Prompt Research: The Successor to Keyword Research

Keyword research was always a proxy discipline: from two or three typed words, infer what a human actually wanted. Entire methodologies existed to reverse-engineer intent from fragments like "crm small business." Prompts dissolve the proxy. People tell assistants their situation in full sentences: what they run, what constrains them, what they are afraid of, what they already tried. The research discipline that replaces the keyword spreadsheet is prompt research, and it produces a different, richer map of demand.

Why the difference matters

"Best CRM" is one keyword. In prompt-space it explodes into distinguishable situations: the solo lawyer who needs conflict-checking, the nonprofit on a grant budget, the sales lead migrating off a spreadsheet, each phrased as a paragraph with constraints attached. Engines answer each situation differently, retrieving different sources and naming different vendors. Competing in prompt-space means knowing which situations you win, which you lose, and which nobody has claimed, at a resolution keyword tools structurally cannot see.

Where prompt data comes from

No tool exports the prompt logs of the engines, so the discipline is assembly:

• Sales and support conversations. Buyers increasingly narrate their AI research ("I asked ChatGPT and it said..."). Harvest the phrasings verbatim; they are ground truth.
• Your search query data, re-expanded. Long-tail queries from Search Console are compressed prompts. Rehydrate them into the situations that produced them.
• Communities. Reddit, Slack groups, and forum threads in your category are people asking buying questions in natural language: prompt phrasing in the wild.
• The engines themselves. Ask them what people ask ("what do buyers typically want to know when choosing X"), then probe the follow-up suggestions engines offer mid-conversation, which reflect real usage patterns.
• Your own prompt tracking. Once the prompt universe is running, wrong-but-interesting answers reveal adjacent situations you had not mapped.

From research to roadmap

Cluster the collected prompts by situation, not by phrasing. For each cluster, run the prompts and record: who gets named, which sources get cited, and whether any retrieved page actually addresses the situation. Three findings drive action. Clusters where cited sources are thin or generic are content opportunities: build the page that genuinely answers the situation, in the formats engines cite. Clusters where competitors dominate reveal which third-party sources you need presence on. And clusters where you get named but mischaracterized are entity and consistency fixes, usually the fastest wins available.

The output replaces the keyword spreadsheet: a living map of buyer situations, the answers currently served, and your share of them. It is smaller than the old ten-thousand-row export and dramatically more honest about where revenue actually comes from.