AI Search

AI Product Descriptions for Food and Drink Brands: What Can Go Wrong, and How to Do It Safely

If you sell food or drink on Shopify, you've probably seen the option to let AI write your product descriptions. Shopify has one built in, Shopify Magic, and ChatGPT can produce a description in seconds. With a catalogue of fifty or a hundred products, the time saving looks obvious.

A UK food brand I worked with took that route. The brand used a ChatGPT-based agent to write product descriptions across its catalogue, and two problems came up. The agent mixed details between similar products, and it repeated the same phrases and structure from one page to the next.

Those descriptions went live. When I spotted the errors, I took examples to the owner and offered to take over the site's SEO myself. He agreed. I still used AI for the work. The difference was what I gave it and what I did with its output. I brought in data from Google Search Console, Semrush and Ahrefs, so the AI worked from real searches, rankings and competitor data, and I decided what went live.

Why AI mixes up products

When an agent writes many pages in one run, information from one product can leak into the next. A range with several close variations makes it worse. The model has no way of knowing what is true about each product, so it writes what sounds plausible.

It also settles into a pattern. Ask for fifty descriptions and you get the same adjectives, the same opening line and the same sign-off fifty times. Your customers notice, and so does Google.

Why this matters more for food

On a clothing site, a mixed-up detail means the wrong fabric in a description. In a food catalogue, it can mean the wrong ingredient or a missing allergen.

UK law treats this seriously. When you sell food online, the Food Standards Agency requires you to give allergen information at two stages: before the customer buys and when you deliver. Your product page carries legal information as well as marketing copy.

AI search adds another layer. ChatGPT, Perplexity and Google's AI answers read your product pages when a shopper asks "does this contain gluten?" or "which of these is dairy-free?". An error on your site can end up in an answer to a customer you never spoke to.

The part nobody asked the AI to do

When I looked at those product pages again, the errors were only half the problem. The agent had delivered what the brand asked for: product descriptions. Nobody had asked for pages that answer the questions shoppers type into ChatGPT, or for facts an AI assistant could quote with confidence. The pages had text on them, and AI search had almost nothing to use. That's when I started building AEO into the brand's product pages.

Research on AI search shows what those pages were missing. The GEO study from Princeton (Aggarwal et al., 2024) tested content changes across 10,000 queries. Keyword density did little. Adding statistics, citing sources and quoting credible authorities raised visibility in AI answers by up to 40%. AI search rewards specific, checkable facts, and generic product copy has none.

For a food product, specific means what's in it, where it comes from, how you make it, what it goes with, and the awards it has won.

Will Google penalise AI-written pages?

Not for being AI-written. Google's guidance on generative AI content accepts AI as a tool, and warns against generating many pages without adding value for users. An Ahrefs study of 600,000 pages found that 86.5% of top-ranking pages contained some AI-generated content. Google judges the page. Thin, repetitive, inaccurate pages lose, however you wrote them.

How I work with AI

I bring the data: what people search for in Search Console, where a site ranks in Semrush and Ahrefs, and who ranks around it. The AI helps me read that data faster, spot patterns, and draft the pages, schema and FAQs that follow from it. I choose which searches matter, check every fact, and approve what goes live.

An agent working alone has none of that context. It doesn't know which of your products sell, which questions your customers ask, or which search terms mean something different in your market.

A safer workflow

Steps 1 to 3 take time, and your team can do them. Step 4 takes knowing how AI search chooses its sources and what shoppers ask about your category. That's the step brands skip, and it's the one I work on with food and drink brands.

FAQ