It’s 4:45 on a Thursday. A rep has a call at 5 with a buyer who just said, “Your competitor told us they do this for less.” The PMM is in a planning meeting. The battle card is somewhere in a shared drive, last updated in the spring.

So the rep does what everyone does now. They open ChatGPT and ask.

This is happening on every sales team, whether or not anyone has approved it. The question isn’t whether your reps use AI to prepare for deals. It’s what the AI is telling them.

What they’re probably hearing

A general-purpose AI tool answers from the open internet. For a question about your product or your competitors, that usually means some mix of:

• Your competitor’s marketing. Their website, their comparison pages and their press releases are public. Your internal positioning isn’t.

• Old information. Pricing pages, packaging and feature lists change. What the model learned may be a year out of date.

• Confident guesses. When the AI doesn’t know, it rarely says so. It fills the gap with something that sounds right.

• Generic advice. Ask how to handle a price objection and you’ll get a sensible answer that could apply to any company. It won’t mention the proof point your team spent a month getting approved.

None of this is the rep’s fault. They have a deal to win and a question that needs an answer now. They used the fastest tool they had.

Why banning it doesn’t work

The first instinct is to tell reps to stop. That rarely lasts. The tool is too useful and too close at hand, and the alternative, waiting for the PMM to reply, is too slow when a buyer is waiting.

The real choice isn’t between AI and no AI. It’s between AI that answers from the internet and AI that answers from what your team has approved.

What a good answer looks like

When a rep asks a question mid-deal, a useful answer has four things:

1. It comes from your content. Your positioning, your pricing, your customer stories, not a competitor’s website.

2. It shows its source. The rep can see where the answer came from and how current it is, so they know it’s safe to repeat to a buyer.

3. It admits what it doesn’t know. “I’m not sure, and here’s who to ask” is far better than a confident wrong answer on a live call.

4. Someone can see the questions. If ten reps ask the same thing this week and nobody has a good answer, the PMM should know.

That last point is the one most teams miss. Every question a rep asks AI is a signal about what they need. Right now those signals disappear into a chat window nobody else can see.

Where to start

You don’t need a new tool to begin. Try this:

• Ask your reps. Find out which questions they take to AI and which tools they use. You’ll learn a lot in one team meeting.

• Ask AI yourself. Put your top five rep questions into ChatGPT and read the answers as a buyer would hear them. Note what’s wrong, outdated or missing.

• Fix your most-used content first. The questions reps ask most often tell you which battle cards and FAQs matter. Start there.

We built Marnie because we kept seeing this gap. Reps ask Marnie in Slack and get answers from the content their team has approved, with sources, and the PMM sees what reps are asking. But whatever you use, the goal is the same: when your reps ask AI, make sure it gives them the right answer.