A DTC heritage skincare brand, founded 2018

The challenge

The brand had built a clean, profitable organic channel over six years. Roughly 38 percent of monthly revenue came from non-branded search, mostly informational queries around sensitive skin, ingredient explainers, and routine builders that funneled cleanly into product pages. When AI Overviews rolled out broadly to their core query set in early 2026, the floor dropped. Organic sessions on the top 40 informational posts fell 64 percent in six weeks. The team's first instinct was to publish more content, faster. They added two writers and pushed weekly output from four posts to nine. Nothing recovered. By the time we got the call, the founder was looking at a $312k revenue gap in the previous quarter and a finance team asking whether the channel was structurally broken. The honest answer was that the channel had not broken, but the rules had changed and the brand had not.

What we shipped

The first decision was to stop publishing new posts entirely for six weeks and rebuild the existing top 40. We treated each as a citation-target page rather than a click-target page, on the working assumption that AI Overviews would keep absorbing the informational click and the job was now to be the cited source inside it.

The rebuild had four parts. First, restructured intros: every post led with a one-sentence direct answer, followed by a short structured summary that mapped to common follow-up questions. Second, ingredient and product schema across the catalog, with proper Product markup including aggregateRating, plus HowTo and FAQPage on the routine guides. Third, an author and reviewer entity graph: the founder is a trained cosmetic chemist, and we built her out as a proper named entity with a sameAs graph linking to her ORCID, two academic papers, and three external publications. We added a "medically reviewed by" line with a real licensed dermatologist whose credentials we could verify.

Fourth, we built an llms-full.txt at 1.4MB combining the ingredient encyclopedia, the routine guides, the safety data, and the heritage origin story. We submitted the brand to the three ingredient databases that AI Overviews most often cites and got accepted to two. On the product pages, we added structured comparison blocks for the three most-asked "X vs Y" prompts. Finally, we set up tracking for AI Overview appearances using a vendor tool plus manual weekly checks on 60 priority prompts.

The numbers

The traffic recovery was slower than the revenue recovery, and that surprised us at first. What we eventually understood was that the AI Overview was doing real qualification work. Users who clicked through after seeing the brand cited inside an AIO arrived already convinced on the ingredient story and the founder's credibility. They added to cart at nearly twice the rate of pre-AIO organic, and they bought more units per order because the AIO had often framed a full routine rather than a single product. Branded search nearly doubled by month six, which we read as users who saw the brand cited, did not click immediately, and came back later through brand. The customer support team also reported a noticeable drop in ingredient-safety questions in chat, because the AI conversations were already handling that layer. Return rate on organic-acquired customers dropped from 11 percent to 7 percent over the engagement, which the operations team tied to better pre-purchase education on ingredient interactions.

What we'd do differently

We were too cautious in month one about touching the original post copy. The founder had written most of those posts herself and there was real attachment. We spent two weeks negotiating edits we should have made in two days. Once we shipped the first ten rebuilds and showed citation movement, the rest got approved in a day. We also should have started the dermatologist reviewer relationship in week one, not week four. That single byline change correlated with a measurable citation lift on the safety-sensitive posts and we left three weeks of compounding on the table. We did not start the structured ingredient submissions early enough either; the two databases that accepted us had a six-week review queue we could have entered on day one.

What's next

The engagement converted into an ongoing monthly retainer covering content maintenance, citation tracking, and quarterly schema audits. The next focus is international: the brand is launching in two European markets in Q3, and the LLM citation playbook needs to be rebuilt for each language. German and French AI responses on skincare lean on different authority sources, and we are mapping those now. We are also building a product launch playbook so that new SKUs land with full citation coverage from day one rather than catching up six months later. The goal for the next two quarters is to push citation share on the priority prompts past 65 percent and to extend the work into TikTok Search and Pinterest, which are showing similar AI-summarization behavior. A small parallel project covers retailer and marketplace surfaces, since Amazon and Sephora are both building their own retrieval layers and the same authority signals carry over.