Fast is easy now. An AI tool can give you a proposal, a blog post, a contract summary or a client email in the time it takes to pour coffee. That’s a real advantage for a one-person business, and it’s also the trap. When drafting costs almost nothing, the bottleneck moves to the one step the tool can’t do for you: deciding whether the thing is right.
You have no editor, no legal department and no colleague down the hall to catch the slip. You are the quality department. Here is how to run it without giving back all the time you saved.
When Speed Is the Wrong Metric
Speed is a fine goal for work that is internal, low-stakes and easy to undo. It’s the wrong goal when any of these is true:
- It’s public. Once it’s on your website or in a client’s inbox, you can’t recall it.
- It contains facts or numbers a reader might act on. Prices, dates, legal or tax claims, statistics, quotes.
- It makes a promise. Deadlines, scope, guarantees, refunds.
- It’s a first impression. A pitch to a dream client is not the place to sound like everyone else.
- It’s expensive to fix later. A wrong figure in a quote is a margin problem. A wrong clause in a contract is worse.
Cal Newport’s argument in Deep Work is that the rare skill is producing at a high level of both quality and speed, and that both depend on focused effort rather than hurried effort. Our sister site The Growth Reads breaks the book down in its Deep Work review. AI gets you the speed; the quality side is still on you.
Two Cautionary Cases
These are public and well documented, and they explain why “the tool wrote it” is never a defense.
The lawyers and the invented cases. In 2023, two New York lawyers and their firm filed a court brief that cited cases that did not exist, which they had gotten from a chatbot while doing legal research. On June 22, 2023, Judge P. Kevin Castel fined them $5,000 and ordered them to notify the judges whose names were falsely attached to the fake opinions. The sanction was driven largely by how they behaved after the problem surfaced, but the root cause was simple: nobody opened the cases and read them.
The airline and its chatbot. In Moffatt v. Air Canada (2024), a British Columbia tribunal found the airline liable after its website chatbot gave a customer wrong information about bereavement fares. The airline had argued the chatbot was effectively a separate entity responsible for its own statements. The tribunal rejected that and treated the chatbot as part of the company’s website, finding the airline hadn’t taken reasonable care that the information was accurate.
The lesson for a solo business is plain. Whatever your tools say or write in your name, you said it.
A Four-Pass Checking Routine
Don’t re-read the draft hoping errors will leap out. Check in passes, each looking for one kind of problem.
Pass 1: Facts
- Highlight every claim a reader could verify: names, dates, places, laws, statistics, quotes, product features, anything that starts “studies show.”
- Check each against a primary source: the official site, the statute, the original paper, the company’s own documentation. Open any link the tool gives you. Links and citations can be invented or point to pages that don’t say what’s claimed.
- If you can’t verify a claim in a couple of minutes, cut it or soften it. A smaller true statement beats a bigger shaky one.
You can also set the tool up to help. Anthropic’s guidance on reducing made-up answers suggests giving the model explicit permission to say it doesn’t know, asking it to ground claims in quotes from documents you supply, and having it retract claims it can’t support. The same page is clear that these techniques reduce the problem rather than eliminate it, and that you should always validate critical information. Treat them as a first filter, not a replacement for your check.
Pass 2: Numbers
- Recalculate every total, percentage and date range yourself, in a spreadsheet or calculator, from the original inputs.
- Check units and currency, and whether the figure is per month or per year.
- Compare any number in the draft to the number in your source. Transposed digits and rounded-up figures are common.
- Reconcile: does the sentence in the intro agree with the table at the bottom?
Pass 3: Voice
- Read it aloud. If you’d never say it, change it.
- Hunt generic filler: stacked adjectives, “unlock,” “elevate,” openers that could start any article ever written.
- Check it against your style guide (see Teaching Your AI to Hunt) and fix the pattern in your saved prompt so you aren’t editing the same tic forever.
Pass 4: Promises and risk
- Does the draft commit you to anything: a date, a price, a result? Would you honor it?
- Does it include anything confidential that shouldn’t leave the building?
- Does it claim expertise, results or credentials you don’t have? Remove them. See the ethics of success for the honesty side.
- Would you be comfortable if the recipient forwarded it to their lawyer? If not, rewrite it. The old advice in be careful what you write still holds.
Match the Checking to the Stakes
You can’t give everything the full treatment. Triage:
- Low stakes (internal notes, brainstorms, rough outlines): skim and use.
- Medium (client emails, social posts, routine updates): a voice pass and a quick fact and number check.
- High (published articles, proposals, contracts, anything with prices, legal, tax or health claims): all four passes, then leave it overnight and read it once more. For legal and tax matters, a qualified professional reviews the final version.
It also helps to put a risk label on the task itself, before you draft. A task where a mistake is cheap, reversible and visible is a good place to let the tool run. Pieter Levels, who builds his products solo, says he now does nearly all his coding with an AI agent working straight on his live server, and that it broke his sites only twice in a year, each time for around ten seconds. He also keeps backups, and says a bigger company should test on a staging server first. His setup works because mistakes are quick to spot and quick to undo. A client proposal has no undo button.
Don’t Let the Checker Be the Same Brain
Asking the same AI tool “is this accurate?” is not independent verification. It’s useful for flagging suspicious spots, such as “list every claim here that needs a source,” and you should use it that way. But the verdict comes from a source document, a calculator or your own eyes.
If you’re sending client data into any tool as part of this, remember the basics from stopping data leaks: know what the tool does with what you paste.
The Takeaway
Use the speed. Spend a fixed slice of what it saves on checking, scaled to the stakes: facts first, then numbers, then voice, then promises. Make “I verified it” the standard for anything with your name on it, not “it sounded right.” Want a printable version of the checklist alongside the rest of the toolkit? Grab the free “One-Person Business AI Stack” checklist.
General information, not professional advice. Legal and tax questions should go to a qualified professional.
Sources
- LawNext, “Court imposes sanctions on lawyers who filed bogus cases after relying on ChatGPT for legal research” (Mata v. Avianca; June 2023): lawnext.com
- McCarthy Tétrault, “Moffatt v. Air Canada: misrepresentation by AI chatbot” (2024 BCCRT 149): mccarthy.ca
- Anthropic, “Reduce hallucinations”: platform.claude.com
- Pieter Levels, “I’ve been coding almost solely on my VPS with Claude Code for almost a year now” (June 2026): levels.io