Articles·AI agents

“I didn’t do one.” What an AI research agent does before a sales meeting.

Chris Cordon · 4 min read
A producer in a parked car checks his phone while a robot in the passenger seat hands him a one page brief.

An AI research agent looks up a prospect or client before you talk to them, checks what you already know about them, and hands back sourced facts plus the questions you should ask. It fixes the prep most salespeople skip, because doing it by hand takes longer than the gap between meetings. How long did a meeting prep brief take me by hand? That’s easy. I didn’t do one.

What did meeting prep look like before a research agent?

Before a research agent, my meeting prep was Google, about 15 minutes before the meeting. I’d scratch down a couple notes and then wing it. Plenty of producers prepare the same way. Real research on a business takes longer than the time between calls, so it gets skipped.

Now it takes about five minutes, and I don’t do the searching. The research agent goes out, finds relevant data on the person and the business, and uses the Sandler pain funnel to give me six questions to ask before I leave the meeting.

Why is AI research hard to trust?

AI research is hard to trust because a general AI assistant will fill a gap with something that sounds right. Ask a chatbot about a small agency and it may hand you a founding year, an employee count, or a website address it made up. Walk into a meeting with one wrong fact and you’ve spent your credibility in the first minute.

The other problem is forgetting what you already know. A web search has no idea you emailed this person in March.

Wanna fix it?

Meeting prep you never have time for gets fixed with an AI research agent that checks your own records first, researches with sources, labels what it can’t confirm, and ends with questions tied to real facts. Here’s how mine works. It runs on Claude.

  1. Know who you’re meeting. The agent sorts the target first, for example an agency prospect or an employer group. Each type gets its own checklist.
  2. Check your own records first. It searches my shared memory and my CRM for any prior contact before it touches the web.
  3. Research a set list. For an agency prospect, it looks for ten things, including who owns the agency and for how long, lines of business, carriers and platforms, how clients reach them, hiring signals, reviews, and signs of manual work.
  4. Every fact needs a source. Each fact traces to a web page, a post, a memory entry, or a CRM record. Anything it can’t trace gets written as “unknown, ask.” My research agents are banned from inventing links.
  5. Questions tied to facts. The six questions follow the Sandler pain funnel, and each one has to point at something specific from the research. Generic discovery questions get thrown out.

It finishes with the one thing I should know before I dial.

What does it look like once it’s running?

Once it’s running, prep is a request and a five minute wait. I ask for research on a company and a person. The agent does the searching, and I get back facts with sources, gaps marked “unknown, ask,” and six questions.

The same research agent works before a cold call, or before a conversation with an employer group about their benefits. The meeting brief that comes out of it, what goes on the page and when it should land, has its own writeup: [how to prepare for a sales meeting in 30 minutes](/articles/sales-meeting-prep).

The part I trust most is the “unknown, ask” line. A gap in the research becomes a question for the meeting, and nobody walks in with a made up fact.

Can you do this yourself?

Yes. Before a meeting, open ChatGPT, Claude, or Perplexity with web search turned on. Ask who owns the business, what changed there in the last year, and what they say about themselves on their website. Then add one line: “Cite a source for every fact, and write ‘unknown’ when you can’t find one.” Click through two of the sources yourself.

That covers one meeting a day. It stops working when you have several, when your own email history needs to be in the mix, or when you catch yourself skipping the source check. One meeting, prompt it by hand. A calendar full of them, build the agent.

Quick answers

What is an AI research agent?
An AI that researches a person or business for you, checks your own records, and returns facts with sources.
Is an AI research assistant the same thing?
Close. An assistant answers questions as you ask them. A research agent runs a whole checklist and hands back finished research.
How long does AI meeting research take?
About five minutes for mine.
How do you stop AI research from making things up?
Require a source for every fact, and have it write “unknown, ask” when it can’t find one.
What questions should it prepare?
Six, built on the Sandler pain funnel, each tied to a specific fact from the research.
What should it check first?
Your own CRM and email history, so you never ask a prospect something you already know.