Responsible Use of AI Chat Tools

A practical staff guide for the Warnell School of Forestry and Natural Resources

This guide is a practical companion to UGA policy and guidance. It does not replace university policy, a supervisor’s direction, course rules, sponsor requirements, research protocols, or professional standards. When they differ, the controlling requirement takes priority.

Start here

• Keep a person responsible for every decision and every final product.
• Do not enter sensitive or restricted UGA data unless the tool and use have been specifically approved.
• Use AI to assist thinking and production, then check the result against reliable sources.
• Choose the least resource intensive model and reasoning level that can do the task well.
• Stop, summarize, and begin a fresh chat when the conversation becomes long, confused, or shifts to a new purpose.

A five question check before you begin

How an AI chat session works

A large language model, or LLM, produces a response by predicting useful sequences of words from patterns learned during training and from the material available in the current chat. It does not understand a subject in the same way a person does, and a confident answer can still be incomplete, biased, outdated, or wrong.

Why a chat can drift

Each follow-up adds material to the session. The tool may give too much weight to an earlier instruction, mix two projects, repeat an error, or lose important details as the conversation grows. A new chat with a clean summary often produces a more focused result.

What the tool may not know

• Current facts unless the tool has access to reliable, up-to-date sources.
• Local Warnell procedures or the intent behind a decision unless you explain them.
• Whether material is confidential, copyrighted, restricted by a sponsor, or covered by a course rule.
• Whether a citation, quotation, calculation, species identification, regulation, or policy statement is accurate.

University rules come first

UGA guidance makes users responsible for the integrity and accuracy of their work. It also states that permission to access UGA data is not, by itself, permission to place that data in an AI tool. If AI informs a decision, explain how it was used; do not allow AI to make automated university decisions. UGA guidance also calls for disclosure when AI helps create public-facing content.

Use AI as a force multiplier

AI is most useful when it expands a staff member’s capacity to explore, organize, draft, compare, or check work while the staff member supplies the purpose, subject knowledge, judgment, and accountability. Start with a small, reversible task and compare the result with your normal process.

The task test

A task is a better candidate when it is low risk, easy to check, reversible, and based on information you are permitted to use. The need for human review rises with the consequences of an error.

A simple pilot approach

1. Choose one repetitive or time-consuming task with low consequences if the first draft is imperfect.
2. Record the current time, quality requirements, and review steps.
3. Try the task with non-sensitive information and a clear prompt.
4. Measure time saved after checking and correcting the output.
5. Keep the approach only if quality, accessibility, and accountability are maintained or improved.

Create prompts that produce useful work

Good prompting is clear assignment design. The goal is not a magic phrase. Give the tool enough context to understand the work, define the result, and know how its work will be checked.

The CARE prompt pattern

Reusable prompt template

Weak and improved prompts

Improve the session in small steps

• Ask for an outline or questions before requesting a full draft.
• Provide one good example when format or tone matters.
• Correct errors explicitly and restate the controlling instruction.
• Ask the tool to separate facts from suggestions and identify assumptions.
• For important work, request a verification checklist rather than accepting a confident answer.

Choose the model and reasoning level

Model names and limits change. Use the categories shown in your approved tool rather than relying on a particular product name. Start with the least intensive option that reliably meets the task. Move up when the work is genuinely complex or the smaller option fails a clear quality check.

Use a lower level to prepare higher level work

A smaller model can help define the problem before an expensive or limited model performs the hard analysis. This works best when the preparation itself is easy to review.

1. Ask a smaller model to turn your notes into a structured brief with goals, constraints, known facts, unknowns, and desired output.
2. Review and correct that brief yourself.
3. Give the corrected brief and source material to the stronger model for analysis or drafting.
4. Use the smaller model again for low-risk formatting only after the substantive work is approved.

When to move up or down

Maintain usage thresholds

Usage limits vary by tool, license, model, and time. A practical threshold is a team rule for when to continue, simplify, switch models, start a new chat, or stop. The aim is to protect capacity for work that benefits from it and to avoid spending more time prompting than the task is worth.

