250,000 prompts analyzed across 250 companies in LatAm: the guide to improving your AI visibility
How to measure and improve visibility in ChatGPT, Gemini and Claude with Lumos, using GEO methodology and comparable metrics for marketing.
By Pablo Arroyo · LinkedIn · Published · Updated
Lumos was born in January 2026. Since then we have analyzed more than 250 companies and over 250,000 prompts, so what follows is what we have learned working with companies and partners across LatAm during these months.
A few months ago, we were with Macal, one of our first clients, reviewing how ChatGPT responded to "what are the best options in Chile" for their category. The brand, a clear leader in its market, simply did not appear. The AI was recommending competitors, several of them foreign and not even operating strongly here. That is when we realized the hypothesis we had was indeed valid: companies no longer just want to show up on Google, they need to appear in AI answers, and that is new terrain with its own rules. How things went for Macal afterward is something I cover in their full success story.
I'm Pablo, founder of Lumos. This guide brings together what we have learned measuring and improving brand visibility in ChatGPT, Gemini and Claude. No empty promises and no magic formulas: the concept made clear, how the engines work under the hood, a comparison table, and the metrics that actually matter for marketing.
Why now: 2026 changed the rules
2026 marked a turning point in search. Two announcements, months apart, finished changing the rules of the game.
The first: at Google I/O 2026, Google made its AI Mode the default search experience, on desktop and on mobile. They described it as the biggest change to search in over 25 years. The classic ten blue links did not disappear, but they were relegated to a "Web" tab, and the analyses point to zero-click rates close to 93%. The consequence is direct: the goal is no longer to rank first but to be cited inside the AI answer. If you are not there, you are invisible even if you hold position one. (Sources: Google I/O 2026; TechCrunch, "Google Search as you know it is over," May 2026.)
The second: ChatGPT started showing advertising. OpenAI launched its ads in February 2026 and already has a self-serve ad manager. The ads appear in labeled boxes at the end of the answer and do not alter the content. What does this mean for you? That AI has stopped being an experiment and become a real commercial channel, and that being in the organic part of the answer, the part people read and trust, is worth more than ever. (Source: OpenAI, "Testing ads in ChatGPT," 2026.)
In practical terms: your customers are migrating to asking AI, Google is pushing them there, and the space is already starting to be monetized. The window to win organic visibility, before it fills up, is now.
What is AI search visibility
AI search visibility is the degree to which your brand appears and is cited within the answers generated by engines like ChatGPT, Gemini, Claude and Google AI Overviews. It is no longer just about being in a list of ten blue links: now what matters is that the model mentions you, summarizes you, or recommends you.
Three things worth keeping clear:
- It is presence and attribution: how and when you are named, whether there is a clickable link, and how prominently you appear in the answer.
- It is not just "first place." It covers summaries, cards, carousels, source panels, and even the questions the model suggests afterward.
- It applies to everyone: B2B and B2C brands, media, institutions, digital commerce, and services.
And how is it different from traditional SEO and AEO? In that the decision happens inside the generated answer, not in an organic listing. AEO seeks direct answers within the search engine; AI visibility spans the entire conversational and generative world (ChatGPT, Gemini, Claude and Google AI Overviews), each with its own way of citing.
How the engines operate and where your brand appears
The flow, broadly speaking, is always similar: the user asks, the model interprets, sometimes it goes out to look for sources on the web, drafts an answer, adds citations or cards if appropriate, and presents it with follow-up questions.
Whether your brand appears depends on three things, no more and no less:
- That your information is retrievable (that the bot can read it).
- That it is trustworthy (signals of authority, data, references).
- That it is useful for solving what the person is asking.
It is worth keeping in mind that not all engines respond the same way:
- ChatGPT sometimes answers from memory (its internal knowledge) and sometimes goes out to browse and adds citations. When it answers from memory, it cites no one, and that explains why your content appears some days and not others.
- Gemini relies fairly heavily on web results and tends to show sources when it retrieves from the web.
- Claude leans toward more explanatory answers and shows citations when it activates search.
- Google AI Overviews blends the AI summary with the usual search engine, with linked cards inside the SERP.
This last point is key, and I repeat it because it tends to be hard to grasp: when the model answers from its internal memory, it cites no sources. Your content only competes when the engine decides to go out and search. That is why visibility looks intermittent, and why the goal is to raise the frequency with which you appear, not to expect 100%.
