Latest / The DTC Marketing Podcast with Fexingo: Direct-to-Consumer Brands, Shopify, and E-Commerce Growth / How DTC Brands Use AI to Write Product Descriptions
Transcript
- Lucas: Let me start with a number that stopped me: Brooklinen used GPT-4 to rewrite product descriptions for two thousand SKUs in under forty-eight hours. A human copywriter doing that same job? Probably two weeks minimum, maybe three. Luna: Two thousand SKUs in two days. That's not just faster, that's a completely different scale of operation for a DTC team. Lucas: Exactly. And Brooklinen isn't some outlier. I've been tracking this across maybe a dozen DTC brands over the last six months. The ones that have figured out how to use generative AI for product copy are seeing a measurable lift in organic traffic to product pages — we're talking fifteen to twenty-five percent improvements in impressions from search. Luna: But I have to ask — does the copy actually sound good? Because I've read ai generated descriptions that feel like they were written by a committee of robots who have never touched a towel. Lucas: That's the million-dollar question. The difference between crap AI copy and decent AI copy comes down to two things: the prompt structure and the human edit pass. The brands doing it well treat the AI as a first draft machine, not a final output. Luna: So what does a good prompt look like? Walk me through it. Lucas: Sure. A bad prompt is something like, 'Write a product description for a linen sheet set.' The AI returns generic fluff — 'luxurious comfort,' 'premium quality,' that kind of thing. A good prompt specifies tone, length, target keywords, features, and the audience. Lucas: Here's an actual prompt I've seen work: 'Write a two-hundred-word product description for a queen-size linen sheet set in the color slate gray. Tone: warm, conversational, slightly aspirational. Include the keywords: breathable linen, sleep temperature regulation, easy-care fabric. Highlight: pre-washed for softness, available in six colors, machine washable. Write for a millennial homeowner who values sustainability but cares equally about aesthetics.' Luna: That's a lot of guidance. And I bet the output is way more usable. Lucas: Night and day. I tested this with a friend who runs a small bedding brand — his old prompts produced copy that sounded like a five-dollar dropshipper. After we rewrote the prompt structure, the AI started generating copy that needed maybe twenty percent editing before it was publishable. Luna: Twenty percent editing is still significant. But if you're scaling hundreds or thousands of SKUs, that edit cost is way lower than writing from scratch. Lucas: Exactly. The brands I see doing this well have a two-step workflow: Step one, generate first drafts using a structured prompt that includes the brand's style guide and SEO keywords. Step two, a human editor reviews each batch — usually a senior copywriter or content manager — and runs a three-question test. Luna: What's the three-question test? Lucas: First: Does this description contain any factual claim I can't verify? AI hallucinates details about materials or dimensions sometimes. Second: Does this sound like a person wrote it, or a thesaurus? If the word 'elevate' appears more than once, cut it. Third: Does this include at least one specific sensory detail — like the feel of the fabric, the sound of the zipper, the weight of the product? Luna: That third one is interesting. Sensory details are exactly what generic AI copy misses. I remember reading a description for a leather backpack that said 'premium full-grain leather' five times but never mentioned the smell or the way the leather softens over time. Lucas: Right. And those are the details that convert. I've seen A/B tests where adding one sensory sentence lifted conversion rate by eight to twelve percent on product pages. Luna: What about SEO? Does Google penalize ai generated content? Because there's been a lot of noise about that. Lucas: Google's official position as of early twenty twenty-six is that they don't penalize AI content per se — they penalize low-quality content, regardless of how it's produced. In practice, I've seen ai written pages rank just fine as long as they're useful and not just keyword stuffing. Lucas: In fact, some brands are using AI to specifically target long-tail keywords that were previously not worth writing a custom description for. Think about a niche variant like 'king-size linen sheet set deep pocket slate gray' — that's a search term with low competition but high intent. An AI can generate a targeted description for that variant in seconds. Luna: That's smart. DTC brands often have massive catalogs with hundreds of color and size combinations. Before AI, you either wrote one generic description for the whole line or you hired an army of freelancers. Lucas: And the freelancer route is expensive and hard to maintain consistency. I talked to a brand that spent forty thousand dollars a year on product copy for their catalog. They switched to AI plus one part-time editor and got better results for under fifteen thousand. Luna: That's a real cost saving. But I wonder — are there any downsides besides the quality risk? Like, does AI copy sound too similar across brands? Lucas: That's a real concern. If every DTC brand is using the same underlying model with similar prompts, you start to see a homogenization of voice. The solution is to feed the AI examples of your brand's unique writing — past emails, your about page, your social media captions. That helps the output match your specific tone. Luna: So you're basically fine-tuning the model on your brand voice without actually doing technical fine-tuning. Just prompt engineering with examples. Lucas: Exactly. Some teams even create a 'brand voice document' that they paste into the system prompt every time. It includes things like: avoid jargon, use contractions, start half of sentences with a conjunction, never use the word 'innovative'. Luna: I love that. I'm going to steal the 'never use innovative' rule. Alright, let me ask you this: should brands disclose that their product descriptions are ai generated? I see some debate about that. Lucas: It's a gray area. Most brands I know don't disclose it on the product page — they treat it as a production tool, like using a spell-checker. But a few are experimenting with transparency as a trust signal. For example, a small skincare brand includes a line at the bottom: 'This description was drafted with the help of AI and reviewed by our team.' And they've gotten positive feedback. Luna: That feels right. It's honest without being defensive. Okay, so if a DTC brand wants to start using AI for product descriptions tomorrow, what's the first step? Lucas: Two things. One: audit your current product catalog and identify the SKUs that have weak or missing descriptions — the ones with fifty words or no keywords. Those are your low-hanging fruit. Two: write one really good prompt that includes your brand voice, top keywords, and a few example descriptions you love. Test it on ten products, edit the output, and see if the quality meets your bar. Luna: Start small, iterate. That makes sense. And then scale once you trust the process. Lucas: Exactly. And if these marketing conversations have sparked something you've actually used, I'll mention that a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many episodes. No ads, just listener support. Luna: Yeah, it's a small group, but it makes a big difference. Keeps the show independent and lets us pick topics we actually find interesting. Lucas: Exactly. So if you've gotten value, that's the place. Otherwise, back to the AI stuff — I think the next frontier is using AI to A/B test multiple description variants automatically. Some platforms are already doing that. Luna: Oh, that's clever. Generate three versions of the same description and let the data pick the winner. Lucas: Right. And because the generation cost is near zero, you can test dozens of variations for a single product. One brand I know tested eleven different descriptions for their hero product and found a version that outperformed the control by twenty-two percent in conversion rate. Luna: That's a huge lift. And they probably wouldn't have had the resources to write eleven versions by hand. Lucas: Exactly. So the ROI on this is pretty clear. But I want to be careful not to oversell it — AI is not a replacement for a good copywriter. It's a force multiplier. The brands that treat it that way are winning. The ones that treat it as a magic button end up with product pages that sound like they were written by a robot. Luna: That's the key takeaway: AI as a tool, not a crutch. Alright, any final piece of advice for our listeners? Lucas: Yeah. If you're going to start, do it this week. Pick ten products, write a good prompt, generate descriptions, edit them, and put them live. Measure the organic traffic change over the next thirty days. You'll probably see a bump. And if you do, you'll know it works. Luna: Low risk, high potential reward. I like that. Thanks, Lucas. Lucas: Thanks, Luna. That's it for this episode.