How GEO helped Runpod 4x new paying customers from ChatGPT

Runpod grew new paying customers 4x in 90 days by treating ChatGPT as an acquisition channel, tracking the prompts its buyers actually ask and fixing the pages AI could not fully read. ChatGPT now sends about 40 new customers a day, and AI visitors convert at 8%.

Runpod is a GPU cloud for training, fine-tuning and deploying AI models. Its buyers are AI-first by definition, so AI search was the obvious place to win them.

The challenge

Runpod's ideal customer is technical and high-intent: DevOps engineers, indie AI builders, academic researchers and platform teams weighing alternatives to AWS, CoreWeave and Google Cloud. These people ask ChatGPT which GPU cloud to use before they ever visit a vendor site.

The team started working on AI search in early 2025. Those early efforts soon plateaued at around 300 to 400 new paying customers per month.

The problem was visibility into the channel itself. The team was tracking only 50 prompts, which could not cover the range of personas, funnel stages and niche technical topics its buyers search. And some content was not being cited at all, with no clear reason why.

To grow, they needed four things:

• Prompt tracking at the level of individual models, mapped to personas and funnel stages
• Visibility across multiple LLMs without the cost ballooning
• Exportable data to steer content strategy and bring writing in-house
• A way to find the technical gaps stopping pages from being cited

What they changed

In April 2025 the team rebuilt its GEO program around three moves.

1. Widened prompt coverage from 50 to 300. The new set targeted the high-intent, niche queries that match Runpod's technical buyer. Each prompt was tracked for mentions, sentiment and ranking, so the team could see exactly where it won and where it was absent.

2. Focused on the engine that mattered. ChatGPT accounts for approximately 98% of Runpod's referral traffic. Rather than spreading effort evenly, the team prioritized prompt-level performance there.

3. Fixed what AI could not read. A site audit surfaced rendering problems that were blocking citations. Several articles were only partially indexed: AI saw the title but not the body. The team updated metadata, restructured the affected pages and corrected formatting. That unlocked full visibility in LLMs.

With performance data per prompt, the team also brought content production fully in-house, writing to close the specific gaps the data showed.

The results

Within 90 days, ChatGPT went from a plateaued experiment to one of Runpod's top-performing growth channels.

• 4x growth in new paying customers per month
• About 40 new customers per day from ChatGPT, on a two-week close cycle
• 8% conversion rate, with 2,100 conversions from roughly 28,000 visitors
• Prompt coverage up from 50 to 300 tracked prompts

The conversion rate is the number to notice. Visitors arriving from an AI answer have already been told you fit their need, so they show up ready to buy.

What you can copy

Track the prompts your buyers ask, not a sample. Fifty prompts could not describe a technical, multi-persona market. Three hundred could. Build your prompt set by persona and funnel stage, and include the niche questions only real practitioners ask.

Find your dominant engine and start there. For Runpod, ChatGPT drove approximately 98% of referral traffic. Check your own referral data before splitting effort across every model.

Audit for partial indexing. If AI can see your headline but not your body copy, it has nothing to quote. Check that your key pages render fully without heavy JavaScript, with clean metadata and clear structure.

Let prompt data drive the content calendar. Write to fill the gaps where you are missing from answers, then measure whether those prompts move.

Measure customers, not just mentions. Runpod judged the program on paying customers and conversion rate. Tie AI visibility to revenue and the budget conversation gets much easier.