Hundreds of professionals have been publicly embarrassed after trusting generative AI; don’t be next.
My article on prompts showed some ways to help shape AI responses, provide more guidance for better results, and avoid some levels of hallucination. However, prompts are not enough. Research identifies significant risks inherent to AI output (Bender et al. 2021). Examples in the media include lawyers submitting court briefs that cite non-existent cases (Adkisson 2025). Between public failures and personal experience, more than 50% of Americans are more concerned about AI than excited by it, and only about 10% are more excited (Pew 2026). However, the main problem isn’t that people are using AI; it is that they are likely using the tool incorrectly.
After the Prompt: A Practical Approach
Prompts are the first part to interacting with AI, but prompts won’t solve the problem with the next part: the output. “[P]ractical usage patterns often show a preference for efficiency over explainability, with users favoring quick, actionable solutions from AI systems” (Spatola 2024). In other words, there is a great tendency to take the first answer as the answer.
AI models predict likely sequences of words based on patterns in training data; they cannot verify truth claims. According to Bender and Gebru (2021), responses with errors and fabrications can still sound logical and authoritative. Treat every AI assertion as a hypothesis, not fact. If a claim cannot be verified and traced to reliable sources, don’t use it.
To counter built-in biases and illogic, start a new query with a neutral or adversarial tone. Prompt for explicit assumptions and counterarguments. An example of this could be, “List three reasons whatever-suggested-claim might be wrong and provide linked sources.” Significant changes in answers can indicate uncertainty. This also gives the user additional information about the sensitivity the model may have in the prompt phrasing.
Several models include project folders to store documents and media specific to a project. Use these folders to archive your exchanges for reference. They can be useful when creating prompts in the future.
AI can accelerate professional workflows. Use some of that saved time to verify outputs, as that is where the real value is.
References
Adkisson, Jay. 2025. “Lawyers Caught Submitting AI Briefs Face Worse Than the Court’s Monetary Sanctions.” Forbes, October 27, 2025. https://www.forbes.com/sites/jayadkisson/2025/10/27/lawyers-caught-submitting-ai-briefs-face-worse-than-the-courts-monetary-sanctions/.
Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” Proceedings of the 2021 Conference on Fairness, Accountability, and Transparency, 610–623.
Merriam-Webster.com Dictionary, s.v. “slop,” accessed June 3, 2026, https://www.merriam-webster.com/dictionary/slop.
Pew Research Center. 2026. “Key Findings About How Americans View Artificial Intelligence.” March 12, 2026. https://www.pewresearch.org/short-reads/2026/03/12/key-findings-about-how-americans-view-artificial-intelligence/
Spatola, Nicolas. 2024. “Practical Usage Patterns in AI Systems: Efficiency over Explainability.” Patterns 5 (7): 100879. https://doi.org/10.1016/j.patter.2024.100879