Answer Volatility

Answer volatility is the tendency of AI assistants to give different answers to the same question asked twice. Ask which physiotherapist to see in a given city, ask again an hour later, and you may well get a different set of names with different sources behind them. This is not a fault awaiting a fix. It is a property of how these systems generate language.

Several things drive it. The models carry deliberate randomness in how they compose text. Which pages get retrieved at the moment of the question can vary. The systems themselves are updated constantly, and different assistants disagree with one another in the first place. The result is a readout that moves whether or not anything about you has changed.

The practical consequence is that a single check tells you very little. Look once and find yourself absent, and that is weak evidence. Look once and find yourself named, and that is weak evidence too. Only a pattern across repeated observations over a stretch of time means anything — which is precisely what one anxious evening cannot produce.

This is also why patience belongs to the method rather than to your personality. After you change your site, the pages must be crawled again, and the new content has to work its way into how these systems represent you. Through that stretch the ordinary wobble is larger than any improvement you have made, so early checking cannot separate progress from noise. Fix the thing, then leave it alone long enough for the answer to mean something.

The human translation is asking three people at a bus stop to recommend a restaurant. You would not conclude much from any one reply, and you would not go back every morning to ask again.