Competitor Battlecards for a Small B2B Team: AI-Drafted, Rep-Checked

A small B2B team can build a competitor battlecard by collecting six to ten public pages per rival (pricing, product, reviews, job adverts, announcements), asking AI to draft each section with a source and a confidence label on every claim, then having one rep who has met the competitor verify the card in a 20-minute review before anyone uses it.

The weak point is never the drafting. AI will write a confident card in two minutes, including a statement that the rival "lacks single sign-on" which stopped being true last quarter. The process below is built around that: evidence in, labelled claims out, a human check, a dated update schedule, and a way for the people on calls to tell you the card is wrong.

Follow me on Instagram@sagnikteaches

What goes on a card a rep will open mid-call

Reps do not read battlecards; they scan them while a buyer is talking. A card that takes more than a minute or two to use tends to get ignored, so the design target is one page. The sections that earn their space, in the order a rep reaches for them:

Connect on LinkedInSagnik Bhattacharya

1. One-line positioning. Who the competitor sells to and how they describe themselves, in their words.
2. Where they win. Honest strengths. A card that denies any strength loses credibility the first time a buyer mentions one.
3. Where we win. Three to five points, each with evidence a buyer can check.
4. Landmine questions. Questions a rep can ask the buyer that make our strengths matter, without naming the rival's weakness.
5. Objections and answers. The five things buyers actually say when this rival is in the deal, with a reply written in a human voice.
6. Pricing as publicly shown. What is on the page, the date it was captured, and what is not known.
7. Proof points. A customer-approved quote or a figure from your own records, never one lifted from a rival's case study.
8. Do not say. Claims that sound good and cannot be backed up.
9. Card owner, last verified date, next review date. Three lines at the top that make the card trustworthy.

Pricing, strengths and "do not say" are the sections that most often get skipped by AI-only cards, and the ones reps value most. Everything else in this tutorial is about producing that page with evidence behind each line.

Subscribe on YouTube@sagnikteaches

The evidence pack: where public research actually comes from

AI drafts only as well as what you put in front of it. If you ask a chat assistant for "a battlecard on Competitor X", it uses whatever it remembers or finds, with no record of where a claim came from. Instead, build an evidence pack: a folder per competitor with saved pages, so every statement has a source you can open. A good pack holds 6 to 10 items and takes about 45 minutes to assemble.

Two rules keep the pack clean. First, save the page as a PDF or paste the text into a document with the web address and the date captured on top; a link alone breaks when the page changes. Second, stay on the public side: read pages anyone can open, sign up with your own work email if a free trial is public, and never pose as a customer or ask a contact at the rival for information. Check a site's terms before pasting large amounts of its review text into a tool.

If you have not yet done a broad scan of who your rivals are, competitor research with AI in one afternoon comes first; this tutorial picks up once you know which two or three rivals deserve a card.

A first card, start to finish: a nine-person security installer

An illustration. A nine-person company installs access control and CCTV for offices and sells annual maintenance contracts, mostly to facilities managers. Three of its last five lost quotes went to the same larger rival, call it Rival A. The owner and the two sales reps spend a Thursday afternoon on the first card.

• 45 minutes: build the evidence pack. Saved: Rival A's pricing and packages page, its service-level page, its help articles on remote monitoring, three job adverts, two press items, and a dozen review snippets the owner read and copied by hand.
• 10 minutes: the reps add what they know from calls, as a bullet list of what buyers said ("they quoted a four-hour callout", "their quote was cheaper in year one, dearer after").
• 5 minutes: run the drafting prompt below.
• 25 minutes: the owner and one rep review the claim ledger, line by line.
• 15 minutes: rewrite the answers in the reps' voice, then drop the card into the shared drive and the CRM's notes template.

That is about 1 hour 40 minutes for the first card, and a second card for another rival takes closer to an hour because the prompt and layout are reused. Measure what you save by counting how long reps used to spend hunting before a call; if it was 15 minutes a deal and you have 12 live deals with this rival, that is 180 minutes, or 3 hours of hunting, against 1 hour 40 minutes to build the card, so the first card has paid back its build time well inside those 12 deals.

