Challenge
An online brand selling products in a sensitive, health-related niche. Before buying, customers ask a lot of follow-up questions: ingredients, allergens, who the product is for, what to replace it with. And more and more often they ask AI: Google AI Overviews, AI Mode, or chatbots.
What made it hard
- A YMYL topic. The subject is health-related, so search engines and AI systems only cite sources they trust.
- Broken structured data. At the start, only 93 pages had valid markup, and 684 pages had errors. For AI systems, this means product and article data is read poorly or not at all.
- Indexing problems. 1,906 pages were outside Google's index.
- A shift in demand. Part of search is moving to zero-click answers. The brand needed to be present in those answers, not just in the classic results.
The goal: make the site easy for AI systems to understand and increase the brand's presence in Google's AI answers.
What we did
We worked on a GEO model: technical foundation first, then content for the AI answer format.
Structured data audit and fixes
We audited structured data across the entire site and systematically fixed the errors. The goal was simple: every important page should pass correct data to search engines about what is on it. We brought the number of pages with errors down to practically zero, and the number of pages with valid markup multiplied.
Index cleanup
We went through the pages outside the index, separated the important ones from the technical ones, and reduced the number of pages Google was not including in its index.
Content for AI answers
We reviewed the pages that were already appearing in AI answers and widened that circle. A structure with clear answers to customers' questions, easy-to-read sections on ingredients and product use, and logical links between pages.
Measurement
We separately tracked impressions in Google AI features, AI's share of all impressions, the number of pages used in AI answers, and, alongside that, classic clicks and sessions from LLMs to see the full picture.
Results
Report for June-July 2026 compared with April-May:
- Impressions in Google AI features: 168,250 → 1,088,226 (+546.7%).
- AI features' share of all impressions: 3.4% → 26.1%.
- Pages used in AI answers: 621 → 841.
- Pages with valid structured data: 93 → 914.
- Pages with structured data errors: 684 → 1.
- Pages outside the index: 1,906 → 1,307.
To be honest about the downsides: classic clicks fell 12.9% over the same period, and sessions from LLMs fell 8.7%. We attribute this to a shift in demand: users get the answer directly in Google's AI block and do not always click through to the site. That is exactly why presence in these answers became critical: a brand that is missing from the AI answer loses both the impression and the click.
What makes this project special
This is one of the cleanest examples of GEO in our practice: the main effect came not from ads or new links but from the technical quality of the data. Once the markup was correct on hundreds of pages, Google's AI features started showing the brand many times more often. In a niche where customers are especially picky about sources, this is an advantage competitors will find hard to replicate quickly.
April-May 2026 (before) vs. June-July 2026 (after): +546.7% Source: Google Search Console, client report, April-July 2026
Published: · Updated: