Correct, and deliberate. Generated content is cached against a fingerprint of the product’s own data. If nothing about the product has changed, asking again returns the stored text rather than generating it afresh, so it costs nothing from your allowance.
Check these in order
Has the product changed?
Change the product, or change the length, tone or model tier you are asking for, and the fingerprint changes, so the app generates properly instead of serving stale text.Did the last attempt fail?
One exception is worth knowing: if a previous attempt fell back to placeholder text because the generation itself failed, that fallback is never treated as a cache hit. The next attempt genuinely regenerates.Did you change your brand voice?
New brand voice examples change the fingerprint for product descriptions, so the next description is generated fresh.
How to fix it
Nothing needs fixing: this saves you allowance. If you want different wording for an unchanged product, change the tone, length or model tier and generate again, or use Edit wording to adjust the text yourself.
If it still is not working
If a changed product keeps returning the old text, tell us which product as described in Still stuck.
Related articles