Comparison table: engines and answer surfaces
So you can see it at a glance, here is how the main engines behave today in terms of how they respond, how often they cite sources, and how they handle Spanish in LatAm.
| Engine | How it responds | Citation/link frequency | Format | Spanish in LatAm |
|---|---|---|---|---|
| ChatGPT | Conversational; lists steps and comparisons | Variable: high when browsing, low when answering from memory | Text with contextual citations, sometimes link previews | Supported, broad availability |
| Gemini | Summaries with bullets and alternatives | High when retrieving from the web | Text plus source cards on the side or at the bottom | Supported, rollout varies by country |
| Claude | Detailed and explanatory, attentive to nuance | Medium to high when search is activated | Text with numbered citations or at the end | Supported, expanding coverage |
| Google AI Overviews | Summary integrated into the search engine, with steps and recommendations | High, via linked cards | Text plus cards and carousels inside the SERP | Supported, market availability evolving |
This changes often: interfaces, policies and citation frequency shift month to month. It is worth reviewing periodically by engine and by country to keep the comparison current and catch changes in time.
Metrics that actually matter for marketing
The cases where AI visibility moves the needle for you are, above all, discovery ones: when someone has not yet chosen a provider and asks AI about the problem or the category. Also "X vs Y" comparisons, how-to guides, and local or regulatory questions.
To measure it for real, these are the metrics we use:
- Share of answer: how much of the answer is attributable to your brand within the block.
- Presence rate by intent: in what percentage of the queries on a topic you appear.
- Citation rate with link: of those appearances, how many bring a clickable link.
- Prominence: how high up and with how much space you appear relative to other sources.
And there is one distinction that to me is the most important: separating branded from non-branded. Queries with your name help you defend yourself, but they rarely bring you new customers. Queries without a brand (the generic ones, the problem-based ones) are the ones that open the top of the funnel. If you only look at branded ones, you will be left with a snapshot that is not real.
Here is a real example: The Qualis, a Chilean premium clothing brand, started with non-branded visibility of barely 2.1%. Three months later, working on citable content and making the site readable for bots, it reached almost 15%. There were no tricks: it was closing concrete gaps, one by one, and measuring every week. That is what makes the difference.
And it is not an isolated case. Macal, a leader in real estate auctions in Chile, went from being nearly invisible in AI answers to tripling its non-branded visibility in three months. We measured it and worked on it week by week, just like with The Qualis. You can see the full Macal success story here.
If you want to compare metrics across engines, regions and languages with rigor, we have published the full methodology (sampling, reproducible prompts, surface labeling) at trylumos.ai/methodology.
How to win visibility without tricks
I'll be honest: there are no shortcuts. What works is largely common sense, but it has to be done well.
On content, cover the full intent: the problem, the solutions, the comparisons, the implementation, and even the ROI. The models reward content that closes the decision cycle, not the kind that leaves the reader hanging.
On authority, show what you are talking about: your own data, a clear methodology, quality references, and examples applied to your industry. Real experience gets cited far more than filler.
On format, write so AI can extract from you: a summary at the start, bullets, short definitions, glossaries, and tables. The clearer the structure, the easier it is for the model to take a fragment of yours and attribute it to you.
And on freshness, keep everything up to date and localized for Chile (regulations, terms, prices). No promises you cannot back up: besides being the right thing to do, it is what sustains your visibility over time.
How to choose an AI visibility platform
If you are going to evaluate a tool, ask it to cover what actually matters:
- True multi-engine coverage: ChatGPT, Gemini, Claude and Google AI Overviews, not just one.
- Serious measurement: reproducible prompts, clear sampling windows, control by region and language, and that they explain how they record attribution with and without a link.
- Useful segmentation: branded vs non-branded and by surface type, so you can prioritize where to invest.
- LatAm context: Spanish, local sources and competitors, not a US tool translated.
And ideally, one that does not stop at diagnosis. Measuring is fine, but what moves the needle is executing the improvements. That is exactly what we do at Lumos.
Frequently asked questions
How do AEO and GEO differ?
AEO seeks direct answers within the search engine. GEO goes further and works on how the generative engines like ChatGPT, Gemini and Claude cite and recommend you. If the decision happens inside the model's answer, you need GEO.
Does AI visibility replace SEO?
No, it complements it. SEO makes you findable on the web, and that feeds the models. GEO works on the new layer: how AI recommends you. The ideal is to look after both.
Can you optimize specifically for ChatGPT, Gemini or Claude?
There is no per-engine trick. What works across all of them is verifiable, well-structured, up-to-date content, with authority and local context. Each engine weighs things differently, but the foundation is the same.
How often do answers change and how should you read the fluctuations?
They change often, and they vary by region and language. To read them well, use sampling, moving averages, and segment by intent and surface. Never draw conclusions from a single query.