An AI visibility report answers a single question: when someone asks ChatGPT, Gemini or Perplexity for a recommendation in your category, does your site appear among the sources, in which position, and who shows up in your place? Everything else, the percentages, the heatmap, the domain list, is a way of breaking that question into parts you can act on.
To explain each part we are not going to use a made-up example. We are going to use two real reports we generated between August and September 2026 for two Argentine brands in very different industries: Mudafy, a network of real estate agencies with its own technology, and Tiendamia, which sells US products with delivery in Argentina. Neither is a Dashcrab customer. The reports come from the same public scan we run for any domain, and we show them because two brands with similar headlines, 24% and 16%, have different problems, and that difference explains more than any definition.
If you are not yet sure what GEO is or why an AI answer cites what it cites, start with the guide from scratch. Here we go straight to the report.
The two reports in one table
The two reports were built the same way: 10 high-intent questions about the category, sent to 5 AI engines (ChatGPT, Claude, Gemini, Perplexity and Google AI Mode). That is 50 answers per brand. For each one we recorded which sources the model linked and in what order.
| What it measures | Mudafy | Tiendamia |
|---|---|---|
| Industry | Digital real estate | Shopping from the US to Argentina |
| Visibility | 24% | 16% |
| Answers with a citation (of 50) | 12 | 8 |
| Share of voice | 2.0% | 1.6% |
| Average position | #5.6 | #5.1 |
| Total citations in the panel | 591 | 488 |
| Top-3 combinations | 4 | 4 |
| Scan date | Aug 19, 2026 | Sep 3, 2026 |
That table alone tells the story: Mudafy appears in one of every four answers, but far from first place. Tiendamia appears less, and almost everything it has is in a single engine. Two similar headlines, two different problems. What follows is what each row means and why it changes what you have to do.
The four figures at the top
The report opens with four numbers. They are four different questions about the same set of answers, and it pays to read them in order, because each one corrects the reading of the previous one.
Visibility. The percentage of answers where your domain appears as a source. It is the headline figure: it says whether you are in the conversation or not. Mudafy, at 24%, is in 12 of the 50 answers. Tiendamia, at 16%, is in 8. What it does not say is how much weight you carry.
Share of voice. The percentage of all citations in the panel that are yours. Here comes the correction: Mudafy has 24% visibility but 2% share of voice. How? Each answer links 10 to 12 sources, so being in one of every four answers with a single link leaves you a small slice of the total. Visibility says whether you are there; share of voice, how much of what the model shows the user is yours.
Average position. The mean place you occupy in the source list when you are cited. In several interfaces the first source is the only one visible without expanding the rest, so a #1.7 and a #8 are two different realities even if both count as citations. Tiendamia averages #5.1 overall and #1.7 in ChatGPT, where one of its three citations came first. Google AI Mode cited it once, in position 14.
Total citations. The universe: every citation to any domain across the 50 answers. It is the denominator of share of voice (Mudafy: 12 of 591, Tiendamia: 8 of 488) and a measure of how many sources you compete against per answer. Between 488 and 591 citations across 50 answers is about 10 sources each time. That is the list you need to get into.
High visibility with low share of voice means you get in but you do not dominate: you are cited once in ten. Low visibility with a high average position, like Tiendamia in ChatGPT (30% and #1.7), means that when the model finds you it puts you first: the problem is that it rarely finds you. Those are two different work plans, and the headline alone does not tell them apart.
Visibility per model and the heatmap
There is no such thing as AI as a single place. Each engine retrieves pages from a different index, with different criteria, and that is why the report splits visibility into five columns. For each model it shows reach (in how many of the 10 answers you appear), first seat (what share of those citations landed in position 1) and average position.
| Engine | Mudafy | Tiendamia |
|---|---|---|
| ChatGPT | 20% · #3.0 | 30% · #1.7 |
| Claude | 30% · #5.0 | 20% · #3.5 |
| Gemini | 30% · #8.3 | 20% · #7.5 |
| Perplexity | 30% · #4.3 | 0% · N/A |
| Google AI Mode | 10% · #8.0 | 10% · #14.0 |
Reach (answers with a citation, out of 10) and average position when cited. The same brand changes story from one column to the next. Tiendamia is a brand ChatGPT recommends first and Perplexity does not know. Mudafy is even across four engines and almost absent from AI Mode, and Gemini cites it three times but far down, at position 8 on average.
