Why do AI / GEO visibility tools show completely different results for the same product?

We started looking into AI and GEO visibility out of curiosity, to understand how visible our product actually is across AI-driven surfaces.

What surprised us wasn’t fluctuation. It was how far apart the results were.

Some AI visibility tools showed almost nothing. Literally close to zero.
Others rated the same product near 15. GEO-focused tools were no more consistent, some placed us around 50, others closer to 70, using seemingly similar inputs.

At first, we assumed we were misconfiguring something. Then we thought it might be timing, models, or prompt differences. But even after normalizing for those, the gap remained.

It started to feel less like “measurement error” and more like a deeper question, are these tools actually measuring the same thing at all?

AI visibility, especially, seems unstable by nature. Small changes in prompts, regions, or model assumptions can swing results from “almost invisible” to “highly visible.” GEO tools feel slightly more grounded, but still far from consistent.

What makes this tricky is that none of the tools clearly explain what their scores are really based on. Are they approximating retrieval probability? Semantic relevance? Likelihood of being cited? Something else entirely?

At this point, the challenge isn’t any single score, but understanding how to interpret them when they diverge this much.

When AI and GEO visibility tools show very different results, how do you decide what signal actually matters?