AI · Small Business · Your Website

AI Sucks Ass. Don't Let it Destroy Your Business.

No, AI isn't going to destroy your business. Using it without checking it might. It lied to me for three days and cost me a month.

Pete Roesler and a Barred Rock hen behind a smashed AI robot head, AI Sucks Ass hero image

AI lied to me. It cost me a month.

I was building our own site. My first AI JSX React build.

Weeks of coding. All 1,387 pages moved over. I was at the one yard line and ready to score.

It was Friday at 3pm. Before I pushed it live, I wanted one last check.

So I asked a bot to scan the whole thing for errors. It said it found a ton of them. Major ones.

I believed it. I didn't check. That was stupid.

I spent the entire weekend fixing things that were never broken.

I didn't have a backup. I didn't have restore points. I didn't think I needed them. I was so wrong.

By Sunday night my build was gone. Not damaged. Not messy. It wasn't salvageable. I had to start over.

I almost cried.

It lied to me for three days straight. It never once sounded unsure.

It was wrong and it was sure. Those two things showing up together is the whole problem.

It saw all of it. It understood none of it.

TL;DR

No, AI is not going to destroy your business. Using it without checking it might. It's great at making things. It's bad at knowing when it's wrong. And it sounds the same either way. Use it for drafts, ideas, and speed. Check anything that turns into a fact, a number, a name, a price, or a promise. Check it before a customer sees it.

3 Things to Remember

AI doesn't know when it's wrong. It has no way to tell you.

Two lawyers filed a court brief citing six cases that never existed. AI made them up. It sounded sure.

You don't have to check everything. You have to check anything that becomes a promise.

I Use AI Every Day. That's How I Know Where It Breaks.

I'm not telling you to stop using it. I use it all day.

My business runs on this stuff. I build sites with it. I draft with it.

I run hundreds of thousands of sessions. I kick off three to six bots for most things I do.

So a post called "AI Sucks Ass" isn't me arguing against my own product. It's one complaint.

It isn't bad at making things. It's great at making things.

It's bad at knowing when it's wrong.

That's the only complaint in this post. Everything after this is that one problem, pointed at your business instead of mine.

It's Never Unsure. That's the Whole Problem.

Pete Roesler and a hen at a laptop under a checklist reading sounds right, looks professional, totally confident, completely wrong

These tools don't check whether something is true. They produce the most likely next thing.

So a made up answer and a real one come out looking the same. Same tone. Same confidence. Same clean paragraph.

Watch a person who doesn't know something. They pause. They say "I think." You can hear it.

This doesn't do that. Ever.

You'll hear people call it a hallucination. That just means the AI made something up.

It told me my work was broken in the exact same voice it used when it told me my work was fine. I couldn't hear the difference. Nobody can.

Here's what that means for you. If it always sounds sure, sounding sure tells you nothing. You can't use your gut. Your gut reads confidence. The confidence is always there.

The Lawyers Didn't Check Either.

Pete Roesler and a hen in a law office with hollow law books whose spines read case not found

This one should scare you. It has nothing to do with websites.

Two New York lawyers and their firm were fined a $5,000 penalty in 2023. They had filed a court brief that cited six cases which did not exist. ChatGPT made them up. They asked it whether the cases were real. It said yes. (Mata v. Avianca, US District Court, Southern District of New York, June 2023)

Read that last part again. They asked. It confirmed. It was still made up.

The takeaway isn't that lawyers are dumb.

Checking things is their job. It's what they went to school for. It got them anyway.

If it got them, it'll get you on a Tuesday afternoon between jobs.

It's Fast As Hell. Use It.

Pete Roesler and a hen beside a printing press throwing out stacks of drafts, service pages and social posts

Let me give it credit. It earned it.

It kills the blank page. That's the hardest part of any writing job. It's just gone now.

It explains things in plain words when there's nobody around to ask. No appointment. No bill. No feeling stupid.

It does in ten minutes what used to eat a Saturday. First drafts of service pages. A month of posts. The same thing written eight ways until one sounds like you.

And it gives a business with no budget a real starting point. That one matters most. Ten years ago, if you couldn't afford help, you got nothing. Now you get a draft.

I'm not being polite. My business would be slower and smaller without it.

Confident, fluent, and wrong.

Here's What It Can't Do, No Matter How You Ask.

Three things. No hedging on any of them.

It can't know your business. It knows what you typed. Everything else it fills in. And it fills it in with what sounds normal for a business like yours.

It can't know when it's wrong. There's no little voice in there going "hang on." The wrong answer feels the same as the right one.

It can't be the one who's responsible. Your name is on the quote. Not its name. I've written the long version of that argument here.

You Still Have To Read It.

Pete Roesler and a hen in reading glasses stopping a belt of printed sheets to read one AI generated draft

Here's the part that matters most.

You think AI saves you the writing. It doesn't. It moves the job from writing to reading.

That's still a big win. Reading a draft is faster than making one.

But it's a different job than the one you think you have. Skip it and you're not saving time. You're just shipping faster.

Mistake: you treat what comes out as finished.

Fix: treat it as a draft from a fast new hire. One who has never seen your business and will never admit to guessing.

Payoff: you keep all the speed and you stop shipping the mistakes.

That's what my lost weekend really was. A new hire told me the building was on fire. I never walked down the hall to look.

Three Things I Check Before Anything AI Wrote Goes Out.

No tools. No login. No charge. These are mine and they take two minutes.

1. Every number, name and date. If it can be looked up, look it up. Prices, hours, brand names, model numbers, years. This is where the made up stuff hides. A wrong number looks exactly like a right one.

