Hi Product Hunt 👋 I built citeshine-audit because I wanted to know how AI assistants describe a product, and whether they get it right, with data I could check myself instead of a score. The first thing I ran it on was my own brand-new company. Results from logged-out ChatGPT web, 13 prompts, once each (2026-10-13): - 0/11 mentions on generic buyer questions - Asked by name, ChatGPT said our domain was a 2021 Shopify phone-case store. That was the domain's previous owner. The report from that run (per-prompt counts and excerpts) is in the repo as an example. The raw answer file for that run isn't published. What it does: - API engines for OpenAI, Perplexity, DeepSeek, Kimi, Grok, Claude and Gemini (your own keys). They are implemented but not yet tested end to end, and none has been used for a published run. So far only the ChatGPT web engine has. - Web-UI mode that drives your own Chrome: ChatGPT (used for the run above), plus experimental Perplexity and Kimi. One tab, human pace, stops at the first captcha or login wall - Mention rate, first-mention order, list rank, share of voice, cited domains - `raw.jsonl` with every answer, plus Markdown/HTML reports What it doesn't do (yet): dashboards, scheduling, alerts, sentiment scores or recommendations. Hosted tools like Profound, Peec AI, Otterly and AthenaHQ cover much of that. This is for people who want the raw numbers. Feedback I'd love: which engines or metrics are missing, and how you'd like brand matching to work beyond substring + aliases. I'll re-run the same prompts with more runs per prompt and post the comparison either way. Disclosure: I also run Citeshine (citeshine.com), a GEO service that helps Chinese SaaS and AI-tool teams selling overseas find and fix what AI gets wrong about them. The tool is free and MIT either way.

citeshine-audit

Find what AI gets wrong about your brand

Find what AI gets wrong about your brand