Keyword Research in the Age of AI

Why classic search data still matters

Google's keyword volume data remains the broadest observable dataset of human curiosity. Billions of queries create durable intent graphs that tell us what people consistently ask for and how they phrase it when they want an answer fast. Even as AI gains traction, nothing else offers that same longitudinal, global view of demand. So the workflows we built on top of this data—clustering, SERP analysis, programmatic structures—should not be tossed aside.

More searches, new behaviors

Large language models changed how people approach research. Instead of a single keyword, a session now looks like a rolling dialog, with follow-up prompts happening seconds apart. Every turn becomes a search, whether it's visible to the user or abstracted away by the agent. That means net-new intent surfaces constantly: clarifications, comparisons, localized variants, adjacent problems the user didn't know to type into a box. The volume of search acts goes up even if traditional search boxes plateau.

Agents have their own keyword language

When agents browse on our behalf, they don't lean on years of "what works in Google" muscle memory. They fire what are effectively blind queries, often verbose and hyper-specific: “best EU SaaS vendors for multilingual knowledge base widgets 2025” is typical agent phrasing. These strings would never show up in Keyword Planner because they were never part of human habit; they’re created on the fly to satisfy a goal. That gap explains why existing data skew human-first while the agent layer feels invisible.

Research for humans, optimize for agents too

1. Keep your quantitative base. Run keyword research the way you always have to capture proven human demand. Prioritize topics with steady volume and clear commercial intent.
2. Model likely agent prompts. If you've ever wired a web-search tool into an agent, you know it constructs queries by combining the user request, context, and constraints. Reverse-engineer that behavior: list the modifiers, specs, and data points an agent would care about for your product.
3. Structure content for machine scanning. Agents hunt for fast signals—explicit numbers, tables, comparison matrices, bullet lists with labels. Layer these into your human-friendly narrative so an LLM can grab and cite them quickly.
4. Target emerging "agent keywords." Even without volume metrics, you can seed these phrases throughout headings, captions, and schema. Treat them like long-tail bets that help you surface when an agent assembles its own query cocktail.