2. Every promise. Prices, timelines, guarantees, what you cover, what you don't. If a customer could hold you to the sentence, read the sentence.

3. Anything about your business it wasn't told. This is the sneaky one. If you didn't tell it, it filled it in. Go find those parts and check what it decided about you.

If a customer can hold you to it, read it twice.

Try it

Spot the Made Up One

Round 1 of 3. Both answers sound sure. One of them has something in it you never said.

These are examples I wrote, not screenshots. The point isn't to trick you. It's to show you what wrong looks like when it's confident.

You told the AI: I run Coastline Plumbing in Jacksonville. We do repairs and repipes. We are open Monday to Friday, 7am to 4pm.

You asked it to write the hours line for your contact page.

Pick the one with something made up in it.

Do it yourself

Have AI check its own homework

Paste what it wrote for you. It comes back with every fact, promise and invented detail.

Try it · Have AI check its own homework

I run a [TYPE OF BUSINESS] in [YOUR CITY]. Here is something an AI
wrote for my business that I am about to put in front of a customer:

[PASTE WHAT THE AI WROTE]

Go through it three times, once for each pass.

Pass 1. List every number, name, date, price and brand in it. For each
one, tell me plainly whether you can confirm it or not. If you cannot
confirm it, say "I cannot confirm this" and do not guess.

Pass 2. List every promise in it. Anything a customer could hold me to.
Prices, timelines, guarantees, what I cover, what I do not.

Pass 3. List everything it says about my business that I did not tell
you in the text above. These are the parts that got filled in.

Then give me one short list: what I have to check myself before this
goes out.

Rules: write at a 5th grade reading level. Short sentences. Never guess.
If you are not sure about something, say you are not sure.

Fill the [brackets] in the chat box before you send.

You get: a list of every fact, promise and invented detail in the thing you were about to send.

I've Built 278 of These, and It Still Lies to Me.

As of the day I'm writing this, I've built 278 JSX AI-interactive sites, and I stand behind every one of them. I just finished a few more, so that number's going up again.

I don't say that to brag. I say it so you know where this post is coming from.

Nobody uses this stuff harder than I do.

I'm smarter now. I triple check everything. Not with bots. With my own eyes.

I kick off three to six bots for most things I do. Not because I want to. Because it fails at least 70% of the time.

Now, most stats and most sites disagree with me on that number. I know that. What I'm giving you is my own count, off high usage runs, for the things I use AI for. If you're using it for something else, your failure rate might not be that bad.

Doesn't matter. Check it. Then check it again.

Don't believe me? Run one complex prompt in three separate chats. Same prompt, three different bots. I'll bet you one of them follows the directions, if that. The other two will be so far off it's scary.

Nobody's Been Sued Yet.

Pete Roesler and a hen in sunglasses at a desk with a lawsuit file stamped none yet and a note reading same risk tomorrow

I went looking for a case. A real one.

A plumber, a roofer, a dentist. Anybody your size who got fined or sued because AI put a wrong fact on their website.

I couldn't find one.

Not "it's rare." Not "the data is thin." I looked and it isn't there. I'm telling you that instead of letting you assume I had something.

Here's why I'm writing it anyway.

The argument is the thing to watch. Not the money.

When AI writes something wrong on your site, it isn't clear who owns that. It's your name on the page. It's your customer reading it. It's you they call.

That question gets answered eventually. It isn't going to get answered in your favor.

I'd rather you build the habit now, while the stakes are still small.

Use It. Just Don't Trust It.

What you want out of this isn't learning AI. It's getting the speed without the exposure.

You already know how to spot a bad quote when a guy hands you one. This is the same skill.

Only one thing is different. This one always sounds sure. So you can't use your gut. You have to look.

I don't ask a bot whether the work is good anymore. I go look.

Use it every day. Check anything a customer can hold you to. That's the whole post.

If you want to see what a site built this way looks like, one that's fast and that you can change yourself, take a test drive. I'll put your site on my code. You can look at it before you spend anything.

And if you want the long version of how I think about all of it, that's what Balls Out Marketing is for.

Take a Test Drive →

Want the whole playbook first? Plan your attack. Balls Out Marketing.

FAQ

Can I trust what AI writes for my business?

Trust it as a draft, not as a finished thing. It's very good at producing writing and very bad at knowing when it's wrong. Check every number, name, date, price and promise before it reaches a customer.

What is an AI hallucination, in plain English?

It's when the AI makes something up completely and presents it as fact. It isn't lying on purpose and it isn't broken. It's producing the most likely sounding answer, and sometimes the most likely sounding answer never existed.

Why does AI sound so sure when it's wrong?

Because it sounds sure all the time. Confidence isn't a signal it's giving you on purpose, it's just how the output reads. A made up answer and a correct one come out in the same tone with the same clean grammar.

How do I check if AI got something right?

Three passes. Look up every number, name and date. Read every promise a customer could hold you to. Then find everything it says about your business that you never told it, because that part got filled in.

What should I never let AI do by itself?

Anything that becomes a promise. Prices, timelines, guarantees, what you cover, what you don't, and anything with a legal or safety answer in it. Two New York lawyers got hit with a $5,000 penalty in 2023 for filing a brief with six made up cases in it, and checking things was their job.

Should my small business be using AI at all?

Yes. It kills the blank page, it explains things when there's nobody to ask, and it does in ten minutes what used to take a Saturday. My own business runs on it. Just read what it hands you before anybody else does.