Catch the wrong fact before your readers do

Hand over anything you’re about to publish (an essay, a post, a memo, a press release) and the assistant checks it the way a professional fact-checking desk would. Every specific claim gets pulled out and verified against live web sources, not from memory.

What it does

• Reads your draft and extracts every checkable claim: numbers, quotes, dates, “firsts” and “largests,” anything a reader could look up.
• Verifies each one against live sources, preferring primary records (filings, official releases, court documents) over hearsay.
• Marks each claim on a clear scale, from verified to flatly wrong, with the source it checked against.
• Applies fixes to the ones that are wrong and hedges the ones that can’t be confirmed.
• Returns a scored report: how many claims checked, how many held up, and an overall confidence level.

The reveal

You paste in a piece you were about to send. Ten minutes later you get it back with the two numbers that were slightly off already corrected, the quote that was misattributed traced to who actually said it, and a green light on the other forty claims. The embarrassing error that would have surfaced after publication never leaves your draft.

What it doesn’t do

• It doesn’t check opinions, arguments, or predictions. Those are yours, and they aren’t facts to verify.
• It never treats its own memory as a source. Every verdict points to something citable and current.
• It won’t rewrite your voice. It corrects facts and flags the rest.

Configuration

• What to check: paste the text, or point the assistant at a document you’ve dropped in.
• Sources to trust: any research feeds or document libraries your team has already connected, plus the public records the assistant reaches on its own.
• How strict: the bar a piece must clear before it earns a pass.

Triggers

On-demand. You bring a draft, ask for a check, and the verification runs in the background while you do something else.