The drafting prompt: make the AI show its evidence

The prompt does three things: restricts the AI to the pack, forces a label on every claim, and separates what the sources say from what you would infer. Paste the saved pages after it, each headed with its address and capture date.

You are helping a small B2B sales team draft an internal competitor
battlecard. Use ONLY the source pages pasted below. Do not use anything
you remember about this company.

Competitor: Rival A. Our company: [one sentence on what we sell and to whom].

Produce a table called CLAIM LEDGER with these columns:
claim | source (page title and date captured) | exact quote or figure |
label (CONFIRMED = stated on the page, INFERRED = my reasoning from
the page, UNKNOWN = not in the sources) | confidence (high/medium/low).

Then draft the battlecard sections: positioning, where they win, where we
win, five landmine questions, five likely objections with replies, public
pricing, do-not-say list. Every sentence in those sections must trace to
a row in the ledger. If the sources do not support a section, write
"No evidence in pack" rather than filling it in.

Never speculate about their finances, customers leaving or staff. Do not
use superlatives. Write replies in plain spoken English a rep could say
aloud.

An illustrative extract of what comes back:

CLAIM LEDGER (extract)
1. Callout within 4 hours on the Premium plan | Service page, captured 2 Nov |
   "4-hour response on Premium contracts" | CONFIRMED | high
2. Remote monitoring costs extra on the Standard plan | Packages page,
   2 Nov | "Remote monitoring: add-on" | CONFIRMED | high
3. Install pricing is "custom quote" | Packages page | "Contact us for
   installation pricing" | CONFIRMED | high
4. They are expanding monitoring staff | Three job adverts for monitoring
   operators | titles only | INFERRED | low
5. They integrate with our cameras | no page found | UNKNOWN

The ledger is the real product. Row 1 and 2 go onto the card as facts; row 4 goes in only as a question for the rep to listen for; row 5 becomes an item for the owner to check before the card says anything about cameras. A drafted card is only as honest as the least-supported row, so review the ledger before you read the polished sections.

What the rep check looks like and how long it takes

The point of "rep-checked" is not proofreading. It is catching three kinds of error AI cannot catch from public pages: claims that are true on paper but wrong in practice, claims that are out of date, and replies that no rep would say aloud. A workable 25-minute review with one person who has been in deals against this rival:

1. Open the ledger first. Strike any CONFIRMED row whose source you cannot open in two clicks. Change any INFERRED row you disagree with to UNKNOWN.
2. Read "where they win" aloud. If it feels too generous, ask whether a buyer has actually said it to you. If yes, keep it. A card is more believable when it admits what the rival does well.
3. Test every objection reply by saying it. Anything that sounds like a brochure gets rewritten in the rep's words.
4. Mark each proof point. Customer quotes need the customer's written approval; internal figures need a date and where they came from.
5. Sign it. The reviewer's initials and the date go at the top. A card with no reviewer is a draft and says so.

This is the same pattern as any AI output a customer might see: a named person looks at it before it goes out. The wider approach to doing that without slowing the work is in setting up human review for AI work.

Three before and afters from a real-looking review

These are illustrative, but they show what review changes.

A brochure reply becomes a spoken one. AI draft: "Our comprehensive maintenance programme delivers superior ongoing value versus Rival A's tiered structure." Rep version: "On their cheapest plan, remote monitoring is an add-on, so ask them for the three-year price with it included. We include it from the first year, so you can compare like with like."

A guess becomes a question. AI draft: "Rival A is struggling to hire engineers." Rep version, in the landmine section: "How quickly can they put an engineer on site for a Saturday fault, and is that in writing?" The first is unsupported; the second does the same work without claiming anything.

A stale feature claim gets removed. AI draft: "Rival A does not offer a mobile app." The rep had seen one on a buyer's phone the week before. Row reclassified UNKNOWN, then checked against the help centre and turned into "Their mobile app arrived this year; ask the buyer whether they have used it."

The week a card goes wrong: a stale claim in front of a buyer

A mistake worth planning for, using an illustrative team. A rep tells a facilities manager that a rival has no out-of-hours phone line, because the card says so. The buyer opens the rival's contact page and reads out the number. The deal does not die on the spot, but trust drops and the rep spends ten minutes recovering.

