Your Keyword Strategy Is the Wrong Unit of Analysis for AI Search

Keyword research has been the foundation of content strategy for 20 years. You find the terms people search for, map them to intent, write content that targets them. It works. Google has trained us to think this way.

AI search breaks the unit of analysis.

Users don’t type “best project management software” into ChatGPT. They ask: “We’re a team of 15, half remote, doing mostly client work. What project management tools would make sense?” The intent is the same. The form is completely different. And the content that gets cited in the second case is not the same content that ranks for the first.

This is what’s called prompt research — and it’s the emerging equivalent of keyword research for AI visibility.

Prompt research asks: what are the actual questions people pose to AI platforms about my category? Not the keywords, but the full conversational queries. “How does X compare to Y for small teams?” “What should I look for when choosing Z?” “Is A or B better for European markets?”

These prompts are the moments of decision. A buyer types this into ChatGPT at the beginning of a vendor evaluation. The answer shapes who makes the shortlist. The companies that appear are the ones who published content that clearly answers those specific questions.

Most companies have never done this analysis. They know their keyword rankings. They have no idea which prompts trigger their brand, which ones trigger competitors, and which ones return an answer that doesn’t mention their category at all.

The disconnect is telling. Traditional keyword research is built around search volume and competition. Prompt research is built around buyer intent at the AI interface. The two have different shapes. A query with enormous Google search volume might generate almost no AI interaction. A conversational prompt with low traditional search volume might drive high-stakes buying decisions through AI.

The practical starting point is manual: query the AI platforms your customers use with the questions they’d actually ask. Record what comes back. Notice who’s cited. Notice whether you’re there.

Then think about what content would need to exist for you to be cited on those prompts. Usually it’s comparative articles, use-case-specific explainers, and direct question-and-answer content — not the keyword-optimized pillar pages that traditional SEO produces.

Beket.ai runs this analysis systematically — querying AI models with the prompts most relevant to your business, scoring what comes back, and identifying where the gaps are. It’s the first step to actually knowing where you stand.