Why it matters: when someone sells you optimization for AI, the question is for which one. The engine that weighs most depends on your category and your country, and the one that treats you worst is where the work is. Which index each one uses and which kind of page it prefers is measured in what each AI model cites.
The heatmap: where the hole is
Below the five models there is a grid of 10 columns (the questions) by 5 rows (the engines). Each cell is an answer, and the number is your position among the sources. Green if you came first, black if second or third, yellow if you were cited further down, grey if you did not appear.
The grid does one thing percentages cannot: it tells apart an absence by topic from an absence by engine. A whole grey column is a topic no model considers you a source for. A whole grey row is an engine that does not retrieve you at all. Two problems, two solutions.
Tiendamia's map is an engine problem. It has a whole grey row, Perplexity's, and scattered cells in the other four. Three of its eight citations are in ChatGPT, in the questions about buying from Amazon in Argentina, Samsung spare parts with local payment, and services that compare prices and import electronics, and in that last one it comes first. ChatGPT finds it and puts it at the top; Perplexity never retrieves it.
Mudafy's map is a topic problem. It concentrates in two columns: properties for sale in Villa Urquiza, where four engines cite it, and digital real estate agencies with up-to-date valuations, where Gemini puts it first. And five columns are grey across all five engines: rentals in Córdoba, price per square meter in Córdoba, asset regularization, investment in downtown Buenos Aires and construction loans. No model considers it a source on those, whichever one you ask.
The questions: the report is worth what the panel is worth
Everything above depends on the 10 questions. If the panel asks what your customer would never ask, the report measures something irrelevant very precisely. That is why the questions are in the report, in full, numbered the same way as in the heatmap, and why you should read them before celebrating or worrying.
The questions are long and specific on purpose, because that is how people talk to a chat. It is not "real estate Buenos Aires": it is "which digital real estate agencies make it easy to sell a property fast in Buenos Aires and offer an up-to-date market valuation". It is not "buy drone": it is "which stores sell professional drones under 250 grams with technical support in Argentina". The intent comes inside the question, and the answer has to resolve all of it.
Two readings that come out of these two panels. First: Mudafy's panel mixes property search questions with questions about UVA mortgage loans and asset regularization. In the loan questions the sources are banks and the government site, not real estate agencies, and Mudafy does not appear. If those questions are not your business, the panel is lowering your number for something you do not care about. If they are, you just found a topic where you are not yet a source. Second: Tiendamia's panel mixes product questions (a PlayStation 5, Xiaomi phones, Samsung spare parts, beauty products) with service questions (how to pay in installments, which courier clears customs, where to find coupons). In the product questions it competes against local stores and the manufacturer; in the service questions, against couriers, forums and Mercado Libre's help pages. Those are two different contests inside the same report, and it pays to decide which one you want to win before writing a single page.
The free preview uses 10 questions generated from the site's description. In AI Visibility the weekly scan covers more than 100 and you can write your own, which is how you make the panel look like what your customers actually ask.
Who shows up in your place?
The next two tables in the report are the ones that drive the most decisions: the most cited domains in your panel and the exact pages the models linked. The first tells you who you really compete with for the model's attention. The second tells you in which format the game is won.
| Panel | The five most cited domains | The brand |
|---|---|---|
| Mudafy | zonaprop.com.ar (53), google.com (40), argenprop.com (21), argentina.gob.ar (20), roomix.ai (12) | 6th with 12 citations |
| Tiendamia | aerobox.com.ar (27), youtube.com (22), instagram.com (18), reddit.com (15), mercadolibre.com.ar (12) | 9th with 8 citations |
Three things jump out of that table. First: half of the domains are not competitors. They are the government, YouTube, Instagram, Reddit, the press, and even links to Google itself, which add up to 40 citations across two questions in Mudafy's panel. A model builds its answer with whatever it finds, and in Tiendamia's panel it mostly finds videos, posts and forums. Being present there, with your own content or with mentions, weighs as much as your site.
Second: competitors you did not have on your radar show up. Roomix, in real estate, has the same 12 citations as Mudafy with a better average position, and does it with a homepage and one article about office-to-housing conversions in downtown Buenos Aires. It is a small brand next to Zonaprop or Mercado Libre, and it gets in anyway because its content answers the question.
