Free audit
Glossary

What is an LLM?

Short answer: an LLM (large language model) is an AI model trained on a very large amount of text to predict and write language. It is the engine inside assistants such as ChatGPT, Gemini and Perplexity's answers, and inside Google's AI Mode and AI Overviews.

How it works, in plain words

  1. Training. The model learns patterns from huge amounts of text. What it "knows" from training stops at a cutoff date and can be wrong or out of date.
  2. Search. Many assistants also search the web while answering, then write from the pages they find. That is how a new business, or a changed phone number, can appear in an answer.
  3. Writing. The model writes an answer word by word. It can name businesses, summarize reviews and link to sources.

For local questions, step 2 matters most: an answer about "the best electrician in Raleigh" usually comes from what the assistant finds that day.

Why answers differ

Reason Effect
Different models and search sources ChatGPT and Gemini can name different businesses
Randomness in the writing The same question can give a different list a minute later
Wording "best" versus "cheapest" versus "near me" changes the list
The user's location and history Some assistants personalize

That is why one answer proves little. Ask the same question several times, on several assistants, and count.

Example

An HVAC company asks one assistant "who should I call for AC repair in Phoenix?" and is not named. It asks 12 times across four assistants and is named in 3. The useful number is 3 of 12, plus who was named instead and which sites were cited. (Example numbers.)

Common confusion

  • An LLM is not a search engine, though many assistants now use one.
  • It does not "rank" businesses the way the map pack does. There is no position to hold.
  • It can be confidently wrong. Check what assistants say about your hours, services and prices.

Keep reading