The cause was predictable: the line came from a three-month-old capture of a page, marked CONFIRMED, never rechecked. Two fixes follow. Put the capture date beside every pricing or "does not offer" claim on the card, so reps see how old it is. And treat negative claims ("they do not offer X") as the highest-risk type: require a second source, such as the help centre, before one goes on the card at all. Absence is the hardest thing to confirm from a marketing page.

Landmine questions and objection replies, with examples

Landmine questions are what make a card persuasive without attacking anyone. A good one is neutral on its face, true whichever vendor the buyer chooses, and shaped so that your strength matters. Three illustrative ones for the installer's card:

• "If a door controller fails on a Friday evening, what happens, and who do you speak to?" Matters because the installer has its own engineers on call.
• "Is remote monitoring included in the price you are comparing, or on top?" Surfaces the add-on gap found in the ledger.
• "After the first year, what happens to the maintenance price, and where is that written?" Matters when the rival's year-one quote is lower and its renewal is not.

For objections, give reps the buyer's phrase and a three-part reply: acknowledge, a short fact, a question back. One example: Buyer: "Rival A is cheaper." Rep: "On the first-year quote they often are. Can I show you the three-year total with monitoring and out-of-hours callout included on both? Then you can compare the same thing." That reply asserts nothing you cannot prove and moves the conversation to a number the buyer can check.

Review sites are an excellent source for the buyer's own words. Mining competitor reviews for gaps and positioning with AI shows how to read them in volume; here, a handful of recurring complaints (slow callouts, price rises at renewal) are enough to write two or three landmine questions.

Pricing on the card: what is shown, what is not

Pricing is where reps most want certainty and where public pages give the least. Write what the page shows, the date, and what you do not know. A filled-in example for the card:

PUBLIC PRICING (captured 2 Nov, source: packages page)
Standard: shown as "from" price, remote monitoring add-on
Premium: "contact us", 4-hour callout
Installation: custom quote
NOT PUBLIC: renewal uplift, discounts, contract length
Rep note: ask buyer for their written quote; do not guess a number.

If you want a recurring watch on rival prices, tracking competitors' prices and offers with AI covers how to capture them on a schedule so the date on your card is never old. The rule for reps is simple: quote only what the buyer shows you, or what is on the page, and say which.

An update schedule a five-person team will keep

Cards decay. The schedule has to be small enough that it happens, so it has two layers: a short monthly check and a quarterly rewrite, plus triggers.

• Monthly, 20 minutes, one owner. Open the rival's pricing page, changelog or product news, careers page and latest reviews. Compare with the card's ledger. Paste both versions into AI and ask: "List every difference between version 1 and version 2 of this page, and which card claims they affect." Update only those rows, and change the "last verified" date.
• Quarterly, 90 minutes. Rebuild the pack, rerun the prompt, and compare the new ledger with the old. This is also when you retire a card for a rival that stopped appearing in deals.
• Trigger reviews. A funding round, a major launch, a price change you spot, or two reps reporting the same new objection in two weeks. Review that card within 5 working days.

Give each card one named owner, not "the team". Put the next review date in the owner's calendar, because a date on a page nobody opens does nothing. The monthly diff step is one of the most useful AI jobs in this whole process, since comparing two long pages by eye is slow and easy to get wrong.

Closing the loop: how sales feedback improves the card

The battlecard only improves if reps can tell you when it fails, and that has to be easier than ignoring it. Three mechanisms, in rising order of effort:

1. A one-line flag. A shared channel or a CRM field called "card was wrong or missing". A rep types one line: "Rival A now includes monitoring in Standard, per buyer's quote". Cost to the rep: 15 seconds.
2. A win-loss field on every closed deal. Choose from a short list of reasons: price, callout time, integration, existing relationship, other. Add "rival" and "card used (yes/no)". After a quarter, the counts tell you which cards matter and whether using one changes outcomes. Keep the fields few; five clicks a deal is the ceiling.
3. A 15-minute review every quarter. Reps bring two deals against each rival. Ask only: what did the buyer say that the card did not cover, and what on the card did you never use? Cut what was never used.

