From Frustrating to Useful: Getting Better Answers from AI

Have you tried using generative AI tools like ChatGPT, Copilot, or Gemini, and been underwhelmed? You might have found the responses generic, hallucinatory, or unusable without significant rewriting. Did you know most large language models (LLMs) offer persistent settings or personalization settings to customize all responses?

Note: Always check your institution’s policy on approved AI tools and usage. Avoid entering confidential, protected, or personally identifiable information unless your institution has approved the tool and appropriate safeguards are in place.

Have you tried using generative AI tools like ChatGPT, Copilot, or Gemini, and been underwhelmed? You might have found the responses generic, hallucinatory, or unusable without significant rewriting.

Did you know most large language models (LLMs) offer persistent settings or personalization settings to customize all responses?

Personalizing an LLM does not replace your subject matter expertise, professional judgment, nor does it remove the need for editing and validation. It can reduce frustration by helping the AI better understand your role, expectations, and preferred communication style, allowing you to spend less time revising responses.

In short: this is not about writing better prompts—it is providing context about you.

Both examples use the same prompt. The only difference is that the second response comes from an AI tool that has been personalized.

The core question: What should your AI tool know about you?

Personalization works best when you tell an LLM who you are (role/persona and context), how you work (style and preferences), and what matters (tone and content). Providing your role allows it to infer expectations and responsibilities. Stating your desired writing style and audience will reduce enthusiastic agreement and produce more acceptable language, while setting content boundaries will get you straight to what you really need.

Here's how four common IR roles might personalize an AI tool:

1. I work in assessment, supporting learning outcomes and program review. Use cautious, improvement-focused language. Avoid claims of impact unless supported by evidence.
2. Write in a professional, neutral tone with clear, non-technical language suitable for executive audiences. Start with a short summary. Avoid emojis, em-dashes, and conversational fillers. Keep explanations concise and structured. Highlight assumptions, risks, and decision points.
3. I am an Institutional Researcher at [institution]. Preferred tools are Python and SAS. Use an analytical writing style and be precise with language. Clearly distinguish between findings, interpretations, and assumptions. If evidence is weak or missing, state that directly. Note how results could be misinterpreted, small n sizes, definition issues, data limitations, and competing explanations. Ask clarifying questions before making assumptions about institutional definitions or data structures.
4. I create T-SQL code and Power BI dashboards. Prefer short paragraphs and bullet points. Note assumptions about model design. Don't remove or alter my comments when analyzing or rewriting code. Identify performance trade-offs when multiple approaches are possible. Note any ambiguities or gaps when creating technical documentation.
5. Whichever AI tool you can use—we all know the IT department has the final say—take those few minutes to significantly reduce editing, re-prompting, and tone correction. The time savings allow you, the expert, to focus on interpretation and analysis, rather than cleanup.

Try it now: personalization by tool

(external links)

ChatGPT

Claude

Copilot

Gemini

Hilary Carter is the Assistant Director of Business Intelligence at the University of Alabama at Birmingham. She oversees data architecture and modeling, reporting and dashboard development in support of institutional research, reporting, and analytics, with a focus on translating complex data for varied audiences.