Battlecard Prompts Your Reps Can Actually Use

Most competitive work is read in silence by people on your own side. A battlecard is not. It gets squeezed into a sentence and said to a buyer. That buyer is very often the one person in the room most likely to know whether it is true, because they have the competitor’s page open and a salesperson from that company in their calendar.

That changes what an error costs. A wrong figure in an internal analysis gets corrected in the next version. A wrong figure in a battlecard gets corrected by the buyer, out loud. And it does not just cost you that claim. It costs you every other claim on the card, because the buyer has just learned something about how carefully your side checks things.

Which is why the drafting is the easy half

Models are genuinely good at the shape of a battlecard. Splitting what you know into sections, cutting it to one screen, and phrasing an objection response in words a person can actually say. That is hours of work, and it is worth handing over. What cannot be handed over is deciding which claims stand on something. The prompts below make the model separate those two rather than blend them.

The objection section is the one place on a card where that split has a second step. Before a response is worth phrasing well, the objection needs a verdict: true, partly true, or false. That is what objection-handling prompts produce first. A model asked for responses will skip straight past that question to three good-sounding answers.

The first drafts the card. The second and third fix the two sections that most often fail in a live deal. The fourth points at a card you already have, and it is the one worth running before you trust anything you inherited.

A battlecard makes three kinds of claim. What the competitor says. What happens when you meet them in a deal. And what you offer instead. Each needs its own evidence, and the card fails in a different way for each one that is missing.

The positioning section is the one most people trust, because it looks like copying rather than analysis. We asked for it twice: once with nothing supplied, once with the competitor’s product page saved that morning.

An invented quote is a different failure from an invented fact

Anyone can check a wrong price in ten seconds, and somebody usually does. A sentence in quotation marks is not read as a claim at all. It is read as evidence, and evidence does not get fact-checked in an internal review, because it looks like somebody already did. That is what makes this the longest-lived error a battlecard can carry.

It also explains why the invented version read better. It was written to sound like positioning, so it came out smoother and more quotable than the real headline, which was three words long. A section that reads well is not evidence that the material behind it was there.

Three passes, and none of them is a rewrite. Take the claims tagged as unsourced and either find a source or delete the line. Resist the urge to soften a line into something vaguer that survives. Check every quoted phrase against the live page, word for word, because those are the lines a buyer can correct. Then read the objection responses out loud. That is the only reliable way to find the ones nobody can say.

The last step is to record when each section was last checked, beside the section rather than at the top of the card. Pricing and packaging claims go stale within a quarter. Positioning holds for a year. One date on the whole card either makes the long-lived parts look suspect, or makes the fast-moving parts look current.

Then put the draft where the team already looks. A sales battlecard template matters less for its layout than for the fact that everyone opens the same one. That is what stops six reps carrying six slightly different versions of a claim. The price line is the one that drifts first, and a competitor pricing comparison is where it gets a capture date attached.

You now have a card where every claim is either sourced or gone. It is accurate today. The problem with this particular document is that it looks no different on the day one of those claims stops being true. And it is read by people who were not there when it was written, and have no way to tell.

The claims decay at different speeds, which is what makes a review schedule hard to set by calendar. A price or a packaging change can kill two lines within a week of a page being edited. A feature claim goes stale quietly when the competitor ships something. Positioning holds for months and then changes all at once, usually alongside a launch. At that point the tone of the whole card is wrong, not one line of it.

The cost of being late is easy to name. A rep in a live deal, quoting a card, to a buyer who has just been told something different by the competitor’s own salesperson. Nothing about the card looked out of date beforehand. That is also what changes when a card stops being opened and starts being read out of a custom GPT: the date on the front of the card no longer reaches anybody.

Noticing those changes as they happen is monitoring rather than drafting, and it is the part a prompt simply cannot do. A model has no memory of the page it read last month. Flares watches competitor pricing, product and positioning pages and turns a change into a signal, so the sections of a card that moved are the ones that come up for review.

What that does not do is write the response. Deciding what a competitor’s new claim means for your positioning, and what a rep should say when it comes up, takes judgement built on knowing your own product. These prompts exist to speed that work up, not to replace it.