Feed the flags back into the next monthly check. An example flow: three flags in six weeks say buyers keep asking whether Rival A's contract auto-renews. That goes into the research prompt as a new question, the pack gains the contract terms page, and the card gets a new objection row. Compare that with a card refreshed on a calendar alone: it answers last quarter's questions.

Guardrails: what the card should never contain

A short list the reviewer checks every time:

• No claim you cannot trace to a public page or to a buyer's own words.
• No guesses about a rival's finances, staffing or customers, and no rumours.
• No private material: a rival's internal documents, a former employee's inside knowledge, or anything obtained by pretending to be someone else.
• No customer names or quotes without written approval.
• No comparative claim in marketing copy until someone with legal responsibility has reviewed it. The card is internal, but reps repeat it.

Keep the card, the evidence pack and the ledger together in a folder with limited access, because the ledger sometimes holds notes from live deals. If you want to use a larger research tool to gather the first pass, a repeatable competitor and pricing brief in Perplexity is one way to source pages, and the output goes through the same ledger and review.

Checking that the cards are working after a quarter

Judge the process on four plain measures, not on whether the cards look good:

1. Use. Ask each rep for the last deal in which they opened a card. If fewer than half can name one, the cards are too long or in the wrong place; move them into the CRM record for the deal.
2. Accuracy. Count the "card was wrong" flags. Zero flags in three months usually means nobody is looking, not that the card is perfect.
3. Age. Every card shows a verified date within the last 35 days. Any card past 60 days gets a banner "stale" until reviewed.
4. Outcomes. Compare win rate in deals against each rival before and after the card, with the honest caveat that a dozen deals is too few to prove anything. Look for direction and for the reasons reps give.

If two measures look weak, shrink the project, not grow it. One good card for the rival you meet most is worth more than five stale ones, and the monthly diff is cheap enough to keep. If you would rather set this up with someone and decide which rivals deserve a card first, that is what my AI implementation consultation covers.

Battlecard questions from small sales teams

How many battlecards does a small team need?

Start with the two or three competitors that appear in your lost-deal notes most often. A team of five reps cannot keep eight cards current, and an out-of-date card does more harm than none. Add a fourth only when a rep reports meeting that rival in two live deals within a quarter.

Can AI write a battlecard from nothing?

It can write one, but it will fill gaps with plausible guesses, which is the worst outcome for a document reps quote to buyers. Give it source pages you collected, ask it to label every claim with a source and a confidence level, and let reps remove or fix anything unlabelled.

Is it legal to compare us with a competitor on a battlecard?

A battlecard for internal use differs from an advert, but reps repeat its lines to buyers. Keep comparisons factual, dated and traceable to a public source, and have a lawyer review any claim you put in marketing copy. This is a practical guide, not legal advice.

What should a rep never say about a competitor?

Anything they cannot show on a public page or a customer's own words. Avoid guesses about finances, staff or customers leaving, and avoid mocking the rival. Buyers trust a rep who says the competitor is good at one thing and explains where you fit better.

How often should a battlecard be reviewed?

Check a short trigger list every month (pricing page, product updates, job adverts, reviews) and rewrite fully every quarter. Also review at once after a competitor's funding or launch announcement, or when two reps report the same new objection in a fortnight.

Further reads

• How to Use AI in a Small Business: Perplexity for Research Jobs — Where a search-first tool fits among your research jobs.
• Best AI for Startup Research: Customers, Rivals and Pricing — Research customers and rivals before you write any card.
• AI Use Case Template: Score Every Idea on One Page — Score whether battlecards beat your next AI project.
• How to Write a One-Page AI Strategy for Your Business — Place battlecards inside a wider plan for AI at work.
• Best AI Tools for Product Marketing at a Small Company: Launch Kit — Write positioning, a launch email and a one-pager from one product brief, then check every claim against customer evidence before it goes out.
• ChatGPT for Sales and Marketing: B2B Battle Cards and One-Pagers — Prompts and a fact-check routine for a competitor battle card and a one-page service sheet in a small B2B firm.
• AI Personalisation in B2B: Tailor Outreach to Each Trade Account — Personalise B2B trade outreach and quotes account by account using structured notes sheets and human-reviewed AI prompts to increase reply rates.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general sales-enablement practice; competitor pages are the examples a reader would collect themselves, so no vendor facts are claimed here.