How to use artificial intelligence responsibly: tips for prompts and verification

Writing clear instructions and checking responses are key practices that differentiate responsible AI use from over-reliance. Recommendations from companies like OpenAI and research from Microsoft help turn these practices into everyday habits for professionals. This article provides guidance on creating effective prompts and exploring AI tools.

Five practices for creating better prompts

A prompt is the instruction a user gives to an AI system. OpenAI’s prompt guide outlines six strategies, five of which can become simple habits:

• Be specific: Provide context, target audience, and desired format. Example: “Summarize this report in five lines for the financial board and highlight the three main risks.”
• Provide reference material: Ask the AI to base its answer on a specific text and cite the passages used. Example: “Use only the document below and indicate where you got each piece of information.”
• Break down complex tasks: Complex activities often lead to more errors. Suggest steps, such as first asking for a list of topics and then a draft for each one.
• Ask for reasoning: Instruct the AI to solve the problem before concluding and review if anything was missed. Example: “List the assumptions used and point out what may be missing.”
• Test changes: An adjustment may improve a specific case but harm the overall outcome. Compare different versions of the prompt using multiple examples to evaluate the impact.

Exploring AI tools

General conversational assistants like ChatGPT, Gemini, and Copilot are useful for generating drafts, summaries, and ideas. Google’s NotebookLM operates differently: its responses are based on sources provided by the user, with citations of passages that allow for accuracy verification.

A practical example is sending three quarterly reports and asking: “According to the sources, what changed in costs between the first and third quarter?” Google notes that it’s possible to enable or disable sources to specify which ones the answer should be based on.

A simple exercise is to send the same prompt to two different tools and compare their clarity, errors, and cited sources.

Verification: the habit that complements usage

A study by Microsoft Research in partnership with Carnegie Mellon surveyed 319 knowledge professionals and found that greater confidence in AI is associated with reduced critical thinking. According to researchers, the focus shifts to verifying information, integrating responses, and supervising tasks.

Three questions help apply this verification: Does the data match the original source? Is the text appropriate for the target audience? Who is responsible for the final result?

An experiment published in PNAS involving nearly a thousand students showed that those who used an open AI and requested ready answers performed 17% worse on a test without the tool. Those who used a version that guided reasoning achieved results similar to the control group.

What’s worth remembering

The recommendations presented are based on best practice guides, not controlled tests. Performance varies depending on the tool and task. Across the board, sources emphasize the importance of clear instructions, reference material, breaking tasks into steps, and verifying results before use.