Latest / The DTC Marketing Podcast with Fexingo: Direct-to-Consumer Brands, Shopify, and E-Commerce Growth / How DTC Brands Use AI Product Descriptions for SEO
Transcript
- Lucas: So Brooklinen — the bedding brand that basically defined DTC sheets — recently did something that I think a lot of Shopify store owners are going to start copying. They used AI to rewrite every single product description on their site. And I'm not talking about a quick ChatGPT pass. They built a custom GPT model trained on their brand voice and pushed out over 700 product pages in less than three weeks. Luna: Three weeks for 700 pages? That would have taken a team of human copywriters at least a few months. Lucas: Easily. And the result was a 23 percent lift in organic traffic within two months. Not because the AI copy was better than human copy — but because they finally had unique, keyword-rich descriptions for every product. Before, they had maybe thirty core descriptions that they just reused with slight tweaks. Google hates that. Luna: Right — duplicate content penalty, or at least a ranking demotion. So the AI solved a scale problem more than a quality problem. Lucas: Exactly. And that's the key distinction I want to make. A lot of DTC founders I talk to are worried that AI copy will sound generic and hurt conversions. But Brooklinen's conversion rate actually held steady — it didn't drop. The traffic bump more than made up for any tiny dip in per-visitor conversion, if there even was one. Luna: But did they test this against a control group? Or was it a full rollout? Lucas: Full rollout, unfortunately. So we don't have an A/B test. But they did monitor brand sentiment in customer service tickets, and they saw no increase in complaints about confusing or off-brand copy. That's a good sign. Luna: I'd still want to A/B test it. But I get that when you're scaling to hundreds of products, you can't hand-write everything. So what was their actual process? How did they train the model on brand voice? Lucas: They fed it about fifty of their best-performing human-written descriptions, plus their brand guidelines document and a list of keywords they wanted to rank for. Then they created a structured prompt that included the product name, the key features, the materials, and the target keyword. The output went through a human editor — one person — who caught obvious errors and made sure the tone matched. Luna: So one human editor versus a whole team. That's a massive cost saving. Lucas: It is. But here's where I get skeptical. If you're a small DTC brand with ten products, is this worth it? You can probably write those ten descriptions yourself in an afternoon and they'll be better than anything AI produces. Luna: Yeah, the break-even point probably sits somewhere around fifty to a hundred products. Under that, human copy wins on quality and cost. Over that, AI starts to make sense. Lucas: And the SEO play only works if you're actually targeting keywords that have search volume. If you sell hand-carved wooden spoons and nobody searches for that exact term, unique descriptions won't help. You need keyword research first. Luna: Alright, let's get practical. What's the actual prompt that works? Because I've tested a few and gotten stuff that sounds like a robot wrote it. Lucas: I've been using a framework from a DTC brand called Buck Mason, which is a clothing brand. Their prompt template is roughly: 'Write a product description for that emphasizes. Use a conversational tone, keep sentences under twenty words, and include the primary keyword in the first sentence and at least once more in the body. Do not use cliches like 'elevate your style' or 'game-changer.' End with a sizing or care note if applicable.' Luna: That's good. The 'no cliches' rule is crucial. So many AI descriptions are full of that empty aspirational language. Lucas: Right. And Buck Mason then runs each description through a grammar checker and a human review. They told me the AI gets it right about 70 percent of the time. The other 30 percent needs manual tweaks — usually around sizing or fabric details that the model hallucinates. Luna: Hallucinations in product descriptions can be dangerous. Imagine telling customers a shirt is machine washable when it's actually dry clean only. Lucas: Exactly why you need that human editor. But the time savings are real. Buck Mason went from three hours per description to thirty minutes per description. That's an 83 percent reduction. Luna: Now, what about Google's helpful content update? Some SEOs claim that ai written content gets penalized. Have you seen any evidence of that with product descriptions? Lucas: I haven't. Google's official stance is that they care about quality, not how it was produced. If the description is useful, original, and accurate, it ranks. The problem comes when people publish thin, keyword-stuffed AI content that adds no value. But a good product description that tells the customer exactly what they need to know — that's helpful regardless of who or what wrote it. Luna: So the risk isn't AI per se, it's bad AI. And bad AI is usually a result of bad prompting or no human review. Lucas: Exactly. And I think we're going to see more tools emerge that are purpose-built for this. Jasper and Copy.ai already have e-commerce templates. But the real game-changer will be when Shopify integrates AI description generation directly into the product page editor. I think that's coming within a year. Luna: Actually, Shopify just announced a beta feature for ai generated meta descriptions. Product descriptions are the logical next step. Lucas: That's huge. If it's built in and trained on store data, the quality could be much better than a generic ChatGPT output. But until then, brands need their own process. Luna: So let me ask you this: for a brand that's listening right now and thinking about trying this, what's the one thing they should do first? Lucas: Audit your current product pages. Check if you have duplicate descriptions or pages with no description at all. If you have more than fifty products with thin content, AI is worth testing. Start with your top ten products, write custom prompts, review the output, and compare the conversion rates against your existing pages. Do a two-week test. Luna: I like that. Low risk, high potential upside. And if it works, you scale. If it doesn't, you've only wasted a couple of hours. Lucas: Exactly. And if it does work, you might see a traffic lift like Brooklinen did. Not bad for a tool that costs twenty bucks a month. Luna: Speaking of tools that cost little but deliver a lot — if these marketing conversations have sparked something you've actually used, whether it's a new prompt technique or a tool you tried because of the show, that's exactly the kind of thing that keeps this podcast ad-free. We'd love your support at buy me a coffee dot com slash fexingo. Lucas: Yeah, even a small contribution helps us keep digging into these topics without sponsors dictating the content. And we genuinely read every message people leave there. Luna: Alright, back to AI descriptions. Lucas, you mentioned Buck Mason's prompt — do you have a sense of what their SEO lift looked like? Lucas: They saw a 15 percent increase in organic impressions over three months, but they're a smaller brand so the sample size is smaller. Still promising. The key takeaway is that AI descriptions are not a magic bullet. They're a scaling tool. If your fundamentals — keyword research, site structure, page speed — are solid, AI can amplify that. But it won't fix a broken SEO strategy. Luna: Right. Garbage in, garbage out. So let's talk about the tools. Which one are you seeing DTC brands adopt most? Lucas: Jasper is popular because of its brand voice feature. You can train it on your existing copy and it remembers style guidelines. Copy.ai is cheaper and faster for bulk generation, but the quality is less consistent. And then there's ChatGPT with custom GPTs, which is what Brooklinen used. That's the most flexible but requires more technical setup. Luna: For a non-technical founder, I'd recommend starting with Jasper. The onboarding is smoother. Lucas: Agreed. And whatever tool you pick, invest time in the prompt. A good prompt is worth more than a good model. I've seen people get mediocre results from GPT-4 and great results from GPT-3.5 just because they had a better prompt. Luna: So the prompt is the differentiator. That's a good place to leave it. If you're a DTC marketer, go audit your product pages tonight. If you've got thin content, AI might be your cheapest growth lever right now. Lucas: And if you try it, let us know how it goes. We'd love to feature your results in a future episode.