AI hallucinations

Happy Friday!

I hope you’ve had a great week.

Here in New Zealand, the sun is shining—just in time for the long weekend we’re heading into. This weekend has come at a key time. We’re coming up to the pointy end of the year and as always, we’ll be ramping things up before we wind down (I’m sure you can relate). A weekend to recharge is just what the doctor ordered—that, and some helpful tools to tackle the end-of-year rush.

By now, you’ll be well aware that I’m always looking for ways to utilise new tools to optimise workflows, maximise results and bring others on the journey to work smarter, not harder. But I’m the first to admit new tech often has pitfalls. Today, I want to touch on one in particular and give some advice on how to avoid it as we head into a very busy time of year.

Here’s what I’ve got for you today:

Concept
😵‍💫 AI hallucinations—what they are and how to avoid them

Resources
🤓 Material I’ve enjoyed lately

Studio News
📰 Leif gets site of the day

😵‍💫 CONCEPT

AI hallucinations—what they are and how to avoid them

AI tools are becoming indispensable for most ecommerce professionals, but they’re not without quirks. Sometimes, they produce information that sounds credible but is entirely inaccurate—a phenomenon known as an “AI hallucinations.” It’s a bit like asking for ideas from a brilliant but occasionally overconfident assistant who sometimes makes things up on the spot.

I came across a statistic that said AI chatbots make up the facts between 3% and 27% of the time. Do I think that’s a reason to stop using AI? Absolutely not. Only by understanding its flaws can you use the tools effectively.

A few weeks back, I gave some tips for writing effective AI prompts. Today, I’m going to expand on that to ensure that AI can be a more dependable part of your workflow. Here are 5 practical ways to stay in control:

1. Match the tool to the task 
   AI models are built for specific purposes. If you use them outside their intended scope, the results can easily go off course.
   
   If you use ChatGPT to generate copy, you’ll get something decent to work with. But asking it to analyse raw customer data might lead to errors. Instead, use analytics-specific tools like Looker Studio for that task to avoid “creative license”.
   
   

2. Be precise 
   The clearer your request, the more likely you are to get an accurate answer. Keep prompts simple and avoid ambiguous phrasing.
   
   Instead of asking an AI tool to "generate product descriptions for a new collection," specify, "Write a 50-word description for each product in our new minimalist jewellery line, focusing on materials and sustainability." The more detail you provide, the better the result will be.
   
   

3. Break things down 
   Big questions can overwhelm AI systems. By asking smaller, focused questions, you’ll receive better insights and spot any potential errors more easily.
   
   If you need ideas for a Christmas marketing campaign, don’t ask, “How can we increase sales in December?” Instead, start with smaller prompts like:

   • “What are popular gifting trends this year?”

   • “How can we promote high-value bundles?”

   • “What messaging works best with last-minute shoppers?”
     
     

4. Set boundaries 
   Use examples, provide specific guidelines, or narrow the scope of your query to steer the AI towards the right kind of response.
   
   If you’re researching trends for a social media strategy. Ask the AI to limit its suggestions to platforms your brand uses, such as Instagram and TikTok, and to ignore any data older than 2023. This helps you filter out noise and get relevant results.
   
   

5. Always verify
   While AI can speed things up, it is still essential to fact-check outputs.
   
   If AI gives you a statistic for a blog or shareholder report, track down the original source before including it (you can ask the AI to direct you to the original quote).
   

AI can be a powerful ally, but success lies in how you direct and validate its responses. If in doubt, use your common sense. If something smells fishy, it probably is.

🤓 RESOURCES

Material I’ve enjoyed lately

📗 Read

2024 Holiday Shopping Trend Predictions by DigiCommerce

This article has some interesting predictions for the upcoming silly season. Seeing which demographics prefer to shop in-store versus online was particularly interesting, as well as the “treat yourself” mentality of millennials. Excuse me while I go find myself a Christmas gift…

🎧 Listen

Hard Fork

New York Times journalists, Casey Newton and Kevin Roose, discuss the latest developments in the tech space. It’s funny, relatable and informative and I like how the presenters take a progressive approach to technology while still bringing it back to human values. It’s well worth a listen.

📚 Bonus Read

Coral created by AI, developed by design

Since we’re on the subject of AI, Leif’s new Holiday collection is a feast for the eyes. Using text prompts, in-house renders and 3D model inputs they’ve created a series of reimagined coral characters. They’re a riot of colour and delightfully surreal—it’s a different kind of “AI hallucination” but one I very much approve of.

And that brings me to this week’s studio news…

📰 STUDIO NEWS

Leif gets site of the day!

On Monday this week, Leif’s new Shopify site was named Site of the Day by the CSS Design Awards.

We are so proud of the work we created in partnership with Jonnie, Candice, Amanda, and the rest of the Leif team. It’s a great example of how conversion tactics can still be design-forward and we’re excited to see how this new asset serves their business.

That’s all from me this week.

Thanks for reading!

Have a great weekend.

Alex Murton
Managing Director & Co-Founder
Studio Almond