Check before you send

The one idea

AI-assisted work is a draft until you have checked it. The trap is that it looks finished — clean, confident, well formatted — long before it is actually correct. Build a 30-second habit before anything leaves your desk.

The three-point check

1. Numbers: totals, percentages, dates, money. Arithmetic is not pattern-matching, so the model can mangle figures even when summarising your own document. Re-check every number against the source.
2. Unsupported claims: anything stated as fact that the model recalled on its own — names, statistics, quotes, "studies show". If you did not supply it, verify it.
3. Tone and audience: is this right for this reader? A technically correct message can still be too blunt, too casual, or aimed at the wrong person.

Let the AI help check itself

You can even hand the draft back for an audit:

"Audit this draft. List every number, every claim not supported by the text I gave you, and anything that sounds off in tone for a nervous client. Do not rewrite it — just flag."

That turns the model into a second pair of eyes — but the final judgment stays yours.

Try it now

Take an AI-assisted message you are about to send. Run the three-point check by hand, or ask the AI to flag numbers, unsupported claims, and tone. Fix what it surfaces.

Your win today

You get the speed of AI without the embarrassment of sending something wrong. One short verification pass catches almost everything that matters.