How to Use AI: A Practical Guide
Use the system you actually have—not an imaginary universal AI
Start with the task
AI tools can help with drafting, explanation, summarization, coding, research, translation, image creation, data work, and many other tasks. But quality depends on the model, product, tools, context, permissions, and the work itself.
A useful workflow is:
- Define the task clearly.
- Give the system the context it actually needs.
- Specify constraints and the desired output.
- Review the result.
- Verify anything consequential.
- Iterate if needed.
Do not confuse fluency with truth
AI systems can produce polished explanations that are incomplete, outdated, or wrong. For important factual work, check the underlying sources rather than treating confidence or writing quality as evidence.
Privacy: the durable rule
Do not paste passwords, government IDs, financial account numbers, health records, other people's private data, API keys, authentication tokens, private keys, or other secrets into an AI service unless your organization has explicitly approved that exact service, account type, and workflow for that data.
There is no universal rule that “all prompts are used for training” or that “no prompts are stored.” Data handling depends on the provider, product, workspace, contract, settings, and feature being used.
Before using sensitive information, check:
- Whether data is retained
- Whether it can be used to improve or train models
- Whether humans may review it
- Where it is processed
- Which administrators can access it
- Whether your organization has approved the service
Use human review where consequences matter
The higher the stakes, the stronger the review should be. Medical, legal, financial, security, hiring, education, and safety-critical decisions can require qualified human judgment and domain-specific controls.
References
Reynolds, L., & McDonell, K. (2021). Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm. CHI EA 2021.
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., et al. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems (NeurIPS).
Zhou, Y., Muresanu, A. I., Han, Z., Paster, K., Pitis, S., et al. (2022). Large Language Models Are Human-Level Prompt Engineers. ICLR 2023.
Citation Note: Some citations are open access (arXiv or DOI, where linked). Others are books or journal articles that may sit behind a publisher paywall. Use the linked DOI or arXiv when available. If you notice any citation errors, please let us know.