AI Visibility maturity: ReBo's 4-level framework

The ReBo framework classifies a company's AI Visibility maturity in four levels, from Level 1, Invisible to AI, to Level 4, AI Reference. Each level describes how far AI systems can read, understand, cite and trust what the company publishes.

Is there a framework to measure a company's AI Visibility maturity?

Yes. The ReBo framework measures AI Visibility maturity in four levels: Invisible to AI, Partially Discoverable, AI Friendly and AI Reference. The classification starts from measurement: target questions run in AI engines and a technical audit of the site.

The four levels follow the four verbs that govern an AI answer. At Level 1, the engine cannot read the content. At Level 2, it reads fragments without understanding enough to use them. At Level 3, it understands and starts to cite. At Level 4, it comes to trust the company as a source on the topic.

The framework is the yardstick of the methodology, and the methodology is the six delivery workstreams of the ReBo method. The levels measure how far the workstreams have progressed. The underlying concept is in what is AI Visibility.

What are the 4 levels of AI Visibility maturity?

Level 1, Invisible to AI: the AI cannot find the content. Level 2, Partially Discoverable: it finds fragments. Level 3, AI Friendly: it reads, understands and starts to cite. Level 4, AI Reference: it cites the company often, ahead of competitors.

Source: ReBo AI Visibility framework, 2026.

At Level 2, Partially Discoverable, part of the content is reachable: a corporate site, a few social posts. The AI knows the company exists and lacks the depth to cite it as a source of analysis. In answers, it shows up at most as a passing mention.

At Level 3, AI Friendly, content lives in static HTML with clear hierarchy, structured data, an FAQ and visible dates. The structure lets a system extract question, answer, author and date without ambiguity, which is what a generated answer needs to cite a source.

At Level 4, AI Reference, verifiable authority is added to the technical base: named authors with credentials, proprietary data, consistent presence across platforms and third-party mentions. This is the level the method pursues for each client, in continuous operation.

What defines a company that is Invisible to AI (Level 1)?

The company's content sits in formats AI reads poorly or cannot reach: PDF, images, pages assembled by JavaScript, material behind a login. On questions about its market, the AI cites competitors, press and public bodies, and the company is left out.

The AI crawlers that feed assistant search, such as GPTBot, ClaudeBot and PerplexityBot, do not run JavaScript. To them, a site that assembles its text in the browser arrives empty. The same goes for monthly content published only as PDF, covered in why AIs can't read PDFs.

A second, silent cause is one site owners rarely see. A CDN or firewall rule can block those crawlers even when robots.txt allows them. The site opens normally in a browser, and the crawler gets an error.

The symptom in real use is the same for both causes. The company produces quality content about its own market, and the AI answers that market's questions citing other names. The 20-minute test shows whether that is the case.

How do you move from Level 1 to Level 3 in 90 days?

With the method's first three workstreams in sequence: a diagnostic with a measured baseline, technical optimization of the site and transformation of existing content into pages AI can read. Ninety days is ReBo's working target, measured against the baseline.

The first workstream, the diagnostic, freezes the target questions and measures the baseline in each engine before anything is published. The second, technical optimization, makes text readable without JavaScript, allows AI crawlers at the server and CDN and declares JSON-LD and llms.txt.

The third workstream, data transformation, takes the existing archive out of PDF and publishes it as pages with a question, an anchor answer, an author and a date. The company does not need to write more: what it already produces starts to exist in a format machines read.

ReBo does not guarantee a position or a citation deadline, because the final choice belongs to the engine. What gets measured, month by month, is progress against the baseline. Level 4 is the compound effect of workstreams 04 to 06, with a target of 6–12 months of continuous operation. The metrics are in how AI Visibility is measured.

What does a research firm need to do for its research to be cited by AI?

Publish each report also as an HTML page, with one question-form heading per topic, a direct answer in the first paragraph, a named analyst and a stated date. The PDF still goes to subscribers, and the page exists for machines to read, excerpt and cite.

A research firm's output is often the most citable content in financial markets, and also the most locked away. A report in PDF, sent by email or kept behind a login, does not enter the competition for citation, however good the analysis is.

The path that preserves the subscription model is publishing the durable part on an open page. The structural thesis, the sector glossary and the valuation methodology become pages with permanent URLs. The full report, with price target and recommendation, stays with subscribers.

Authorship weighs heavily in a regulated sector. The analyst's name, a stated certification, a date and the regulatory notice visible on the page give the model what it needs to trust the source. The conversion step by step is in from PDF to AI-friendly page.

Are investors using AI to choose an asset manager?

AI use is already the majority in Brazil: 77% of Brazilians use or have used an AI platform, according to Bain, fieldwork in February 2026. There is no public Brazilian measurement yet on choosing an asset manager, and ReBo treats that link as a hypothesis.

Source: Bain & Company, Consumer Pulse Brasil 2026, published May 18, 2026. Accessed September 21, 2026.

What ReBo measures directly is the other end: whom the AI cites when the question is about an investment topic. In the September 15, 2026 test, the same question about Brazilian private credit, asked of five AIs in a private session, returned the same two names in all five answers: XP and Fitch.

For an asset manager, the practical consequence does not depend on settling the hypothesis. If an investor asks an AI about the topic the manager specializes in, the answer already picks its sources. The full market analysis is in AI Visibility for financial markets.

Keep reading

To find out today which level the company is at, run the AI citation test. To see how each workstream moves a company from one level to the next, read the ReBo method. For a classification by an analyst, with a technical audit and a baseline, ask for the diagnostic.