aeo

About

Elmo is an open-source, self-hosted platform for optimizing your AI visibility, which is also known as:

• Answer Engine Optimization (AEO)
• Generative Engine Optimization (GEO)
• LLM Optimization (LLMO, which is where the name Elmo is from)

Elmo tracks how AI answer engines like ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews mention, cite, and describe your brand, so you can benchmark competitors and grow your visibility in AI answers.

It's a free alternative to tools like Profound, Peec, and Otterly. You can run it on your own infrastructure, own your data, and audit exactly how every metric is calculated.

Demo

Try the live demo at demo.elmohq.com to see how Elmo tracks prompts and analyzes citations.

Screenshots

What Elmo Tracks

• Visibility scoring — For every tracked prompt, Elmo measures how often each AI answer engine mentions your brand and trends the score over time, per prompt and per model.
• Brand mention tracking — Each answer is checked for your brand name, aliases, and domains, so mentions are counted the same way every time, across every engine.
• Sentiment and position — Every answer that names your brand or a competitor is read for its sentiment toward that company (positive, neutral, mixed, or negative) and where it ranks among the tracked companies named. Sentiment rolls up into a 0–100 score per prompt, per model, and against each competitor, with the exact passages behind it.
• Responses — Search every AI answer for your brand, filter by model, tag, prompt, or sentiment, and click a company's badge to highlight the passage its sentiment was read from.
• Citation analysis — Every URL an engine cites is stored with its domain and position, then categorized: your own domains, competitor domains, social media, Google properties, and institutional sources. You see which pages AI models actually trust for your industry.
• Competitor benchmarking and share of voice — A leaderboard of competitor mention rates next to your own, with your overall share of voice and how it moves over time.
• Query fan-out analysis — When an answer engine grounds a response, it often runs several web searches first. Elmo records those searches, shows how engines rewrite your wording (which words they add, drop, or keep), and which searches your content wins or misses.
• Prompt management — An onboarding wizard analyzes your website, then suggests keywords, competitors, buyer personas, and tracking prompts. You can also add prompts manually, tag them, and enable or disable them individually.
• Opportunities — Elmo turns your tracked visibility and citation data into a prioritized, regularly refreshed list of what to create, pitch, and seed to earn more AI citations.
• Reports — Shareable AI visibility reports (visibility summary, per-prompt breakdown, citation analysis, competitor comparison) viewable by stakeholders without an Elmo account.
• API, SDKs, and MCP — A REST API (with TypeScript and Python SDKs) for brands, prompts, competitors, analytics, sentiment, citations, runs, opportunities, and reports, plus an MCP server so Claude Code, Cursor, Codex, and VS Code can query your data and manage prompts. See the API reference and MCP docs.

How Elmo Measures AI Visibility

Elmo's methodology is simple and fully inspectable in this repository:

1. Prompts are defined per brand — either generated by the onboarding wizard from your website or written by hand. Each prompt is a question a potential customer might ask an AI assistant.
2. A background worker runs each prompt on a schedule (several times a day by default) against every engine you configure. Scraping providers capture the real consumer surfaces — ChatGPT, Google AI Mode, Google AI Overviews, Gemini, Perplexity, and Microsoft Copilot — while direct model APIs (OpenAI, Anthropic, Mistral, OpenRouter) add coverage for Claude, Grok, and other models with web search enabled. The providers guide compares both approaches and their costs.
3. Every answer is normalized and parsed. Each run produces the answer text, the list of cited URLs, the web searches the engine ran while grounding (its query fan-out), and the model version. The text is scanned for your brand's name, aliases, and domains, and for each competitor's, and Jev reads each answer that names one to decide whether it really means that company, its sentiment toward it, and the passage that shows it.
4. Everything is stored in PostgreSQL, including the raw engine output, so any metric can be re-derived and audited later. Nothing is a black box: the mention detection, scoring, and aggregation code is all in this repo, and every judgment keeps the passage it rests on.
5. Metrics are aggregated into trends. Visibility is the share of runs that mention your brand; share of