You're Betting Real Money on One Confident Answer

There's a contract in your inbox. The AI receptionist, the new job-management software, the second location — whatever the real-money call is this month. Your research so far: the vendor's own blog post, a case study the vendor picked, and a question to an AI chat that came back with one confident answer and a statistic you didn't check. Your gut agrees. So you sign.

I'm not mocking that. It's how most real-money decisions get made in a small business, because the honest alternative has always been worse: weeks you don't have, a consultant you can't justify, or twelve browser tabs that all trace back to the same three press releases.

One confident answer isn't research. It's usually the seller's angle, read back to you in a friendlier voice.

That's the trap in asking an AI to check for you. Ask "should I buy an AI receptionist?" and you get one angle — smooth, certain, sourced from whoever wrote the most about the topic. Guess who writes the most about a product. The people selling it.

What an editor would do

Newsrooms solved this long before AI. An editor doesn't run a story on a single source, no matter how confident the reporter sounds. The claim gets confirmed independently, the people who'd dispute it get a call before it prints, and if the confirmation falls through, the story is cut — not softened, cut. Not because reporters lie, but because a single source is always certain and regularly wrong.

Your five-figure decision deserves that desk. So I turned the desk into a skill you install once.

The same question, asked five ways

It's called Storm Research, built on Stanford's STORM method. You ask one business question. Instead of answering it, the skill first works out which experts would actually disagree about your topic — the operator who's run it, the skeptic, the one following the money — and picks a panel of four to six. Each researches your question separately, asking the questions only its perspective would ask. Then the skill maps where they contradict each other, writes it all into one briefing, and does the part that matters most: it re-checks every number against the original source before you read a word.

The scary stat that traces back to a vendor's marketing page gets flagged, not repeated.

What lands is a briefing, not a verdict: a 60-second summary, five findings ranked by evidence quality, a plain list of what's safe to say out loud and what to never repeat — and a banner at the top with an honest tally of how many claims got checked, corrected, or demoted on the way to you.

Don't take my description on faith. The question every service business is hearing pitches about right now — "should I let an AI answer my phone?" — is a real briefing the skill produced, unedited, verification banner and all. Read it before you install anything.

What it touches, plainly

Your question is researched on the public web by your own AI assistant. No business data leaves your machine unless you type it into the question. The skill stores nothing. Every source is cited with a URL and independently re-checked — and what comes back is still a draft: input for your decision, honestly labeled, including what it couldn't verify. The panel is built by the skill and the briefing says so — agreement across it is a strong lead, not proof. The panel reads; you decide.

One honest note: this is the heaviest skill in the library. It wants an AI assistant that can run research agents — Claude Code or the Claude desktop app is the straight path — and it's for decisions worth twenty minutes, not for looking up an address.

Why bother

Because skipping the check is the expensive part. In the book I wrote on AI readiness there's a CEO who spent $180,000 on AI licenses and training and couldn't point to a single dollar of return — the tool got bought before the question got checked (AI READY, Danny de Vries). Next to that, twenty minutes of a panel arguing about your question is cheap. It's also free: I've shipped north of 20,000 software products, and this one asks nothing but the question you were about to bet on anyway.