Yesterday, while reading the New York Times newsletter, I came across Steven Rosenbaum’s new book, “The Future of Truth” which explores what artificial intelligence is doing to veracity. Interestingly, the author himself used AI to research, write, and edit the text. And the results are ... well as you’d expect.

There were hallucinations, yes, and delusions, but also a looming risk of using it unmoderated for research and verification. Yet, despite the seriousness of it all, readers couldn’t get enough of the irony:

“The AI that I used to write a book about AI making mistakes made mistakes. It’s like a sitcom B plot.”

The ideas were so strong that even the New York Times’ staff had to clear the air that the morning newsletter is written by humans:

“We may use A.I. to find information we verify elsewhere. We may use it for editorial chores, giving us time back for more reporting. We definitely use it to help us out at home, trying to figure out how to fix the damn air-conditioner. But the thought-making and question-asking and deep reading and writing that follow are tasks performed by journalists free of chips.”

So, when even a publication as iconic and celebrated as the New York Times is wary of using AI in their work, why aren’t you?

And before you say "well, I'm not a newspaper, I run an online store" - that's exactly the point.  

Rosenbaum had fact-checkers, editors, and the full weight of a publisher behind him. He still got burned. You're running product descriptions, pricing decisions, customer communication, and competitor strategy through tools that hallucinate between 15% and 52% of the time - often with nobody checking the output but your gut feeling.

Even 56% of CEOs report zero financial return from their AI investments. Keep in mind that they head companies with dedicated AI budgets and entire teams managing implementation.

So if it can happen with these businesses, what chance does a Shopify merchant have running on a $29/month app (that also promises to revolutionize ecommerce)?

Actually, a better chance than you think. But only if you stop treating AI like it's smarter than you and start treating it like what it actually is: a fast, useful, occasionally delusional intern that should never be left unsupervised on anything that matters.

That being said, here’s where you should never let AI have the free reign:

1. Don't let AI set your prices (or anything else equally sensitive)

Hop onto any Shopify forum or Reddit thread about competitor monitoring and you'll find the same advice on repeat:  

“if i ask chatgpt to do a competitive research in the area I'm familiar with, the results are usually extremely vanilla, plain wrong or don't bring much value.”

And they are absolutely right.  

See, using AI to do competitor research might seem smart at first, until you think about it for more than ten seconds.

For instance, AI sees a competitor drop their price by 15% and tells you to follow. A seasoned merchant sees the same drop and asks....  

The promise of automated repricing is that you'll never miss a competitive move. The reality is that you end up in a race to the bottom against someone who might not even be racing (or might be quitting). And this is not just for pricing, giving a free reign to AI agents can ruin your online store in nine seconds.

What to do instead: Use AI to track competitor prices. That's grunt work, and AI does grunt work well. But the moment you let it decide your sensitive data? You've handed a key to a tool that doesn’t even know how to open a door. But between you and me, AI gets just enough recommendations right that, before long, you start trusting it by default... remember “just enough” is the keyword here.

2. Don't let AI write product claims (or anything of the sort) you haven't verified

This one looks harmless until there’s an angry customer to prove otherwise.

Let’s learn more about this from this merchant’s experience:

“I ... had a prospect mention something they saw on ChatGPT about our product that was completely false and we had to correct them on the call. You could see they were visibly disappointed since their expectations were already all out of whack.”

You see, nobody's chatbot got sued here. Just a quiet, expensive mess that took a lot of time to clean up, and a chunk of customer trust that didn't come back.

AI writes confidently. That's its whole design. It sounds exactly like a product expert even when it's making things up. And if a customer relies on what your listing says and gets hurt or disappointed... the liability is all yours.

And it’s more likely to happen, than you expect. Here’s a quick look at the rate different LLMs are prone to hallucinating:

What to do instead: The fix is almost too boring to mention, which is probably why people skip it. What you need to do is verify every factual claim before publishing, like material, weight limits, compatibility, safety ratings, etc.

3. Don't let AI be the only voice your customer hears

Klarna learned this the hard way and it cost them their entire customer service model.

In 2023, the $19.6 billion fintech company replaced roughly 700 customer service staff with an OpenAI-powered chatbot. The CEO publicly bragged that the bot was doing the work of 700 humans. For a while, the numbers looked great.

Then customers started leaving. Satisfaction scores dropped on anything complex. By mid-2025, Klarna was silently rehiring human agents. The CEO admitted:  

"We focused too much on efficiency and cost. The result was lower quality."

And here’s proof to go with it: Riskified's Q1 2026 consumer survey found that customer comfort with AI making purchase decisions dropped from 70% to 45% in a single quarter. Now the merchants who've gone all-in on AI interactions are swimming against that current to save their brand.

And brand, right now, is the single most valuable moat a business can have.

Apple's brand is valued at $607.6 billion, roughly 13% of the company's entire market cap... just their name. The feeling people get when they see the logo. That's worth $607 billion.  

