This is when not to use AI because the input itself is too sensitive
Sometimes the risk is not a wrong answer. It is handing a service the very document, conversation, or record the task depends on.
Older material from this site's earlier life. Kept online so links keep working.
Sometimes the risk is not a wrong answer. It is handing a service the very document, conversation, or record the task depends on.
When two estimates do not look like they describe the same job, use AI to compare scope, exclusions, payment terms, and the questions to ask before signing.
Before the laptop, phone, or TV checkout page wins, use AI to compare the specs that actually matter for your life and budget.
Package size, sale labels, and different units can hide the cheaper choice. Transcribe first, then show the normalized arithmetic.
Retype only the logistical facts from a long thread and turn them into decisions, owners, and open questions without uploading the conversation.
Read the axes, baseline, units, sample, and source before letting the most dramatic shape in the chart become the conclusion.
Transcribe uncertain handwriting first, confirm the original yield, then scale only the amounts the recipe actually supports.
Transcribe stacked times, arrows, days, and exceptions into separate candidate rules without asking AI for a legal parking verdict.
Turn colors, branches, transfer circles, zones, and legends into a route question without treating a static map as live service.
Use tightly staged walkway photos to notice ordinary visible obstacles, while unseen risks and safety judgment stay with a real inspection.
Serving size, servings per container, and different units can distort a quick comparison. Make the printed arithmetic visible first.
Turn a tiny care label into a symbol-by-symbol transcription, then confirm every interpretation with an authoritative care key.