Probability-based engine
A probability-based engine writes its answers by predicting likely words, so the same question can get a different answer each time.
A probability-based engine is an AI system that writes its answers by predicting which words are most likely to come next. ChatGPT, Perplexity, Gemini and the other AI platforms all work this way. They don't look up a stored answer. They write a new one every time.
This changes how you should measure yourself. One answer tells you what the platform said once. It doesn't tell you what it usually says.
How it differs from a search engine
A classic search engine matches the words in a query to pages in its index and ranks them. Type the same query twice and you'll usually see the same list.
A probability-based engine builds its reply word by word, choosing from many likely options at each step. A small change in the question, or in the pages it reads first, can send the answer down a different path. Even the same question, asked twice in a row, can come back with a different list of brands.
An example
Ask "What's the best coffee subscription for espresso lovers?" on Monday and the answer might name Kestrel Roasters, Fernhill Coffee and Harrow & Pine. Ask again on Tuesday and it names Kestrel and two brands you've never heard of. Nothing about Fernhill changed. The model just took a different path.
Change one word, so it asks about espresso beginners, and the answer can change again.
What it means for tracking
Because each answer is a sample, you need many of them before you can trust a number. One answer is noise. A week of answers starts to show a pattern.
This is why AnswerOrbit asks each of your prompts once a day on each platform, and bases every number on the last seven days. When we compare two weeks, we only mark a change when it's bigger than chance alone would usually make it. See How we count.