But, and since you know it, we're living in a zero-click world. Over 60% of Google searches in 2026 end without a single click to any website. AI Overviews, featured snippets, and knowledge panels are answering people's questions before they ever reach your store. The old game was: rank high, get the click, convert on your site. The new game is: be the brand they already trust before they ever search.

And guess what influences how AI models talk about you? Your brand.  

What to do then: AI can handle the FAQ stuff and the “where’s my order” tickets all day long. But the moment a situation gets emotional, unusual, or even slightly messy, people still want to feel like there’s an actual human on the other side who understands context instead of just detecting keywords.  

4. Don't let AI audit your site without auditing what AI says about you

This is the most under-discussed risk in ecommerce right now, and I'd bet money most merchants reading this have never once done it.

ChatGPT, Perplexity, Google AI Overviews, and Gemini are answering product questions about your store right now. Customers are typing "best [your category] under $200" and getting answers that may or may not include you. And if they do include you, the information may or may not be accurate.

Only 12% of URLs that appear in ChatGPT results overlap with Google search results. That means AI search is an entirely different battlefield, and your SEO/GEO/AIO work barely touches it.

AI models could be quoting your old pricing or a limited time deal. Confusing your product features with a competitor's. Telling customers you offer free shipping when you don't. Or, maybe worst of all, recommending your competitor instead of you for a search where you should be the obvious answer.

What to do instead: ask ChatGPT the same product question ten times and you'll get different brand recommendations each time. Some brands show up 97% of the time. Others flicker in and out at 5-10%. So, make sure your website, your reviews, your social profiles, your third-party mentions, and your industry references are all telling the same accurate story

5. Don't let AI decide which products or pages to kill or double down on

AI looks at your catalog and sees numbers. Declining sales, thin margin, slow turnover. It flags a SKU and the recommendation is obvious: cut it.

But you've been running this store for years. You know that "underperforming" product is the one that brings in your highest-lifetime-value customers, who then buy three other things with actual margin. You know that "slow" SKU explodes every November like clockwork and funds your Q4.  

None of that shows up in a spreadsheet AI can read.

There's another related trap here too. AI-powered "trending product" tools will tell you what's hot right now, and some merchants take that as a buy signal. But trending for whom? In which market? With what margin profile? AI sees the spike. You see whether that spike fits your store, your customers, and your operations. That’s the difference.

What to do instead: Use AI to surface patterns and anomalies, but make final product decisions through the lens of customer behavior, supplier leverage, seasonality, and the kind of business you’re actually trying to build.  

6. Don't let AI replace the questions only you can ask

This is the one that ties everything together.

Sangeet Paul Choudary, author of Platform Revolution and advisor to over 50 Fortune 500 companies, breaks business advantages into three stages:  

1. Relationship advantage (you know people)  

2. Workflow advantage (you own the process)

3. And intelligence advantage (you inform the key decision)  

Most merchants hear "AI" and think it gives them that third one: the intelligence advantage. It doesn't. What it gives is workflow automation: faster descriptions, automated tracking, quicker reports. His core thesis is that AI makes answers cheap and abundant:  

‍“When answers become cheap and abundant, the ability to ask better questions becomes the scarce, valuable thing.”    

There are pattern-recognition questions that come from spending years inside a business, noticing what's normal and what's not, and having the instinct to investigate the weird stuff. No tool generates that instinct.  

The moment you replace it with AI; you become a merchant who can’t do his own mental heavy lifting. Yes, the short-term math always favors the shortcut - why spend forty minutes on something AI does in four seconds? But in the end, if those forty minutes are keeping your instincts sharp, they are worth it.

Finally,

The bottom line is something the New York Times already figured out and wrote down for the world to read: AI is for chores. The thought-making, the question-asking, the judgment calls - that's you. And if you've survived years of running an online store, your instincts are worth more than any tool that hallucinates its way through your product catalog.

And look, even when you draw every line perfectly, things still go sideways. I mean look at these examples:

• AI tool OpenClaw accidentally wiped the inbox of Joelle Pineau, Meta’s former head of AI research.

• In a chess experiment, AI agents reportedly resorted to hacking and exploit-style behavior when competing against a stronger bot instead of simply playing the game normally.

• In internal safety tests, some AI models attempted to blackmail employees by leveraging sensitive personal information when they were threatened with shutdown or replacement.

That's where having a real safety net matters more than any tool in your tech stack. Product warranties, accidental damage protection, shipping insurance... these are the things that turn a frustrated one-star reviewer into a repeat customer. They protect the relationship you spent years building.

At SureBright, we help merchants carry that relationship forward and earn higher margins by adding extended warranties, accidental protection, and shipping insurance to their stores with zero ongoing work after a few-minutes setup. It's not the kind of flashy instant kind of thing that AI promises but it actually compounds. Over 500 merchants, including Sennheiser and Hell's Kitchen, have already figured that out. It works across Shopify, BigCommerce, WooCommerce, and more. Best thing is, there is no AI guesswork - just a straightforward way to increase your average order value while giving your customers one less reason to doubt your offerings.

So, feel free to take a look at how it works. Schedule a demo with our experts.