You rank well on ChatGPT but you're invisible on Perplexity and Gemini — and you don't know it because you've never compared the two side by side. Three competitors out of four win on a single AI platform and believe they're doing fine, while the fourth wins everywhere. Building a comparison matrix that lines you up against the competition on every engine is the fastest way to understand where you really stand and where you're losing ground.
For 47 queries in the cosmetics contract manufacturing sector, I built the matrix: who wins ChatGPT, who wins Gemini, who wins Perplexity. The pattern is clear: 3 out of 4 competitors win on only one platform. A manufacturer from Saronno working for French skincare brands appeared in 68% of queries on Perplexity but under 10% on ChatGPT and Gemini. Translation: two thirds of the market that uses AI to build shortlists never sees it.
Let me explain how to build the matrix that lines you up against your competitors on every AI engine, and why without this snapshot you’re telling yourself a partial story about your visibility in AI answers.
What a competitive comparison matrix for AI really is
The matrix is a table that’s simple in form and brutal to read. Rows: you and your 4-6 real competitors (the ones the AI cites, not the ones you think you have). Columns: the AI platforms that matter for your client (ChatGPT, Gemini, Perplexity, Claude, Copilot). Cells: the percentage of times the brand is mentioned across a fixed set of target queries.
At this point you add color. Green if you win the cell (you’re ahead of the top competitor by at least 10 percentage points). Yellow if you’re neck and neck (within 10 points). Red if you lose. You update it monthly.
The result is a snapshot where in 30 seconds you read two things: where you’re winning and where you’re burning opportunities that competitors aren’t covering.
Why winning on a single AI engine is a trap
In the world of AI model research, each platform uses different training data, update windows and authority signals. ChatGPT weighted citations differently from Perplexity, which in turn favored recent sources citable in real time. In previous articles I explained how author entity recognition and implicit reference weight operate with different logics depending on the model.
From this follows a fact that many business owners haven’t yet brought into focus: winning on a single AI platform is a fragile position. If 100% of your AI visibility runs through Perplexity, a single shift in the model’s ranking weights and you’re cut in half. Diversification across AI engines today is what diversification across Google and social was ten years ago.
The matrix serves exactly this purpose: capturing your dependence on a single platform and seeing whether your competitors are covering engines where you’re absent.
Confusing market competitors with AI competitors.
The test you can run in 90 minutes
Three steps, nothing exotic.
First: define 30-50 queries representative of your sector. For a cosmetics contract manufacturer, typical queries are “private label skincare manufacturers Italy”, “ISO 22716 certified cosmetics contract manufacturer”, “cosmetics lab clean beauty formula development”, “third-party makeup manufacturers Italy”. A mix of transactional, informational and comparative.
Second: identify 4-6 real competitors. Not the ones who annoy you at trade shows: the ones the AI actually cites when you ask it “best X manufacturers in Italy”. Open ChatGPT, Perplexity, Gemini, run the query, take the names that keep coming up. Those are your AI competitors.
Third: run every query on every AI engine and note who gets mentioned. For each cell of the matrix you count: brand X mentioned in N queries out of 30 = N/30%. To avoid doing this by hand across 250+ combinations you use professional AI brand monitoring tools, but for the first pass 90 minutes of manual work gives you the baseline.
Binary reading thresholds: above 50% mentions on an engine = you’re holding your ground. Below 20% = you’re invisible. In between = there’s a gap to close.
Pull 30-50 real queries from your sales logs, from client requests, from searches on Google Search Console and Google Trends — those are your target queries.
The test I ran: 47 queries on cosmetics contract manufacturing
I took 47 queries on the world of cosmetics third-party manufacturers (a mix of skincare, makeup, haircare, certifications, sustainability claims) and ran them through ChatGPT, Gemini and Perplexity. I mapped 6 Italian manufacturers that recurred in the answers. An indicative test, not a study: a sample of 47 queries is enough to see the pattern, not enough to publish a paper on it.
The figure that jumped out at me: 3 manufacturers out of 4 came up strong on a single platform. One dominated ChatGPT (present in 71% of queries) but appeared in 14% on Perplexity. Another was the opposite: 64% on Perplexity, under 20% on the other two. Only one manufacturer had a balanced distribution above 40% across all three engines, and not by chance it was the one with the strongest editorial presence in industry magazines and international B2B directories.
