A prompt that normally works starts returning "I'm not able to help with that, as I'm only a language model." The request is ordinary and clearly within what the tool does. What is this, and what does Google suggest trying?

The principle — some failures are not the prompt being wrong. Recognising a filter response saves you from rewriting something that was already correct.

Why the key is correct — Google names this pattern directly: a fallback response is one returned when either the prompt or the response triggers a safety filter, and it gives this exact sentence as the example. Its suggested move is to try increasing the temperature, because a different sampling path can produce a response that does not trip the filter.

Why the others are wrong — the sentence reads like a capability statement but is not one, and believing it retires a working prompt. It is not truncation, because the reply is a complete sentence rather than one cut off mid-thought. And it is not a clarity problem, since the same prompt worked before.

Remember this — read the shape of the failure before rewriting. Cut off mid-sentence is a length ceiling; a polished refusal to an ordinary request is a filter.

Sources — Google's Prompt design strategies.