What it means for your advertising
An LLM learns the patterns of language from text such as books, articles and websites, then writes a reply a word or so at a time, each the most likely continuation. That's why it can explain, summarise and draft so fluently, and why it is sometimes confidently wrong (hallucination).
What a model learns in training stops at a cutoff date. To answer questions about now, such as which businesses near you offer a service, AI search tools also search the web and write from what they find, citing it. This is called grounding, or retrieval-augmented generation (RAG), and it's where most of the chance to be mentioned comes from (GEO).
LLMs also power AI agents, and the tools ad platforms offer for writing ads and finding audiences.
An example
Asked about office coffee, ChatGPT's model already knows from training that a 30-person office needs a grinder and regular deliveries. It can't know which Manchester roasters deliver this month, so it searches the web, finds Java Roasters' office coffee page and names it in the answer, with a link.
Java Roasters is a fictional business, and its figures are illustrative.
How we handle it
AI helps us plan campaigns: the Guru uses it to turn a plain description of your customers into an audience profile (described audience), and our team checks what it suggests before anything goes live. More on ChatGPT advertising.