CONCEPT
Generate-First Protocol
The learning-preserving intervention requiring users to produce their own attempt—however incomplete or incorrect—before receiving AI assistance, ensuring the generation event and cognitive network activation occur even when the machine stands ready to provide instant answers.
The generate-first protocol is the simplest and most direct application of desirable-difficulties research to AI-augmented work. The procedure is straightforward: before consulting an AI tool, the user must generate their own response from their own cognitive resources. The quality of the generated attempt is not the criterion—it may be partial, rough, or wrong. The cognitive operation is the criterion: did the user activate their knowledge network, attempt retrieval from memory, engage in the reconstructive effort through which encoding occurs? If so, the subsequent AI-provided answer lands on prepared ground, processed more deeply than it would have been if received without the prior generation attempt. The protocol preserves the generation effect—the finding that produced information is remembered better than received information—in an environment designed to eliminate production through instant provision. Implementation requires overriding the natural preference for immediate answers and the institutional pressure for efficient output, making it psychologically difficult and organizationally costly despite being technically trivial.
