GEO / AI Brand Mention Check Report

I developed an analysis tool that turns SEO keywords into user-question prompts and measures how often a target brand and its competitors appear in GPT and Gemini responses.

It quantifies brand mention rates and share of appearances in AI search experiences such as ChatGPT and Gemini—visibility that search rankings alone cannot show.

Background

As people turn to AI chat and AI search alongside traditional search engines, brands need to know not only their search ranking but also whether they are named in AI-generated answers.

The tool converts high-intent SEO keywords into AI-ready questions and aggregates brand appearances in the responses.

Because AI responses change from one question to another, tracking them manually is difficult.

I built an automated pipeline covering keyword processing, LLM response collection and analysis, and report generation.

My Role

I handled the planning, requirements definition, design, implementation, testing, and report design for this project.

The project took approximately 50 hours to complete and supports both Japanese and English.

Pipeline Architecture

The workflow is divided into three Python modules.

• kw-to-prompt: Translates keywords, generates questions, supports review, and translates questions into English.
• prompt-to-answer: Retrieves GPT and Gemini responses, extracts brands, and determines target-brand visibility.
• answer-to-report: Aggregates results by model, creates consolidated comparisons, and generates HTML and PDF reports.

What the Report Measures

The report aggregates the following metrics.

• Total Prompts: The total number of question prompts included in the analysis.
• Target Brand Mentioned Prompts: The number of prompts whose responses mentioned the target brand.
• Target Brand Mention Rate: The share of all prompts with a response mentioning the target brand.
• Brand Appearances: The number of responses in which each brand appears at least once, shown as a ranking.
• Model Comparison: A side-by-side Yes/No comparison of target-brand mentions for every prompt in GPT and Gemini.

Technical Specifications

The tool is implemented in Python 3.12+.

It uses uv for environment and package management. Its modular structure also makes each part of the workflow easy to update.

• OpenAI / Gemini APIs: Batch retrieval of model responses to questions.
• pandas: Data loading, shaping, missing-value handling, and appearance-rate calculations.
• Jinja2: Produces 16:9 HTML slides from templates.
• Playwright: Loads generated HTML in Chromium and converts it into landscape 16:9 PDF slides.
• pytest / ruff: Testing and code-quality checks.

Design & Implementation Considerations

I designed the tool for repeated use in real projects rather than as a one-off analysis.

Keyword processing, LLM response collection and analysis, and report generation are separate modules, so individual steps can be rerun and new models can be added.

The HTML output uses a 16:9 slide format, helping preserve the layout in both browsers and exported PDFs.

Data Anonymization

The sample report and this case study anonymize all real company, client, and product names, as well as links to the code repository.

All brand names are replaced with generic labels such as "Target Brand" and "Competitor A" to L.

The brand mention rate measures only whether a brand appears in an AI response. It does not measure positive recommendations, recommendation rank, or purchase outcomes.

Business Use Cases

This monitoring report can support decisions such as the following.

• Measure brand awareness in AI search on a regular basis.
• Compare share of mentions with competitor brands to understand positioning in AI search.
• Identify keyword groups and question categories where the target brand is not mentioned.
• Measure change before and after Generative Engine Optimization (GEO) initiatives.