A practical threshold plan

Measure value by completed work

Count the full effort: preparing material, prompting, waiting, checking, correcting, documenting, and obtaining review. A fast draft that requires extensive repair may not save time. Teams can track a small sample of tasks using the worksheet below.

Suggested team habits

• Set a monthly or project-level budget where the tool provides usage reporting.
• Agree which tasks justify premium models and which should use standard tools.
• Share tested prompts and approved workflows so staff do not repeatedly solve the same setup problem.
• Review failed or costly uses without blame and improve the process.
• Do not evade vendor or university limits by creating extra accounts or moving work to unapproved tools.

Keep sessions focused and manageable

Begin a new chat when you change projects, audiences, data sensitivity, or goals. For a continuing project, carry forward a short, checked summary rather than the entire transcript. This reduces clutter and gives you a chance to remove mistakes and information that no longer matters.

When to summarize and restart

• The chat has moved through several drafts or more than one work product.
• The model repeats an error or seems to follow an old instruction.
• You must scroll extensively to find the current goal or approved facts.
• The task changes from brainstorming to final production or from one audience to another.
• You are about to use a stronger model and want to provide a clean brief.
• The session includes information that should not be carried into the next phase.

Prompt for a handoff summary

Check the summary before reuse

• ☐ Correct errors and remove outdated instructions.
• ☐ Remove sensitive, restricted, personal, or unnecessary information.
• ☐ Confirm that links, citations, dates, and numeric values match the sources.
• ☐ State who will review and approve the next output.
• ☐ Paste the checked summary into a new chat and attach only the source material needed for the next step.

Separate work by purpose

Review every AI assisted product

The staff member using the output remains responsible for it. Review should match the risk of the work. Important material needs source checking and, when appropriate, subject matter, policy, accessibility, communications, legal, privacy, safety, or research review.

Final review checklist

• ☐ Accuracy: I checked important claims, calculations, quotations, citations, names, dates, and links against reliable sources.
• ☐ Completeness: I confirmed that the output addresses the actual assignment and preserves required qualifications.
• ☐ Data protection: I did not expose information that the tool or use was not approved to handle.
• ☐ Human judgment: A qualified person made the decision and reviewed the final content.
• ☐ Fairness: I considered whether the output could disadvantage a group or reflect unsupported assumptions.
• ☐ Academic and research integrity: The use follows course, sponsor, publisher, authorship, and research requirements.
• ☐ Copyright and attribution: I checked ownership, permissions, quotations, and required citations or disclosure.
• ☐ Accessibility: The final product is usable by people with disabilities and does not replace approved accommodations.
• ☐ Transparency: I disclosed material AI use when required by UGA, the work context, or the intended audience.
• ☐ Records: I kept the prompt, sources, revisions, and approvals when the work requires documentation or reproducibility.

Common failure modes

Disclosure example

Adapt disclosure to the work and its governing rules. A simple statement may read: “AI was used to develop an initial outline and improve clarity. The author verified the facts, revised the content, and accepts responsibility for the final version.” This example is not a substitute for a required citation or disclosure format.

Quick reference for staff

Decision card

Supervisor discussion guide

• What outcome are we trying to improve?
• What information will the tool receive, and how is that information classified?
• Which UGA-approved tool and account should we use?
• What work must remain with a person?
• How will we verify quality and measure time saved?
• What disclosure, documentation, or approval is required?
• When will we stop, reassess, or retire the workflow?

Sources and continuing guidance

UGA AI Policy and Guidance hub - Current links to UGA AI policy, acceptable use guidance, research guidance, and data protection standards.

UGA Guidance on Acceptable Use of AI - University guidance on data, privacy, transparency, acceptable uses, and accountability. Revised August 2026.

UGA Considerations for Using AI in Research - Research responsibilities, data and intellectual property, verification, disclosure, and discretion.

UGA Teach with AI - Teaching guidance, supported tools, course expectations, assessment, accessibility, and faculty responsibility.

OpenAI Model Guidance - Current product documentation on model capabilities, prompting, and reasoning effort. Product names and settings change; consult the tool you are authorized to use.