A foundation model is a large, general purpose AI model, such as GPT, Claude, Grok, Gemini or Llama, that other products are built on. Most AI recruiting tools rent a foundation model and add their own instructions, data, screens and workflows on top.
AItopic
2 minread
2related terms
In one line
A large general purpose model, such as GPT, Claude, Grok, Gemini or Llama, that other products are built on. Most AI recruiting tools rent a foundation model and add their own instructions, data and screens on top.
Foundation models are trained on very large amounts of text and other data so they can handle many tasks without being built for any single one. The same model can write a job ad, summarize a call or translate a message. Software companies reach these models through an API and pay per use.
When a recruiting vendor says it has AI, it usually means it sends your data to one of these models with a carefully written set of instructions. The value the vendor adds is the workflow, the data connections and the guardrails, not the model itself.
Because the model sits underneath, its strengths and limits pass through to the product. If the model makes things up, the product can too, unless the vendor grounds answers in your records and builds checks around them.
Why it matters to recruiters
Knowing which foundation model sits under a tool helps you compare vendors honestly and ask better security questions. It tells you where candidate data travels, whether it is used for training, and why two tools that look different may give very similar results. It also explains why prices and features shift when the underlying model changes.
An example from a recruiting desk
Two AI sourcing tools are on an agency owner's shortlist. On the demo call she asks each vendor which model they use, where data is processed, and whether customer data trains the model. One vendor answers clearly in writing. The other is vague. She runs a two week trial with the first and adds the data terms to her vendor file.
Tips that help
Ask every AI vendor which foundation model or models the product uses.
Get in writing whether your candidate data is used to train any model.
Ask what happens to output quality when the vendor switches models.
Judge tools on workflow and data access, since many share the same model.
Mistakes to avoid
Assuming a tool built its own AI when it actually rents a foundation model.
Skipping the data processing questions because the vendor is a well known brand.
Frequently asked questions
Does a foundation model know my company's data?
Not by default. It only sees what the tool sends it in each request, unless the vendor connects it to your ATS, CRM or documents. That connection, called grounding, is what lets it answer from your own records.
Is a large language model the same as a foundation model?
Large language models are the most common type of foundation model. Foundation model is the broader term and also covers models that work with images, audio or several types of data at once.
Should my team pick one foundation model?
Most teams use what their approved tools provide. If you buy a general assistant, pick one your security team approves and train people on it, rather than letting everyone use a different free chatbot.
