Knowing that AI cites you in 18% of the queries in your industry doesn't tell you whether you're doing well or badly: in some markets it's an excellent position, in others you're fourth out of four. Without an industry reference point, the number floats in the void and you can't make any sensible decision about where to invest. Building the right benchmark for your market turns a generic figure into an operational compass.
Your AI Share of Voice is 18%. Is that good? It depends on the industry. In luxury fashion, 18% is excellent; in strategic consulting, it’s below average; in local services, it might even be a huge opportunity. Without your industry benchmark, that 18% is a number hanging in the void: it doesn’t tell you whether you’re winning, whether you’re losing, whether you should push or whether you should just defend what you have.
Let me explain it with a case I handled personally. A thermal spa hotel in Salsomaggiore Terme, in the province of Parma, came to me with this line: “we get 22% of citations when people ask about spas in Emilia, it seems low.” It seemed low to me too, before I built the benchmark. After building it, that 22% turned out to be the second absolute position in an AI market extremely concentrated around two players. It wasn’t a result to improve at all costs: it was a position to defend with a different plan.
In the previous articles in this series I showed you how to measure your AI Share of Voice and how to run a gap analysis. Here I’ll explain the piece that sits in the middle: how to turn your raw figure into an operational judgment, by calculating the distribution of mentions across all the players in your industry.
What I mean by an industry benchmark for AI visibility
An industry benchmark isn’t “the market average.” It’s the map of concentration: how many brands split the AI citations in your vertical, in what proportions, and how far apart the first and the last are.
In classic competitive analysis, the Herfindahl index is used to measure how concentrated a market is — the higher it is, the fewer companies dominate. In the answers from ChatGPT, Claude, Gemini and Perplexity the same principle holds, but applied to the citations the model generates when a user asks an industry question. From this follows a direct operational consequence: your 18% means nothing until you know whether the top three brands in your industry take 90% of the mentions or whether the market is fragmented across twenty more or less equivalent players.
In a concentrated market, being fourth is a disaster. In a fragmented market, being fourth is a position of strength.
Why without this figure your goals become fantasies
I often see briefs like “we want to reach 50% AI Share of Voice by year-end.” It sounds ambitious. It’s often impossible, because in many industries the historical leader has accumulated ten years of links, citations on Tier 1 publications and a density of entities that no generative model would dream of leapfrogging in twelve months.
The benchmark serves exactly this purpose: to define a goal that fits within the physics of your market. If in your industry the leader sits at 35% and the top five brands cover 78%, then a realistic goal for you in position 4 isn’t “becoming the leader,” it’s “consolidating that 12% and bringing it to 18% by pushing on the queries where the leader is weak.” It’s a different conversation, and it produces very different editorial plans.
In the articles on E-E-A-T for AI and backlinks as a citation proxy I’ve already explained why certain players are hard to dislodge. The benchmark is the moment when that theory becomes a number.
Confusing the average with the benchmark.
How to build your benchmark in 60 minutes
You don’t need enterprise tools for the first pass. All it takes is a one-hour session, an Excel sheet and a bit of discipline. Here’s the procedure I use when I have to give a client the first honest snapshot of their AI market.
First, define the query perimeter. Choose 15-20 questions a customer in your industry would ask an AI model to discover suppliers. They must be genuine discovery questions, not brand queries. For the Salsomaggiore thermal hotel, the questions were things like “best thermal spa hotels in Emilia-Romagna,” “where to go to the spa in the province of Parma for a weekend,” “hotels with subsidized thermal treatments near Milan.”
Then run the queries on at least three different AI engines. I use ChatGPT, Perplexity and Gemini. For each answer, note on a sheet: which brands are named, how many times, in what order. Mentions in the “cited sources” carry different weight from those in the body of the answer — I keep them separate in two columns.
Finally, calculate the distribution. Sum the mentions for each brand across all queries and all engines, then divide by the total. That’s the industry AI Share of Voice. Order it from highest to lowest and look at the curve: if the top three brands cover more than 70%, you’re in a concentrated market. If the curve is flatter, you’re in a fragmented market. It’s an indicative test, not a statistical study — a sample of 15-20 queries isn’t large — but the pattern emerges clearly.
Choosing whether to attack the leader or to hold the niches where the leader isn’t present — almost always the second is more effective for those in position 2-5.
The Salsomaggiore thermal hotel case: how 22% turned out to be second place
When I applied the method to the client I mentioned, this distribution came out, built on 18 queries across ChatGPT, Perplexity and Gemini.
- Player A (historic national chain): 41% of mentions
- My client: 22%
- Player C (independent property in Tabiano): 14%
- Player D (resort hotel in Castrocaro): 9%
- Another 11 brands split the remaining 14%
The top two covered 63% of the AI market. The market was concentrated, the leader was a chain with twenty years of editorial coverage on thermal tourism publications, and my client was firmly second. The 22% that on instinct seemed low was actually a position to consolidate, not to chase down desperately.
The action that came out of it: no assault on the leader, but content built around the queries where the leader was absent or weak — subsidized spas for ENT treatments, thermal stays for elderly guests with assistance, off-season wellness weekends. These are semantic niches where the historic leader hadn’t accumulated signals, and where a second player with good content could become the leading AI source.
Obviously: this is an indicative test done in an hour. The real analysis, with samples of hundreds of queries and longitudinal tracking over six months, requires professional tools and a serious investment. But to understand where you are and where it makes sense to go, the hour in Excel is more than enough.
The mistakes I see most often when building a benchmark
I always see the same four stumbles, and I’ll list them because they save you weeks of useless analysis.
- Confusing the average with the benchmark. The average doesn’t exist in almost any AI market. The figure that counts is the concentration curve, not a single number.
- Including brand queries. If you ask the engine “thermal hotel Salsomaggiore So-and-So,” it cites So-and-So. You haven’t measured visibility, you’ve measured that the model can do literal matching.
- Using only one AI engine. ChatGPT, Perplexity, Gemini and Claude have very different citation biases. Measuring on only one gives you a partial and often misleading snapshot.
- Not updating. March’s benchmark isn’t valid in September. Models get retrained, publications shift weight, competitors publish. A refresh every 90 days is the bare minimum.
What to do once you have the benchmark in hand
The benchmark isn’t a trophy to frame. It’s the starting point for three operational decisions that completely change your AI visibility strategy.
- Define a realistic goal for the next 6-12 months that fits within the physics of your market.
- Choose whether to attack the leader or to hold the niches where the leader isn’t present — almost always the second is more effective for those in position 2-5.
- Decide how much to invest in content, digital PR and authority signals: in a concentrated market you need much more, in a fragmented one you need much less to move the needle.
This is the thread of visibility in AI answers: being there isn’t enough, you have to know how much your presence weighs compared to those playing the same game as you. The benchmark is the tool that turns your isolated figure into a readable competitive position.
In the upcoming articles in this series I’ll take you inside the techniques of AI visibility forecasting, the maturity models for understanding what level you’re at as an organization, and the use of AI visibility as a competitive moat defensible over the medium term.
