LLM Watch, November 2026: What the Latest AI Models and Studies Mean for AI Search Visibility
Six short notes from the November 7 LLM Watch update: the top AI models, model prices, AI search clicks, where AI answers come from, AI writing and one task for this month.
This week in one line
AI answers now come before links, and your own website is their main source.
The notes below come from the November 7 update of LLM Watch, our tracker of AI models and AI search research. Each note ends with what it means for your leads.
Top AI models: three labs within 3 points
Google, Anthropic and OpenAI now lead the Artificial Analysis Intelligence Index v5. The next company is 7 points behind.
- Google Gemini 4 Carbon
- 60
- Anthropic Claude Opus 5.5
- 59
- OpenAI GPT-6 Astra
- 57
- Meta Muse Wonder Full
- 50
- Kimi Kimi K3.1 Pro
- 46
- xAI Grok 4.7
- 35
- Qwen Qwen3.8 Max
- 34
- DeepSeek DeepSeek V4.1 Flash 1017
- 33
- GLM GLM-5.3
- 32
- MiMo MiMo-V2.6-Pro
- 30
- Mistral Mistral Large 4
- 26
- Moonshot opened the weights of Kimi K3.1 Pro on November 4. Mistral did the same for Mistral Large 4 on October 29.
- The newest models go to small groups first: Gemini 4 Carbon, GPT-6 Cyber and GPT-6 Rosalind.
Why it matters: No single AI tool wins. Your buyers use different tools, so check your visibility in more than one.
AI model prices: 100 times apart
- GPT-6 Luna (OpenAI): $0.10 per 1 million input tokens
- Claude Haiku 5.5 (Anthropic): $0.10 per 1 million input tokens
- MiMo-V2.6-Flash (MiMo): $0.14 per 1 million input tokens
- Qwen3.8-Flash (Qwen): $0.15 per 1 million input tokens
- GLM-5.3-Flash (GLM): $0.15 per 1 million input tokens
- DeepSeek V4.1 Flash 1017 (DeepSeek): $0.30 per 1 million input tokens
- MiMo-V2.6-Pro (MiMo): $0.435 per 1 million input tokens
- Mistral Large 4 (Mistral): $0.70 per 1 million input tokens
- Gemini 3.8 Flash (Google): $0.75 per 1 million input tokens
- GPT-6.1 Terra (OpenAI): $1 per 1 million input tokens
- Muse Spark 1.3 (Meta): $1.25 per 1 million input tokens
- DeepSeek V4-Pro-0813 (DeepSeek): $1.32 per 1 million input tokens
- GPT-6.1 Sol (OpenAI): $2 per 1 million input tokens
- Claude Sonnet 5.5 (Anthropic): $2 per 1 million input tokens
- Gemini 4 Carbon (Google): $2 per 1 million input tokens
- Gemini 4 Argon (Google): $2 per 1 million input tokens
- Grok 4.7 (xAI): $2 per 1 million input tokens
- Qwen3.8-Max (Qwen): $2 per 1 million input tokens
- Kimi K3.1 Pro (Kimi): $2 per 1 million input tokens
- Kimi K3 (Kimi): $3 per 1 million input tokens
- Claude Opus 5.5 (Anthropic): $4 per 1 million input tokens
- Muse Wonder (Meta): $4 per 1 million input tokens
- GPT-6 Astra (OpenAI): $10 per 1 million input tokens
- Claude Fable 5.1 (Anthropic): $10 per 1 million input tokens
Lowest: GPT-6 Luna and Claude Haiku 5.5 at $0.10. Highest: GPT-6 Astra and Claude Fable 5.1 at $10.
Cheap models put AI answers in more places: site chats, browser assistants and AI agents. Our own site chat runs on Claude Haiku 5.5 with a spend cap of $5 a day.
Why it matters: More software reads your website and makes choices for a buyer. Each one uses only the facts that it can read.
