In a chained sequence of prompts, what defines the relationship between one step and the next?

The principle — Chaining converts one hard request into a pipeline of easy ones, where each stage starts from a concrete artefact rather than from an intention.

Why the key is correct — Google describes chaining as making each step a prompt and chaining the prompts together in a sequence, stating that the output of one prompt in the sequence becomes the input of the next, and that the output of the last prompt is the final output. Anthropic describes the most common instance of this pattern as self-correction: generate a draft, have the model review it against criteria, then have it refine based on the review. Each stage is a separate call, which is what makes the intermediate result inspectable.

Why the others are wrong — Resending one prompt is retrying rather than chaining. Re-issuing the original request at every step rebuilds the overload that decomposition removed. And answering all the steps together is simply the large prompt again.

Remember this — A chain passes work forward. If nothing produced by step one enters step two, it is not a chain.

Sources — Google, Prompt design strategies.