Product Pages: Nearly an Hour to Under 5 Minutes. Organic Traffic: Nearly +30%.
A direct-to-consumer e-commerce company in Austin, TX operating three brand websites with thousands of active SKUs, mid-eight-figure annual revenue, and 120 employees.
A direct-to-consumer e-commerce company in Austin, TX operating three brand websites with thousands of active SKUs, mid-eight-figure annual revenue, and 120 employees.
What They Were Facing
A direct-to-consumer e-commerce company in Austin, TX operating three brand websites with thousands of active SKUs had a content bottleneck that was limiting growth. The company, with mid-eight-figure annual revenue and 120 employees, was adding new products faster than their content team could write descriptions, shoot lifestyle copy, and generate SEO metadata. At any given time, roughly 40% of their product catalog had incomplete or missing metadata, titles, and descriptions. Product page creation was painfully slow. For each new SKU, a content writer spent close to an hour crafting a description, writing SEO-optimized title tags and meta descriptions, generating alt text for product images, and creating structured data markup. The math was brutal: at 200+ new pages per seasonal launch, each launch required a dedicated content sprint that consumed the entire team for three weeks. The inconsistency problem was just as damaging. Three brands meant three sets of voice guidelines, and the content produced by five different writers ranged from on-brand to barely recognizable. Product descriptions on one brand's site used different formatting, feature ordering, and terminology than the same category on another brand. The SEO metadata gap was hurting them in search results, and they knew it, but never had the bandwidth to go back and fix the existing thousands of pages. The content team was burned out. They'd been hired to do creative work (brand storytelling, email campaigns, social content) and were instead spending most of their time on the most repetitive kind of writing imaginable: product description after product description, formatted identically, with minor variations.
40% of product catalog had incomplete or missing metadata, titles, and descriptions
Each new SKU required close to an hour of content writing for descriptions, SEO metadata, and structured data
Seasonal launches of 200+ pages consumed the entire content team for three weeks
Three brands with inconsistent voice across five different writers
Content team burned out on repetitive product descriptions instead of creative brand work
How We Solved It
We mapped the content creation workflow end-to-end and identified what information was already available in their product information management (PIM) system: specs, dimensions, materials, features, categories, and pricing. For most products, the vast majority of what a content writer needed to craft a description already existed in structured form. The writer's job was largely translating specs into prose and applying brand voice. That's exactly the kind of task AI handles well. We built a content generation pipeline that takes structured product data from the PIM and produces complete content packages: product descriptions in the correct brand voice, SEO title tags, meta descriptions, image alt text, and JSON-LD structured data markup. Each brand has its own voice profile that controls tone, vocabulary, sentence structure, and formatting conventions. The system doesn't produce identical outputs with different brand names slapped on. It genuinely writes differently for each brand. For quality control, we built a review workflow where generated content is staged for human approval before publication. The content team reviews and edits rather than writes from scratch. In practice, the majority of generated descriptions need only minor edits (tweaking a word choice, adjusting a claim), while the rest require more substantive revision, usually for products with unusual features that the system doesn't capture well from specs alone. We also ran a backfill operation on the existing catalog. Over three weeks, the system generated metadata for the 1,600+ pages that had been missing SEO elements. The content team reviewed these in batches, and the updated pages were deployed in a staged rollout to monitor search performance impact. Beyond product pages, we built a support knowledge base that indexes the company's return policies, shipping information, product care instructions, and warranty terms. Their customer service team uses this to answer common questions faster, and the same content feeds the FAQ sections on product pages.
Mapped content workflow end-to-end identifying the vast majority of description inputs already in the PIM system
Content generation pipeline producing descriptions, SEO tags, meta descriptions, alt text, and JSON-LD per brand voice
Each brand has its own voice profile controlling tone, vocabulary, sentence structure, and formatting
Review workflow staging generated content for human approval with the majority needing only minor edits
Backfill operation generating metadata for 1,600+ pages missing SEO elements in a staged rollout
Measurable Outcomes
Quantifiable improvements delivered within the project timeline
Reduced from nearly an hour to under 5 minutes including human review
Seasonal launches compressed from 3 weeks to a couple of days for 200+ pages
Organic search traffic increased nearly 30% within 90 days of catalog backfill
Metadata coverage improved from around 60% to over 99% across all three brand sites
Majority of writing capacity freed for brand storytelling and campaigns
Support knowledge base with average query resolution under 15 seconds
The nearly 30% organic traffic increase came primarily from the metadata backfill on existing pages. Hundreds of product pages that had been invisible to search engines because they lacked title tags, meta descriptions, and structured data were suddenly indexable and ranking. The traffic gains appeared within 6-8 weeks of the staged deployment and continued climbing through the 90-day measurement window. The real win, according to the VP of Marketing, was getting the content team back. Before the automation, they had three content writers spending most of their time on product descriptions. After, those same writers produce more brand content in a month than they had in the previous quarter. The holiday campaign they produced during the first quarter post-launch significantly outperformed the prior year.
Implementation Timeline
A structured approach from discovery to deployment
Analyzed existing content and product data sources
Weeks 1-2Analyzed existing content and product data sources
Brand voice tuning and generation system development
Weeks 3-4Brand voice tuning and generation system development
Human approval workflow for generated content
Week 5Human approval workflow for generated content
Generated metadata for 1,600+ pages missing SEO elements
Weeks 6-8Generated metadata for 1,600+ pages missing SEO elements
Indexed return policies, shipping info, product care, and warranties
Week 9Indexed return policies, shipping info, product care, and warranties
Complete rollout across all three brand sites
Week 10Complete rollout across all three brand sites
Frequently Asked Questions
How do you maintain distinct brand voices across three different websites?
Each brand has a voice profile that functions like a style guide the system follows. We built these profiles by analyzing dozens of existing approved descriptions per brand, identifying patterns in tone, sentence structure, vocabulary, and formatting. The profiles capture differences like whether a brand uses casual contractions or formal language, whether it leads with benefits or features, and what kind of adjectives it favors. During testing, brand managers reviewed blind samples and correctly identified which brand each description was written for the vast majority of the time.
What about products where the specs don't tell the whole story?
Some products, especially new category entries or items with unique selling points that aren't captured in specs, need more human input. The system flags these based on configurable rules (new category, missing key attributes, significant price premium over similar items). For flagged products, a content writer gets a partially generated draft with notes on what the system couldn't determine from the data alone. It's a starting point, not a finished product.
How did you handle the metadata backfill without disrupting existing search rankings?
We deployed the backfill in stages: 200 pages per week, starting with the lowest-traffic pages. This let us monitor for any negative ranking impacts before touching high-performing pages. We also preserved any existing metadata that was already performing well. The system only generated content for pages where metadata was missing or flagged as thin by the SEO audit. No well-performing page was changed.
Can the system generate content for new product categories it hasn't seen before?
It can, though the quality is lower for categories without training examples. When a genuinely new category launches, we recommend having the content team write 10-15 descriptions manually. Those get added to the brand's voice profile as category-specific examples, and subsequent descriptions in that category are significantly better. Most of the time, though, new products fit into existing categories and the system handles them without any additional training.
Services Used in This Project
Ready for Similar Results?
Schedule a discovery call to discuss your specific challenges and learn how we can deliver measurable outcomes for your organization.
Schedule Discovery Call