Keyword Research Beyond Tools for AI Search | Venkata Pagadala

With Venkata Pagadala, AI and search product leader at AT&T · Streamed live · 43 min

What this episode covers

Venkata Pagadala explains why keyword tools miss real demand: long conversational queries show zero volume, and customer pain points in reviews never appear in tools. He recommends persona-based fan-outs, mining reviews and communities, and single-purpose AI agents to collect and classify the data.

Key takeaways

  • ✦Treat zero volume on long, multi-part queries as a tracking gap, not zero demand.
  • ✦Map buyer personas and mine real questions from Reddit and YouTube comments.
  • ✦Use one job per AI agent and store results with timestamps.
  • ✦Use your own customer reviews and sales data to find gaps.

Key moments

  1. 12:26Pain points missing from tools
    “I saw the pain points were never spoken in keyword tools.”
  2. 13:52Persona-based query fan-out
    “Persona-based query fan-out is a method I'm following right now.”
  3. 14:46Zero-volume long queries
    “When the query crosses three words or four words, usual traditional tools say zero.”
  4. 15:40Think beyond SEO
    “The real SEO starts when you start looking not like an SEO.”
  5. 28:19Reviews as a source
    “Reviews are like a gold.”

Quotes are from the episode transcript, lightly cleaned of auto-caption errors.

Questions answered in this episode

Why does keyword research miss demand for long queries?
Traditional tools often show zero volume for queries longer than three or four words because they are not tracked, but the speaker argues these queries still reflect real demand, citing a car-buying example done through ChatGPT.
How can I build an AI agent to research my market?
Create a single-purpose agent in Claude with a defined objective, such as finding podcasts or speaking opportunities, then store the classified results in Excel or a Postgres database with timestamps.
How do you measure whether an AI agent is working?
Measure whether it gets the job done without noise, how many tokens it consumes, and its success rate at agent and multi-agent level, comparing cost against returns.

More episodes on AI Search & LLMs