the right category." What they give instead is a map of where the picture breaks down, and that map is the actual output.
What AI Share of Voice Still Cannot Tell You
The cuts above show where a company disappears: which buying consideration, which persona, which stage of the purchase. What they can't show is why.
A near-zero share of voice on a specific need could mean the company has nothing to say on that question. It could mean something real exists but was never framed in a way that answers it directly. Or the content could already exist and be framed well, but not appear in sources AI treats as authoritative on that topic. Three different problems, three different fixes, and the measurement alone doesn't tell you which one you're looking at.
Tracking tells you where to look. It doesn't tell you what to do with what you'll find there.
Closing
AI is becoming the first place buyers form an impression of a company. Two different things can go wrong with that impression.
The first: the company might not show up for the exact questions that decide whether it even enters the running before a buyer starts comparing vendors.
The second: even where the company does show up, the language AI uses to describe it may not come from the company at all.
AI repeats whatever it finds and treats as authoritative, whether that's a company's own site or a directory profile written for a completely different purpose. That is why AI share of voice works best as a starting marketing metric, not as a complete measure of how AI understands a brand.
Our audit maps both. Share of voice by persona and by buying consideration, where a company goes dark and for whom, the sources shaping its AI description, and how far that description sits from how the company actually positions itself.