Teia Studio's GEO IGO platform audits ChatGPT, Gemini, Claude, and Perplexity in real time — and the pilot data raises serious questions about how much companies can trust what generative AI says about them.

by Kasun Illankoon, Editor in Chief at Tech Revolt

[For more news, click here]

There is a question that almost every marketing executive, hospital administrator, and brand manager in the world has quietly started to wonder about, and almost none of them have a reliable way to answer: when someone asks ChatGPT or Gemini about their company, what exactly does the AI say?

It is a reasonable thing to wonder. Generative AI has moved, with remarkable speed, from a curiosity to a primary information source for millions of people making decisions about products, services, healthcare providers, and institutions. Unlike a Google search, which returns a list of links and leaves the interpretation to the user, a generative AI model delivers a single, confident answer presented as settled fact. There is no page two. There is no competing narrative sitting one click away. There is just the response, offered with the rhetorical authority of something that has processed the entire internet.

The problem is that the response is not always accurate. And until very recently, no one was systematically checking.

That is the gap that São Paulo-based startup Teia Studio has set out to close. This week, at Web Summit Rio 2026, the company is presenting Teia GEO IGO, a platform it describes as the first real-time auditing system for what generative AI models say about brands, institutions, and services. The name stands for Intelligence Governance and Observability, and the distinction it draws from the more familiar concept of Generative Engine Optimisation is both deliberate and instructive.

Teia Studio is presenting its GEO IGO platform at Web Summit Rio 2026, competing in the summit's premier PITCH competition.

GEO, as a discipline, has grown quickly in recent years alongside the rise of AI-powered search. It essentially asks the same question that traditional SEO asked about Google: how do you make sure your brand appears prominently in the results? Teia Studio's founders argue that the discipline has a significant blind spot. Optimising for visibility in AI-generated responses is only half the equation. The more urgent question is whether what the AI actually says, once it does mention you, is accurate.

"The GEO market already arranges companies' digital storefronts to appear in AIs, but until now, no one has been checking in real time what these AIs are actually telling people. That is exactly what Teia GEO IGO measures," explains José Vásquez, founder and Chief Data and AI Officer of Teia Studio.

The Audit Gap Nobody Wanted to Talk About

The platform operates on a continuous weekly cycle. It begins with a company's official website, already indexed by Google, as its verified source of truth. Standardised questions are then sent simultaneously to ChatGPT, Gemini, Claude, and Perplexity. Every response is logged in full, including the date, time, and specific model, and the system then audits the results against the official record: whether the brand was mentioned, whether the information is accurate or fabricated, how consistently the four models agree with each other, and how the narrative shifts over time.

To translate those findings into something actionable, Teia Studio developed what it calls KAPIs, or Key Algorithmic Performance Indicators, a proprietary framework that converts AI perception into auditable metrics. The methodology has been submitted for patent protection with Brazil's national intellectual property body and is grounded in a peer-reviewed scientific paper published on Zenodo.

The framing of IGO as a distinct discipline rather than an extension of GEO is worth taking seriously. The company's argument is that the market has invested heavily in understanding how to get brands into AI responses without investing anything comparable in understanding what those responses actually contain. That asymmetry, Teia Studio contends, represents a material risk that most organisations have not yet quantified.

What the Pilot Data Actually Found

Between March and April 2026, Teia Studio ran a structured pilot programme with four real brands across four sectors, auditing each against the four AI models in weekly cycles. The results from the public health sector case study, which covered 640 analysed interactions across two audit cycles, are the figures the company is now presenting publicly, and they are worth examining carefully.

The audit detected 58 hallucinations in total, of which 18 were classified as high-severity, meaning the AI had generated information that was entirely invented. Between the first and second week of the audit, the total number of hallucinations increased by 1,250 percent, jumping from four occurrences to 54. Perplexity accounted for 83 percent of the severe hallucinations, generating fabricated medical registration numbers, incorrect geographic locations, non-existent websites, invented scientific studies, and false contact information.

"With TEIA GEO IGO, companies begin to see, in a structured way, what AIs are saying about them — and how that narrative compares to their competitors. This completely changes strategic decision-making capabilities across marketing, reputation, and commercial operations." — José Vásquez, Founder and CDAO, Teia Studio

Perhaps the most structurally significant finding is not the volume of errors but the degree of inconsistency across models. In one out of every three queries, the four AI systems gave contradictory responses to the same question about the same institution. On average, the models mentioned the brand in only 44 percent of responses to questions that were explicitly about it. In none of the cases did any model offer a caveat about the possibility of inaccuracy.

For a hospital, the implications are direct and serious. Fabricated medical registration numbers and invented contact information are not abstract reputational concerns. They are the kind of errors that could lead patients to the wrong clinic, contact the wrong professional, or act on fabricated clinical guidance. The healthcare sector is simply the most visible illustration of a problem that applies with varying degrees of severity across every industry vertical in which generative AI is now consulted as a primary information source.

A New Discipline for a New Information Environment

There is a broader context in which Teia Studio's platform lands at a particularly significant moment. The conversation around AI governance has, for most of the past two years, focused primarily on the training side of the equation: what data large language models are trained on, whether that data is licensed, what biases it may encode, and how models should be regulated at the point of creation. Comparatively little structured attention has been paid to the output side in real-world commercial contexts: what these models say about specific entities, how often they are wrong, and how those errors propagate through the information environment before anyone notices.

Teia Studio is positioning IGO as the discipline that addresses that gap. The proposition is not that generative AI is unreliable and should therefore be distrusted. It is that the outputs of these systems, when they relate to specific brands, institutions, or professionals, require continuous monitoring in the same way that any other public-facing communications channel does. The fact that the channel is automated and operates at scale makes the case for monitoring stronger, not weaker.

Based on its audit findings, the Teia GEO IGO platform also generates prescriptive guidance, recommending specific technical actions on official domains and social profiles to help AI models recognise authoritative content as a primary source. The approach treats AI hallucinations not as inevitable noise but as a correctable signal, one that becomes correctable only once it has been systematically measured.

Teia Studio is competing in the PITCH competition at Web Summit Rio 2026 and has been selected as an ALPHA Startup for the event.

For US brands and enterprise technology leaders watching from abroad, the Teia Studio story carries a specific relevance. The AI monitoring problem is not a Brazilian problem or a Latin American problem. It is a global one, and the organisations best positioned to navigate the next phase of the AI information environment will be those that invest as seriously in understanding what AI says about them as they have in understanding how to appear in what AI says. That infrastructure is only now beginning to be built.

Related Articles:

Sovereign AI Is Moving From Government Policy to Enterprise Procurement and This Partnership Shows How

Europe's Largest Private Tech Provider Is Betting on the Middle East's Infrastructure Moment

Before Physical AI Can Scale, Someone Has to Fix the Data Layer