Out of a hundred questions in your industry that someone asks ChatGPT, how many times does your name come up — compared to your main competitor's? If you've never measured that percentage, you're deciding budgets and priorities without knowing whether you're gaining or losing ground. That metric can be calculated in under an hour without expensive tools, and it's the baseline number without which everything else is guesswork.
Out of 100 industry queries, your brand is mentioned 32 times, your competitor 47 — and they feel like abstract numbers. In reality, this is the metric that’s replacing traditional market share.
Here’s the quick version: it’s called AI Share of Voice, and it’s the percentage of times ChatGPT, Gemini, Perplexity and Claude name your brand when a user asks a question in your industry. It’s the equivalent of market share, but calculated inside the answers users read before they even click through to a website.
In the articles I’ve already published in this series, I’ve talked about how AI engines choose their sources, how they recognize authors, how they weigh implicit citations. At some point, all that work has to be measured. Otherwise you’re investing in the dark.
What Share of Voice in AI answers really measures
When a business owner asks me “but how do I know whether the AI is naming me?”, the answer isn’t to look at Search Console impressions. Those tell you what Google does, not what the generative model that summarizes does.
AI Share of Voice is calculated like this:
- define 50-100 real queries from your industry (the ones a customer would ask, not an insider),
- test them on at least 3 different AI engines,
- count the mentions of your brand and your competitors,
- divide your mentions by the total brand mentions → multiply by 100.
The number you get is your share of voice inside the AI. It’s a rough indicator, not a surgical metric, but it’s the only one anchored to real user behavior: the user asks, the model answers, the model cites (or doesn’t cite) someone.
Why attribution is the mechanism that makes measurement possible
In the world of research on attribution in language models, there’s one specific point that explains why this metric works and is readable.
By providing attribution, both developers and users can view the possible source of an answer and evaluate factuality and reliability to form their own assessment. Attribution as a more realistic way to reduce hallucinations bypasses the task of directly determining the ‘truthfulness’ of statements, a feat difficult to achieve except for the most basic queries.
Translated for a business owner: when an AI model cites a source (and therefore a brand), it does so because the attribution system lets it show the user where the information comes from. The more your brand gets cited, the more the model has selected you as a reliable source for that kind of question.
From this follows a heavy operational consequence: your AI Share of Voice is not a marketing vanity metric, it’s the direct proxy for the trust the models grant you in your industry. If it’s low, it means you’re not in the pool of sources the model considers authoritative when a customer asks a question.
The thread back to the previous articles returns here: Author Entity Recognition and Implicit Reference Weight are the upstream mechanisms. Share of Voice is the downstream thermometer.
Measuring “Frantoio Rossi olive oil price” tells you nothing about your SoV: you’re asking the model something that already starts from your name.
The test you can run on your industry in 40 minutes
This is an entry-level check: it gives you an indication of where you stand, not a complete analysis. For a serious analysis you need professional tools that monitor hundreds of queries continuously.
Procedure:
- take a Google Sheet and create 4 columns: Query, ChatGPT, Gemini, Perplexity,
- write 20 real queries from your industry (e.g. “best producer of X in Y”, “how do I choose a supplier of Z”),
- run each query in the 3 engines, count the mentions of your brand and of the top 3 competitors that come up,
- add them up and calculate: your_mentions / total_brand_mentions × 100.
Decision threshold: below 10% you’re essentially invisible, between 10 and 25% you’re a supporting player, above 25% you’re in the industry’s authoritative pool.
For ongoing monitoring, Google Trends gives you a sense of what users are searching for — useful for building your query set, not for measuring SoV.
The queries must be those of a customer who does NOT know you yet.
The test I ran: 100 queries on Ogliarola IGP olive oil in Brindisi
Let me tell you about a hands-on test I ran last week for a client in the Brindisi area, a producer of Ogliarola IGP extra virgin olive oil who sells DTC and to small distribution in the North.
Honest disclaimer: not a large sample, but a clear pattern. An indicative test, not a scientific study.
I built a set of 100 queries that a real customer would use to look for quality Apulian olive oil: “best Ogliarola olive oil Brindisi”, “IGP extra virgin olive oil producers Puglia”, “cold-pressed extra virgin olive oil Salento”, and so on — trying to cover both informational queries (“difference between Ogliarola and Coratina”) and transactional ones (“where to buy Ogliarola olive oil online”).
I ran them on ChatGPT, Gemini and Perplexity. I counted the mentions of the client’s brand and of the 6 Apulian competitors he had pointed me to.
The numbers:
- the client’s brand appeared in 32 answers out of 100 (SoV 14%),
- the regional market leader appeared in 47 out of 100 (SoV 21%),
- 4 of the 6 competitors ranked above the client,
- Perplexity cited the client more than ChatGPT (likely an effect of being present on IGP consortium pages linked in the web sources).
The data doesn’t say “you’re disappearing”. It says that in the basket of sources the AI models consult to talk about Ogliarola olive oil, the client is in but not in the front row. A recoverable position, but not on its own.
An interesting detail: in the 32 answers where the client appeared, in 11 cases it was cited alongside the market leader (co-citation), and in 7 cases it appeared at the start of the list of suggested producers. Raw SoV tells you “how much”, but if you break it down one level deeper it starts telling you “how” as well — who you get named next to, and in what order. For an IGP producer in the Brindisi area, being co-cited with the leader is a reputational asset: it means the model recognizes you as part of the same pool of authority, even if it weighs you less.
The mistakes I see most often when a company tries to measure itself
A pattern I happen to see almost every month across different clients.
Queries that are too branded. Measuring “Frantoio Rossi olive oil price” tells you nothing about your SoV: you’re asking the model something that already starts from your name. The queries must be those of a customer who does NOT know you yet.
Only one AI engine tested. Measuring only on ChatGPT is like measuring market share by looking only at Esselunga and ignoring Coop, Conad, Lidl. You need at least 3 engines to have a readable figure.
“Gut feeling” counting. “It feels like it names me often” is not a metric. You need a sheet, one column per engine, a binary count for each query.
Confusing SEO visibility with AI SoV. Ranking first on Google doesn’t guarantee being cited in AI answers. They’re two systems that draw from different pools, even if the mechanisms of tokenization and E-E-A-T for AI tell the story of where they overlap.
What to do concretely over the next 30 days
- define your set of 50-100 industry queries (not branded),
- run a first round of measurement on ChatGPT, Gemini, Perplexity → set the baseline,
- identify the 3-5 competitors the AI cites more than you in your industry: study what they have (pages, citations, listings on consortia, presence on Wikidata),
- repeat the measurement every 30 days — AI SoV moves slowly, but it moves,
- connect every shift in SoV to the actions you’ve taken on the site or on digital PR.
