LLM Watch Launches, and the AI Models Behind AI Answers
Our first issue: LLM Watch goes live, 18 AI releases in three weeks, small models that make decisions, and one task: write your 20 buyer questions.
This week in one line
The AI tools that your buyers use change every few weeks. Track the tools, not every model.
This is the first issue of A/GEO Research Weekly. Each week we send short notes on new AI models and new studies on AI search. Each note says what it means for your leads.
LLM Watch: one place for the models behind AI answers
Today we launched LLM Watch and its Dashboard View. LLM Watch tracks the major AI companies, their current models and their public benchmark scores. It also lists studies on how AI tools pick their sources. Every number links to its source.
Why it matters: These models write the answers in ChatGPT, Claude, Gemini, Meta AI and Grok. When a buyer asks for a business like yours, the model decides which businesses to cite.
New models: 18 releases in three weeks
Our release log has 18 entries from September 21 to October 12. The main ones:
- Anthropic: Claude Opus 5.5 (September 22), Claude Sonnet 5.5 (September 28) and Claude Haiku 5.5 (October 7).
- OpenAI: GPT-6 Luna (September 22) and GPT-6.1 Sol (September 29).
- Google: Gemini 4 Argon (September 30) and the Nano Banana 2.1 image model (October 6).
- Meta: Muse Wonder and Muse Wonder Full (today).
- Others: Grok 4.7 from xAI (September 21) and Mistral Large 4 (October 6).
Why it matters: Your buyers do not all use one AI tool. A check in one tool does not tell you what the others say.
A new kind of model: small models that decide
Three companies released small models that make one fast choice, for example yes or no, or which option fits a rule:
- Amazon released Strands Decider 2B (October 1).
- OpenAI opened the Decisions API as a beta (October 6).
- Microsoft released Microsoft-Decision-1 (October 9).
Why it matters: AI agents use fast choices to filter options for a buyer. An agent can only pick from the facts that it can read on your website.
Subscribe to A/GEO Research
Get one short email each week: new AI models, new studies on AI search, and one task for your website. Each note says what it means for your leads.
Study of the week: AI Mode cuts clicks by 18.8 points
Researchers put 1,100 US Google users into three groups for seven days: normal Google, Google with the AI features hidden, or AI Mode only. AI Mode cut clicks to other websites by 18.8 percentage points. People in AI Mode also trusted Google less.
An earlier Pew Research Center study points the same way. Pew tracked the browsing of 900 US adults. People clicked a result link on 8% of visits with an AI summary, and on 15% of visits without one.
Why it matters: Fewer people click through to a website. The AI answer itself must cite your business and state your facts.
This week's task: write 20 buyer questions
- Write 20 questions that buyers ask before they buy. Use their words, not yours.
- Include your city or area. Example: "Who is the best plumber in San Jose for a slab leak?"
- Keep the list in one sheet. Next week we use it.
Want to see the models behind the answers? LLM Watch shows the current models, their scores and the latest research.
Sources
- A/GEO Research, LLM Watch release log, checked October 12, 2026
- Wang, Gleason, Bart, Wilson and Metaxa, AI in Search Reduces Publisher Referrals Without Improving User Experience, arXiv 2608.18352, August 18, 2026
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 22, 2025