How we got 879 transactions from AI thanks to GEO
Generative Engine Optimization (GEO) is not about the future. It is already here today. In this case study, we will show how our team used GEO optimization to generate 879 transactions in just 5 months for a US-based tech e-commerce store, primarily from ChatGPT traffic.
Context and task
In January 2025, we were approached by an online electronics store from the US specializing in the sale of products such as headphones, power banks, smart watches, and other smartphone accessories. They already had SEO promotion and contextual advertising, but the client wanted to enter a new channel – LLM search. The main goal was to:
- Get into the answers of ChatGPT, Perplexity, Copilot, etc.
- Get targeted traffic without additional costs
- Ensure trackability of transitions from each platform
What we did — 5 GEO solutions
| GEO Action | Implementation Details | Goal |
| 1. Product JSON-LD Markup | Added Product, HowTo, FAQPage, Review | Enable LLMs to see structure and “read” product specifications |
| 2. GEO-FAQ | Created 5-7 questions per category: simple, short, similar to user input | Become a reliable source for chatbot answers |
| 3. Block Optimization | Paragraphs ≤ 46 words, lists of 7 items, clear CTAs | Ideal for content chunking in LLM responses |
| 4. ChatGPT Testing | Entered queries like real users, analyzed AI-generated responses | Identified which blocks appear in AI recommendations |
| 5. UTM Tracking | Added distinct tags like utm_source=chatgpt etc. | Accurately measured performance of each LLM-based traffic source |
Result: how transactions from GEO channels increased
The data below is the result of 5 months of work. Everything you see is organic traffic from generative models.
Transactions by referral source (January–May 2025)
| Source | Transactions |
| ChatGPT | 773 |
| Perplexity | 75 |
| Microsoft Copilot | 16 |
| Google Gemini | 14 |
| Mistral | 1 |
| Total | 879 |
ChatGPT is the clear leader here, accounting for 88% of transactions.
Monthly dynamics
| Month | Transactions | Dynamics |
| January | 114 | — |
| February | 114 | 0 % |
| March | 172 | +51 % |
| April | 309 | +80 % |
| May | 170 | −45 % (after the peak) |
The biggest jump came in April, when we updated our FAQ, added a HowTo structure, and introduced a separate subdomain for technical solutions.
Why did it work?
- ChatGPT and Perplexity use structured data. If your website has logically presented, easy-to-read, and well-marked content, LLM is highly likely to include it in its response.
- We “adapted” to the logic of the models. We created content that could be easily integrated into text responses.
- The content was regular and fresh. Within 10–14 days after the update, we saw a reaction in the graphs.
Insights and recommendations from MIM:AGENCY
| What to Do | Why It Matters |
| Update structured data monthly | LLMs cache outdated data |
| Phrase queries like your customers | Increases the chances of being included in AI-generated responses |
| Account for new platforms (Gemini, Copilot) | Their share is growing faster than ChatGPT |
| Break content into small chunks | Models are more likely to cite a short, concise text |
| Set up separate GEO analytics | It’s a distinct channel – don’t mix it with SEO or Ads |
Conclusion
Generative Engine Optimization is no longer an experiment, but a full-fledged channel for attracting traffic. In practice, we have proven that a properly structured website adapted for LLM can generate hundreds of targeted transactions per month.
ChatGPT works as a new “search engine,” and those who adapt first will get a lot of traffic.
If you are an e-commerce business in a complex niche, especially a technical one (spare parts, equipment, medicine, construction), GEO promotion may become the most effective new source of customers as early as 2025.
