Short answer: eleven direct model channels sampled via their model APIs plus a Brave-search-grounded GPT-5.6 channel. Nothing else — and we won't pretend otherwise.

Covered today

  1. GPT-5.6 (OpenAI, direct) — the model family behind ChatGPT, via its model API.
  2. Claude Sonnet 4.6 (Anthropic, direct) — via its model API.
  3. Gemini 3 Flash (Google, direct) — via its model API.
  4. Claude Opus 4.6 (Anthropic, direct) and Gemini 3.1 Pro (Google, direct) — frontier-tier cross-model comparison.
  5. DeepSeek V3, Kimi K3 (Moonshot), GLM-5.2 (Zhipu) — leading models in the Chinese AI market, via their model APIs.
  6. Llama 3.3 70B, Qwen3 30B, Mistral Small 3.1 — open-weights channels for broader comparison.
  7. GPT-5.6 + Brave Search (grounded) — the model answering with live Brave web results, approximating how retrieval-augmented assistants compose answers.

Not covered

Model-API answers are not the same as the consumer apps: ChatGPT web, Gemini app or Perplexity answers add retrieval, memory and interface layers we cannot reproduce through an API. We do not query consumer surfaces automatically — ChatGPT web is sampled through a manual human-collection channel, and Perplexity, Copilot and AI Overviews are not covered. If a tool tells you it measures "your visibility in ChatGPT" it should be able to show you the actual answers it sampled — we hold ourselves to that standard, so every answer in your report is inspectable and labeled with the exact channel it came from.

Why the results still generalize

  • Assistants share their diet. Search-grounded assistants build answers from live web results; our grounded channel reproduces that mechanism with real Brave results. If your site and the sites that cite you are legible to this pipeline, you are structurally positioned for other grounded assistants too — they retrieve from the same web.
  • Direct-model presence reflects your public footprint. Models internalize the public web during training. A brand invisible to eleven independent direct channels usually has a footprint problem that affects all models, not a vendor-specific one.
  • The failure modes are shared. Blocked crawlers, missing structured data, absent FAQ content, no presence on cited directories — the issues our diagnosis and Citation Gap surface hurt you on every AI surface.

What does not generalize: exact mention rates. Treat your numbers as a rigorous sample from defined channels, not a universal "AI score".

Roadmap

Two expansion routes are already in motion. First, a manual human-collection channel: a human operator runs your prompts in real consumer interfaces (starting with ChatGPT web) in a normal browser, weekly — those samples appear in reports explicitly labeled as manual. Second, China's leading AI assistants (Doubao, Kimi, DeepSeek and others) are on the coverage route, following the same rule. Fully automated coverage of consumer surfaces (ChatGPT web, Perplexity, Gemini) remains on the public roadmap. When a channel ships you'll see it labeled in reports — we add channels by actually querying them, never by extrapolation. Watch the changelog.