The Short Answer
Your confidence should scale with the stakes. For low-stakes tasks like generating discussion prompts or brainstorming activity ideas, AI is reliable enough to use with a light review. For anything factual, technical, or that students will rely on for assessments — always verify independently before using it.
A Simple Risk Tiers Framework
Tier 1 — Use with light review (low stakes)
These tasks are low-risk because errors don’t cause significant harm and are easy to catch:
- Generating discussion questions
- Drafting email templates
- Brainstorming activity ideas
- Summarizing a concept you already know well
- Rephrasing your own content
Confidence level: High — but still read it before you use it.
Tier 2 — Verify before using (medium stakes)
These tasks involve factual claims that students or participants may rely on:
- Explanations of concepts you’re less familiar with
- Statistics or research findings
- Historical information or timelines
- Best practice recommendations in your field
Confidence level: Medium — cross-check with a reliable source before sharing.
Tier 3 — Always verify independently (high stakes)
Never rely on AI alone for these:
- Medical, legal, or financial guidance
- Specific citations or direct quotes (AI fabricates these regularly)
- Anything that informs student grades, placement, or formal assessments
- Current events or recent research (post knowledge cutoff)
Confidence level: Low — treat as a starting point only, verify everything.
The Citation Problem Deserves Special Attention
AI regularly invents citations — real-sounding author names, journal titles, and page numbers that don’t exist. If you ask AI to cite sources, you must look up every citation manually. This is one of the most common mistakes educators make when first using AI in their teaching.
A Quick Gut-Check Before Using AI Output
Ask yourself: “If this is wrong, what happens?” If the answer is “nothing serious,” use it. If the answer is “a student gets incorrect information, or I look uninformed,” verify it first.
The Practical Habit
Read every AI output before you use it — not to catch every error (you won’t), but to catch obvious ones and flag anything that feels off for a quick external check. Over time, you’ll develop a reliable instinct for which AI outputs are safe to use as-is and which ones need a second look.