Fine-tuning
Fine-tuning is extra training that adapts an AI model that's already been trained to a narrower job, like a company's support questions.
Fine-tuning is extra training given to an AI model after its main training, to adapt it to a narrower job. The model already knows language in general. Fine-tuning shows it many examples of one kind of task, so it gets better at that task.
Marketers come across the term because it sounds like a way to change what ChatGPT or Gemini says about your brand. It isn't one, and it helps to know why.
How it works
Training a large model from scratch takes huge amounts of text and computing power. Fine-tuning starts from that finished model and trains it a little more on a much smaller set of examples.
For example, a coffee company could fine-tune a model on a few thousand of its past support emails, so that its own help chat answers questions about grind sizes and delivery in the company's tone. The model keeps everything it knew before, and picks up the habits in those emails.
What it can't do for your brand
When a business fine-tunes a model, it gets its own copy. The public assistants people use, like ChatGPT, Gemini and Perplexity, are run by the companies that make them, and only those companies can retrain them.
So you can't fine-tune your way into their answers. What you can change is what they read: your own site, and the independent pages about you that they search and cite.