Over a few weeks you notice ChatGPT's replies have changed in tone, and a phrasing trick you relied on no longer produces the same result. Which reading of the situation is best supported?

The principle — ChatGPT is a moving product, and a workflow tuned tightly to one model's habits is resting on something that was never promised to stay still.

Why the key is correct — OpenAI states plainly that it regularly updates ChatGPT by introducing new models and retiring older ones, that these transitions can sometimes feel disruptive, and that changes in how a model responds, even when subtle, may affect workflows or how you experience conversations. It also states what does survive: your prior conversations can carry over to newer models, including background you have shared about your interests, preferences and goals, so while tone or style may change, that retained context keeps the conversation grounded. Both halves of the observation are accounted for, which is what makes this the best-supported reading.

Why the others are wrong — a restricted tier is not how any of this is described, and a plan change would be visible in your subscription settings rather than inferred from tone. Memory personalises rather than overriding what you have explicitly asked for, and you can read exactly what it holds in the memory summary, so a change felt across many kinds of request is not well explained by it. And a phrasing trick that stops working is the ordinary consequence of a different model having different habits, not evidence that your wording was targeted.

Remember this — this is the practical argument for stating what you want plainly rather than relying on a clever phrasing that happens to work today. Explicit instructions about format, audience and length survive a model change; an incantation tuned to one model's quirks does not. OpenAI also points to personality and customisation features as the supported way to shape responses over time, which is the durable version of what a phrasing trick was doing informally.

Sources — OpenAI's What to expect when models change article.