Foundation model (also called a base model) is a large AI model trained on broad, general-purpose data that serves as the starting point for building specialized AI products. GPT-4, Claude 3, Gemini, and Llama are foundation models — they’re trained on vast amounts of internet text to develop general language understanding, then adapted (through fine-tuning, RLHF, or system prompts) into specific products like ChatGPT or Google AI Overviews.
Foundation model → Product pipeline
Foundation model (pre-training on broad web data)
↓
Alignment tuning (RLHF, safety training)
↓
Fine-tuning (optional — specialized domain or behavior)
↓
System prompt configuration
↓
RAG / retrieval integration
↓
End-user AI product (ChatGPT, Perplexity, AI Overviews, etc.)
Each layer in this pipeline can affect how your brand is represented. A brand visible in the foundation model’s training data has a baseline representation. Fine-tuning and system prompt configuration can amplify or suppress that representation in specific deployments.
Why the foundation model distinction matters
Base model vs. fine-tuned model: The same foundation model fine-tuned for different purposes can behave very differently. GPT-4 fine-tuned for a medical platform may suppress commercial brand recommendations entirely; the same base model powering a general assistant will recommend brands freely.
Model updates change your baseline: When OpenAI releases GPT-4o or Anthropic releases Claude 4, the underlying foundation model changes — training data, size, and alignment all shift. Your brand’s representation in that model may change with it, even if your own content hasn’t changed. This is the source of unexplained visibility shifts that sometimes appear in monitoring data after a major model release.
Open-source foundation models: Llama (Meta), Mistral, and others are released openly, allowing any organization to build products on top of them. As more AI search products are built on open-source models, your brand’s visibility in those deployments depends on whatever fine-tuning and retrieval configuration the builder applies — creating more fragmentation in AI brand visibility.
Monitoring across foundation models
LLM Metrix tracks visibility across the primary consumer-facing AI products, which represent different foundation models at different stages of their development pipeline. When a model provider announces a major model update, expect a monitoring review — that’s the right moment to audit whether your brand’s position has shifted and why.