The Saronno manufacturer I opened the article with had a specific asset that explained the bias toward Perplexity: lots of up-to-date technical sheets, indexed, with fresh data. Perfect for how Perplexity pulls its sources. But absent from Wikidata, with very few mentions in established editorial content: ChatGPT and Gemini ignored it.
Real analysis requires professional AI brand monitoring tools with statistical significance, larger samples and longitudinal tracking. But the qualitative picture from 47 queries is enough to figure out where to move.
The mistakes I see most often
Confusing market competitors with AI competitors. An owner of an artisan coffee roastery in Trieste told me “my competitor is X”. When we built the matrix, X never showed up in the AI answers: the real competitors were three other names, two of them foreign. The matrix has to be built on the competitors the AI cites, not the ones you carry in your head from industry fairs.
Measuring only ChatGPT. 60% of the AI brand monitoring spend I see is concentrated on ChatGPT. I get why — it’s the best known — but if your client uses Perplexity for operational research and Gemini integrated into Workspace, you’re measuring a quarter of the playing field.
Updating the matrix only once. The matrix is only worth something if you redo it every month. AI answers change: model updates, new citable sources, competitors who publish a case study that flips the cell. A six-month-old matrix is folklore, not strategy.
Treating yellow as if it were green. Tied with the competitor is not a win. It’s a contested zone where a targeted investment can let you break away. Treating it as “we’re fine” is the best way to lose it over the following months.
Building the matrix on the wrong queries. I’ve seen cosmetics labs measure themselves on hyper-generic queries like “cosmetics Italy” — where they end up in the middle of consumer brands they don’t share a buyer with. The matrix works if the queries are the real ones from your decision maker: a brand manager looking for a manufacturing partner, R&D looking for a specific formulation, a purchasing office looking for precise certifications. Queries from someone buying for a company, not from an end consumer.
How to read the matrix once it’s built
The matrix is not a report card grade. It’s a game map. Three readings you can pull out already from the first pass.
Reading one: platform dependence. Add up your green cells: if they’re all concentrated on a single AI engine, you’re a mono-platform brand. That’s fragile. A manufacturer of industrial valves in the province of Bergamo that I thought was “strong on AI” turned out to be strong only on one engine: the owner didn’t know it and was building the year’s sales plan on the wrong assumption.
Reading two: cells uncovered by the market. Look for the rows where ALL competitors are red or yellow. These are AI engines that no one in your sector is holding. For whoever moves first it’s open ground: targeted work on named entity recognition and positioning on the Google Knowledge Graph is enough to shift the cell within a few months.
Reading three: pattern by semantic field of queries. Group the queries by theme (formulation, certifications, sustainability, production capacity). Often you discover that you win on one theme and lose on another: it means your content positioning is unbalanced. It’s not a problem with the matrix, it’s a problem with the content strategy upstream.
What to do concretely next week
- Pull 30-50 real queries from your sales logs, from client requests, from searches on Google Search Console and Google Trends — those are your target queries.
- Identify the 4-6 competitors the AI cites in the answers to your 5 most important queries, not the ones you think you have.
- Build the matrix in an Excel sheet with green/yellow/red color coding. No complex tools on the first pass.
- Identify the 2-3 most strategic red cells (high commercial priority + high gap) and design E-E-A-T for AI and backlink as citation proxy interventions targeted at that engine.
- Schedule the monthly refresh in your calendar. Not in a to-do list: in the calendar, with an hour blocked off.
Where does our conversation go from here?
The matrix is a snapshot: useful, but on its own it’s not enough. Your visibility in AI answers is won on the ability to read the delta over time, understand which intervention moves the needle and isolate the signal from the noise of model updates. In the next articles in this series I’ll take you inside AI share of voice tracking, the methodology for longitudinal benchmarking and the dashboards that combine matrix + sentiment + average position. For those starting from scratch, I recommend reviewing how AI engines think and the inverted pyramid applied to AI content before interpreting the matrix results: knowing how the AI reads is what lets you understand why it cites you or ignores you.