Third, and the most uncomfortable: in the cross-border shopping category, the domain with the most citations after the couriers is Wise, a fintech, with two blog guides: how to buy from Amazon in Argentina and the customs limits. Eleven citations across seven questions. Tiendamia sells exactly that and has eight. The page that explains how it is done beats the page that sells it.
The cited pages: the format brief
The list of exact URLs is where the report stops describing and starts prescribing. Look at which of your pages were cited and which of everyone else's, and you will see the pattern without anyone explaining it.
Three Mudafy pages were cited: the homepage, the price-per-square-meter map by neighborhood in Buenos Aires, and the listing of properties for sale in Villa Urquiza. A data page and a listing page by neighborhood, both answering one concrete question. Mudafy has dozens of listings and guides, but the model chose the two that match a question in the panel. And when the panel asked about the price per square meter in Córdoba, there was no equivalent page: grey cell.
No Tiendamia page made the most-cited list: its eight citations were spread across different URLs, none with more than one. What does appear are Mercado Libre's help pages on international purchases, a third-party import tax calculator and the Wise guides. Look at the contrast with its homepage: it says "Buy from the US from Argentina, easy and safe" and repeats "Pay quick and easy with Mercado Pago or MODO" under every product, but it is a grid. When the panel asked which options exist to pay for international purchases in installments or with Mercado Pago, the answer was on that homepage, in a line repeated dozens of times, and no engine cited it: they cited whoever explains it in a paragraph. The panel asks for explanations and the category answers with guides. The winning format is in plain sight.
A useful detail from this table: several URLs cited by ChatGPT carry the
utm_source=openai parameter. It is the same marker you will see in your GA4 when ChatGPT sends
you traffic, and the most direct way to confirm that a citation turned into a visit. How to measure that
traffic is in traffic from ChatGPT, Claude and Gemini.
The opportunities and what each brand does
The report closes with three general opportunities, ordered by priority. They are not a tailored plan, they are the minimum reading that comes out of the data. With the three cases in view, this is what each one means.
High: invisible in key questions. The list of questions where you did not appear in several engines. It is the content brief. For Mudafy it is 8 of 10, and the ones most aligned with its business are comparing price per square meter in Córdoba (it already has the page for Buenos Aires: it needs the Córdoba one), finding rentals in Nueva Córdoba and Cerro de las Rosas, and properties with appreciation potential in the capital. For Tiendamia it is all 10: in every question there is at least one engine that does not cite it, and the format that wins in its category is the guide on how to import and how to pay, not the product page. The panel has already picked the first page to write: how to pay for international purchases from Argentina in installments or with Mercado Pago, the question where its own homepage had the answer and nobody cited it.
Low: strength detected. The question-by-engine combinations where you are in the top 3. Mudafy and Tiendamia have 4 each. The recommendation is to defend that content: keep it updated, do not change its URL, do not bury it behind a redesign. A page several engines already cite, like Mudafy's Villa Urquiza listing, is an asset that costs far more to build than to keep.
Medium: optimize conversion pages. Most citations in the panel go to homepages and corporate pages. The pages that capture purchase intent in a generative engine are the ones that answer the commercial question: neutral comparisons, prices, terms. In these two panels the evidence is Mudafy's price-per-square-meter map, a data page that answers one concrete question, and the Wise guides in cross-border shopping.
The model cites pages that answer the question in text. It does not cite catalogs, it does not cite a promise repeated under every product, it does not reward brand size: Roomix and Wise prove it. Visibility is built one question at a time, with one page per question, and defended by not touching what already works.
If you want to know what your own domain's report looks like, sign up for Dashcrab for free and add it: the full report has more questions, the list of competitors and the cited pages, and you can write your own prompts. And if you are at Mudafy or Tiendamia and want the full report, write to us.
The two reports were generated with Dashcrab's AI Visibility scan on August 19 and September 3, 2026, with 10 questions per brand sent to ChatGPT, Claude, Gemini, Perplexity and Google AI Mode. Model answers change over time: the numbers in this article are a snapshot of those dates, not a permanent measurement. The two brands did not take part in the study and are not Dashcrab customers.