Posts

Design an AI Task With an Explicit Unknown Output

An extraction task becomes fragile when every field must contain a confident answer. Define an explicit Unknown state for information that the source does not establish. Keep Unknown separate from No and Not applicable. Set up a practical example For a fictional event note, “Parking is not mentioned” does not mean parking is unavailable. Ask the assistant to return the source wording, the extracted value and a status label. Specify how it should handle blank fields and conflicting statements. Test the prompt with one clear fact, one omitted fact and one contradiction. Write the expected status for each before running the task. If the source says two different start times, the output should flag the conflict rather than silently choose the later sentence. Verify before proceeding Review whether the assistant preserves those distinctions in its summary. A table full of Unknown values may accurately reveal that the source is insufficient. The next action is obtaining information from the ...

Distinguish a Tool Recommendation From an Actual Test

A recommendation can be based on documentation, a limited trial or a recurring personal workflow. These are different kinds of evidence. State which one you have before writing a conclusion about an AI tool. Set up a practical example Create three headings in your notes: Documented, Observed and Not checked. Under Documented, record inspected vendor statements and their scope. Under Observed, describe exactly what you did and what happened. Leave untested features in the third section. For a fictional note-taking app, reading that export exists does not prove the export fits your destination. A useful test would open the actual exported file and inspect structure, missing content and formatting. Do not report that test as completed if you only saw the export button. Verify before proceeding Write the recommendation around the reader's task and the evidence you have. Use “The documentation states” for a documented claim and “In this trial” for an observed result. Avoid implying a br...

Build a Small Glossary Before Asking AI to Explain Technical Notes

Simplifying a technical note can introduce error when an assistant replaces a precise term with a familiar but different one. Prepare a small glossary and an audience statement before requesting the rewrite. Keep definitions tied to the source material. Set up a practical example For a fictional project note, distinguish a draft, a scheduled release and a public release. Explain which terms must remain exact and which can be followed by a plain-language explanation. Avoid asking for a universal definition when the term has a specific meaning inside your project. Ask the assistant to rewrite one paragraph first. Require it to preserve the relationships between components and to mark any term it cannot explain from the supplied notes. An unanswered question is more useful than an elegant analogy that changes the mechanism. Verify before proceeding Compare the rewrite with the original. Check whether a possibility became a guarantee, whether a prerequisite disappeared and whether the anal...

Keep an AI Draft From Inventing Personal Experience

A draft may sound more persuasive when it claims that the writer tried a product or learned a lesson firsthand. Do not publish that claim unless it reflects a real, documented experience you are authorised to describe. A fictional example should remain visibly fictional. Set up a practical example Search the draft for phrases such as “I tested,” “we found,” and “in my experience.” For each one, identify the underlying observation, date and scope. If there was no test, rewrite the passage as a proposed method or an illustrative scenario. For a fictional productivity article, replace “This saved me two hours every week” with a worked example showing the steps the method removes. Do not turn the example into a measured time saving. The reader can still evaluate whether the workflow suits their situation. Verify before proceeding Check photos, captions and charts for the same problem. An illustrative screenshot or generated scene should not imply access to a real customer, laboratory or pr...

Use a Change Table to Review an AI Rewrite

A clean rewritten document can hide changes you did not request. For a substantial edit, ask for a short change table beside the proposed copy. Review meaning, facts and commitments before spending time on sentence rhythm. Set up a practical example Create columns for Original passage, Proposed passage and Reason. For a fictional service notice, protect dates, URLs and the distinction between a confirmed event and a possible event. A revision from “may be delayed” to “will be delayed” is not merely a style improvement. Focus the table on material changes rather than every comma. Ask the assistant to flag removed qualifications, added examples and altered instructions. Its table may omit a change, so compare the final copy with the original as well. Verify before proceeding When you reject a change, state the rule it violated. “Preserve uncertainty where the source is uncertain” is more reusable than asking for a vaguely softer tone. Apply the rule to the whole document and check other ...

Make an AI Handoff Note Useful to the Next Person

A handoff note should let the next person continue without guessing what happened. Ask for current state, verified actions, evidence locations and remaining work. Keep plans and completed actions in separate sentences. Set up a practical example For a fictional website update, “Contact page drafted” is different from “Contact page published and opened successfully.” Include the actual page or file reference when available, along with any unresolved limitation. Do not describe a preview as the live result. List what must be preserved before the next edit: existing posts, scheduled dates, account identity and backups. Add the smallest useful reproduction steps for a failure. An unexplained “does not work” note forces the next person to repeat the investigation. Verify before proceeding Read the handoff against the real service state before sharing it. Remove stale instructions that refer to an abandoned version. Keep credentials and unnecessary personal information out of the note; use a...

Review an AI File-Renaming Plan Before Anything Moves

A file-renaming task should begin with a proposed mapping, not with moving files immediately. Ask the assistant to show the current name, proposed name and reason for each change. Keep ambiguous files unchanged until you can identify them confidently. Set up a practical example Use a fictional folder containing report.pdf, report-final.pdf and report-final-2.pdf. The names alone do not prove which document is latest. Have the assistant list the evidence needed to distinguish them instead of deleting or overwriting the apparent duplicates. Review the mapping for collisions. Two source files must not silently become the same destination name. Check extensions, date formats and whether another application depends on the existing path. Preserve a backup and use the recovery mechanisms appropriate to your actual environment. Verify before proceeding Approve a small reversible example first. After execution, compare the actual folder with the approved mapping and open representative files. A...