NiubiGEO audits brand visibility in AI answers using the user's own keys

NiubiGEO is a new open-source tool that audits how AI models answer questions about a company's product, which competitors they name instead, and which sources they cite. It has picked up 185 stars in the last seven days and comes as a Docker image and a Node.js command-line runner.

What it audits

The tool takes a domain and a list of questions, runs them across the AI providers the user has keys for, and produces a report that answers plain business questions. Does AI recognise the product. How AI describes the brand. Who the confirmed competitors are. Which questions bring up competitors more often. Which important questions miss the brand. Which sources support these conclusions.

The full AI answers stay in the report, collapsed by default, so a claim about how a model treats the brand can be opened and read against the exact prompt and citations behind it. NiubiGEO separates confirmed competitors from loosely related names, and every conclusion links back to the question, the model answer, and the citation sources.

Providers it can use

The project supports OpenRouter as a single key that gives access to many models, plus the native APIs for OpenAI, Anthropic, Google Gemini, Perplexity, and DeepSeek. An OpenAI-compatible provider path for custom gateways and local services is listed as planned. Because the user brings the keys, cost and rate limits belong to the account that runs the audit rather than with the tool.

How it compares itself with commercial platforms

The README includes a side-by-side comparison with Profound, Peec AI, Otterly.AI, Semrush AI Visibility, Ahrefs Brand Radar, and AthenaHQ. Each row gives the case for the commercial product first, then the case for NiubiGEO, with the comparison dated 2026-09-03 and a note that vendor capabilities change and should be verified on each vendor's site. The README's view is that a hosted platform gives you mature workflows and proprietary data, and the self-hosted tool gives you control over questions, models, and evidence.

The scope, in plain terms

The audit ends at a report. It shows which prompts put competitors ahead of the brand and where AI answers describe the wrong product, but it makes no change to the site itself. Reports save to a local runs/ directory by default, so the raw JSON is available for follow-up work.

An AI visibility check that the team can run on its own hardware, against its own list of prompts, and read the exact model answer behind every conclusion is different from a hosted score. It makes GEO work checkable: two runs a month apart against the same prompts show whether a change to the site changed the answer, and disagreements between two providers become visible rather than hidden in an average. The pattern also matters beyond this one tool, since a repository of prompts and expected mentions is something a team can maintain and diff, in the same way they maintain and diff tests.

Source

Primary source: Albert-Weasker/niubigeo on GitHub.