AI search: more AI answers, fewer clicks
- 88% of commercial keywords showed a Google AI Overview in August 2026, up from 45.75% in September 2025. Semrush via Search Engine Land, October 22, 2026. 400 US keywords in banking, ecommerce and software.
- 18.8 points fewer clicks to other websites in Google AI Mode. Wang, Gleason, Bart, Wilson and Metaxa, arXiv 2608.18352, August 18, 2026. Field experiment with 1,100 US Google users.
- About 4 sources per AI Mode answer, down from about 16. Only 9.3% of the pages cited in the summer kept their citation. Maeva Cifuentes, The Search Spot (Scrunch data), October 23, 2026. 8 B2B software brands, 407 prompts.
- 13.8% of ChatGPT visitors converted, against 3.02% from Google. But ChatGPT sent less than 1% of Google's visits. Marie Haynes via Search Engine Land, October 21, 2026. GA4 data from 100 websites over 12 months.
Why it matters: If your leads went down while your rankings stayed the same, check AI answers first.
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.
Where AI answers come from: your website
About 78% of AI citations go to company websites. Small brands start behind.
- Global brands
- 73%
- Regional brands
- 44%
- Niche brands
- 11%
Two fixes have proof:
- Structure: a change to page structure alone raised citations 17.3% across six AI engines. (Yu, Yang, Ding and Sato, March 31, 2026)
- Facts: the pages that shaped AI answers most had more facts to quote. (Zhang, He and Yao, April 28, 2026)
Why it matters: The main source of AI answers is a website that you already own. Fix it first.
AI writing: models sound more alike
- Models from the last three years give less varied answers over time. (Patel, Crossman, Aggarwal and Wenger, August 19, 2026)
- About 16% of the sources that AI search tools cited were AI-generated. (Allaham and Diakopoulos, May 22, 2026)
- Rules added one turn at a time cut quality: models kept only 71.2% of what one full brief gave. (Singh, Eyasir, Yaqoob and Manavalan, September 27, 2026)
Why it matters: Your own facts, stories and voice make your content different. Give AI one full brief.
This month's task: check 20 buyer questions
- Write 20 questions that buyers ask before they buy. Include your city.
- Ask each one 3 times in ChatGPT, Perplexity and Google AI Mode.
- Record who gets cited and which sources the answers use.
- Fix your own site: services, prices, area and hours in plain text, question headings and FAQ schema.
- Repeat next month and compare.
Start with step 4 on your home page. The free AI Visibility Score runs 37 checks in about 20 seconds and shows what to fix first.
Do you want us to run all 180 answers for you and 3 competitors? See the AI Visibility Benchmark ($499, delivered in one week).
Sources
- A/GEO Research, LLM Watch and Dashboard View, checked November 7, 2026
- Artificial Analysis, Intelligence Index v5, November 7, 2026
- Semrush via Search Engine Land, October 22, 2026
- Wang, Gleason, Bart, Wilson and Metaxa, AI in Search Reduces Publisher Referrals Without Improving User Experience, arXiv 2608.18352, August 18, 2026
- Maeva Cifuentes, AI Mode Cut 70% of Its Sources. Here's What It Kept., The Search Spot (Scrunch data), October 23, 2026
- Marie Haynes, ChatGPT sends less than 1% of Google's traffic: Data, via Search Engine Land, October 21, 2026
- Pratyush Kumar, Generative Engine Optimization at Scale, June 18, 2026
- Yu, Yang, Ding and Sato, Structural Feature Engineering for Generative Engine Optimization, March 31, 2026
- Zhang, He and Yao, From Citation Selection to Citation Absorption, April 28, 2026
- Patel, Crossman, Aggarwal and Wenger, Are LLMs becoming similarly creative?, August 19, 2026
- Allaham and Diakopoulos, Synthetic Sources?, May 22, 2026
- Singh, Eyasir, Yaqoob and Manavalan, The Effects of Incremental Instruction Delivery on Language-Model Creative Writing, September 27, 2026