Aurora was a five-year-old DTC home goods brand with a small but committed audience. Their analytics looked enviable in the top half of the funnel — strong organic traffic, high engagement, healthy add-to-cart rates. Below the cart, everything fell off a cliff.
The product detail pages had been written once, at launch, by a junior copywriter on a deadline. They described the product. They did not sell the product. Three or four short sentences each. Every PDP read like a spec sheet.
The team had a hypothesis — "we need better copy" — and the entrance to a different problem: writing 1,400 product descriptions, manually, with brand voice, with seasonal variation, and re-writing them every time the data changed. Their two-person marketing team did not have the bandwidth, and an agency quote came in at six figures plus retainer.
We connected the engine to Aurora's analytics, search console, Shopify, and heatmap stack. Within 48 hours it had clustered the PDP catalog into five voice patterns based on category, price band, and existing PDP performance.
The engine wrote a candidate rewrite for each PDP, ran it past its internal critic against Aurora's brand voice document and three negative-control prompts, then deployed the rewrite to 18% of traffic for a 48-hour evaluation window. Variants that beat baseline by more than two percent rolled out to 100% of traffic; variants that didn't were silently rolled back.
Outcomes are public on this page. The orchestration graph, the operator names, the prompts, and the evaluation harness stay under NDA. That is the deal.
100% wool throw blanket. 50" x 60". Hand-loomed. Available in three colors. Machine washable on cold.
Heavy, honest wool that drapes like it remembers somewhere colder. Loomed by hand in earth tones that read as found, not bought. Wash it cold. Keep it for a decade.
Within forty-eight hours of the first rewrite cohort going live, PDP conversion across the 18% variant traffic was running 186% above baseline. The variant was promoted to 100% of traffic. The engine kept iterating — 73 autonomous cycles in the following twelve weeks, each one a small rewrite, a small test, a small win.
Aurora's marketing team didn't add anyone. The founder reads a weekly memo and approves a handful of strategic changes per quarter. The cost of the engine, on a per-rewrite basis, came out to under five cents per product change.
I stopped writing PDP copy. The engine does it. I read the Friday memo and ship a strategic call. My team is half the size it was eighteen months ago. We're shipping six times the change.— Placeholder · Founder, Aurora Home Goods