Picture this: you're sitting across from a client — a dentist, a contractor, a boutique law firm — and you've just told them their AI visibility score improved this month. They ask the natural follow-up: "How do you know?"
If the honest answer is "we asked an AI and it said so," you have a problem. Not a small one.
AI assistants are remarkable. They're also, by design, probabilistic. They don't retrieve a stored fact and hand it to you — they generate the most statistically likely next token, given everything they've been trained on. Most of the time that produces something useful. Some of the time it produces something confidently, fluently wrong.
The industry has a word for it: hallucination. The AI companies themselves have a word for it too — they put it in their own terms of service. "May produce inaccurate information." "Verify important facts independently." The disclaimers are right there in the fine print of the very tools some people are using to measure AI visibility.
And it gets more specific than just wrong answers. Ask the same question on Tuesday. Ask it again on Thursday. You may get a different business recommended, a different reason given, a different confidence level expressed — with no change to the underlying data at all. That's not a flaw to be patched. That's how the system works.
Variability like that is not a basis for a client recommendation. It's not a basis for a score. It's not something you can put in a report and defend.
The AI Ranking Factors framework is built on a different foundation entirely. Every one of the 141 factors we test returns a binary answer: true or false. Either the signal is present or it isn't. Either the structured data is valid or it has errors. Either the NAP information matches across sources or it doesn't.
That test runs the same way every week. The result doesn't shift based on phrasing, timing, or model temperature. If a factor passed last Tuesday and it passes this Tuesday, the answer is the same — and if it changed, it changed because something in the real world changed, not because the measurement tool was having a different day.
That's what deterministic means: the same input produces the same output, every time. It sounds obvious. In GEO and AEO, it's rarer than you'd think.
There are a few moments in a client relationship where you absolutely cannot afford to be fuzzy:
Deterministic testing makes every one of those conversations easier. You're not hedging. You're pointing at the data.
Let's be clear: the entire point of the AI Ranking Factors framework is to help local businesses show up better inside AI assistants — ChatGPT, Gemini, Perplexity, and the rest. We spend a lot of time understanding how those systems process information and what signals they trust.
But understanding AI and using AI as your measurement instrument are two entirely different things. A structural engineer studies how earthquakes behave. She doesn't use an earthquake to measure a building's foundation.
The signals we test — schema markup, citation consistency, content structure, authority indicators, and 136 other factors — are the inputs that AI assistants pull from when they decide which businesses to surface. Our job is to audit those inputs rigorously, not to ask an AI if it likes you today and report back.
When the framework flags that a business is missing a critical structured data type, that's a concrete finding. When the citation audit shows an address mismatch on three directories, that's a fixable problem with a before and an after. When the score moves month over month, every point of movement traces back to specific factors that passed or failed — not to a model's mood.
That's the kind of reporting that builds trust over time. Not because it always shows progress, but because it always shows the truth.
Your client hired you to tell them what's real. The least you can do is make sure your tools are held to the same standard.
— J. Brent Tuttle