The objection usually goes: “AI assistants change too fast. Models update constantly, every assistant is different, and what works today may not work in six months. Let’s wait until it settles.” It sounds prudent. It assumes that acting early carries risk and that waiting carries none.
What changes when assistants update, and what doesn’t
A new model reasons better and has newer training data. Assistants change how they search the web and which sources they show. Answers can shift from week to week. All true.
What doesn’t change is what answers are made from: pages that explain things clearly, accurate facts about your products, and other reputable sites that talk about you. Work on those holds up across model versions, because it’s about what’s on the web, not about one model’s quirks.
- Your own pagesFacts in plain text, accurate prices and stock, structured product data. Assistants with live search read these today and will tomorrow.
- Other people’s pagesReview lists, retailers, magazines and communities. These feed both what models learn and what they find while answering.
Why constant change is a reason to measure
If answers shift from week to week, you can’t know where you stand from a one-off check, and you certainly can’t know from not checking. Change is the argument for measuring carefully: enough prompts, several asks each, a likely range on every rate, and the same prompts repeated over time so you can see what’s moving.
Waiting also has a cost you can’t see. While you wait, the people asking assistants for recommendations are hearing about whoever is named today.
A small first step
Starting doesn’t need a large programme.
- MeasureWrite prompts the way people ask, mostly without naming you. Ask several assistants several times and see where you’re named, who is named instead and what the answers cite.
- Fix one gapPick the clearest gap: a missing fact, a wrong price, a review list you’re not on. Fix what you can on your own pages.
- Measure againAsk the same prompts later and compare the ranges. If they overlap, the change isn’t established yet, so keep measuring rather than drawing conclusions.