I built an open-source platform to see how ChatGPT, Perplexity and Gemini talk about your brand

When someone asks ChatGPT "what's the best tool for …", your product is either in the answer or it isn't. Classic SEO tools can't tell you which, and the tools that can ("AI visibility", "GEO", "AEO") are mostly closed SaaS priced per prompt.

So I built AutoSEO, an MIT-licensed platform you can self-host that puts AI visibility and classic SEO in one place.

👉 GitHub: https://github.com/codextde/autoseo

What it measures

You define the prompts your buyers ask ("best CRM for small agencies", "is X better than Y"), pick markets and engines, and AutoSEO runs them on a schedule across 16 AI engines: ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and AI Mode, Copilot, Grok, Mistral, DeepSeek, Meta AI, Qwen, Kimi and a few more.

For every answer it extracts:

• Mentions and position: are you named, and where in the answer
• Citations: which URLs and domains the engine cites, and whether any are yours
• Competitors: every other brand mentioned, so you get share of voice and head-to-head comparisons
• Sentiment: what the engine praises or criticises, on an 11-aspect scorecard
• Query fan-outs: the background searches engines run before answering, which turn into content ideas

Next to that there's the SEO toolbox you'd expect: keyword research, rank tracking, a crawler-based site audit with Lighthouse, backlinks, Search Console and GA4, plus AI bot traffic from your server logs (GPTBot, ClaudeBot, PerplexityBot…).

Architecture choices that might interest you

Bring your own AI subscription. Most AI work (analysis, content drafts, simulated engines) can run through a small local agent that drives your own Claude Code or Codex CLI. The agent only makes outbound HTTPS requests (no open ports), runs every job in a fresh CLI session in its own temp folder, and updates itself from signed releases. API keys for Anthropic, OpenAI, OpenRouter, Perplexity, Gemini and others are the fallback.

Honest data sources. Engine answers come from DataForSEO, the vendor APIs or the local agent. If an engine has no configured provider, it can be simulated by a model with web search, and those results are labelled as simulated everywhere they appear.

Everything is an API. There's a REST API (OpenAPI spec included) and an MCP server with 100+ tools over streamable HTTP with OAuth 2.1, so you can ask Claude Code or Cursor "which prompts did we lose to competitor X this week?" and it queries your own instance.

One env var to deploy. Next.js 16, React 19, PostgreSQL 17 and Drizzle, shipped as a single Docker image (ghcr.io/codextde/autoseo). You set DOMAIN, and everything else (email, AI providers, OIDC SSO, roles, limits) lives in an admin panel. Migrations run on boot.

curl -fsSL https://autoseo.codext.de/install | bash
# or: docker compose up (see deploy/ in the repo)

After setup you can open a demo project with 90 days of generated data, so you can click around without spending provider credits.

What's next

The public repo is brand new, so I'm looking for feedback, especially on:

• measurement methodology (prompt sampling, reporting run-to-run variance)
• which engines and integrations you'd want next
• the self-hosting and upgrade experience

If you'd rather not run a server, there's a hosted version (AutoSEO Cloud, $50/month), which is how development gets funded. The self-hosted version has every feature.

⭐ If this is useful, a star on GitHub helps other people find it.