Build a battlecard from Gong calls with Claude, checked against competitor evidence

A battlecard from Gong calls starts from what buyers said out loud. Claude Academy's battle card use case builds from closed-lost deal notes in a CRM, which record what a rep wrote down. A transcript keeps the buyer's own words, the objection as it was raised and who raised it. That is the evidence a rep can quote back in the next call.

Gong's connector is built for that job. Its Help Center says it lets Claude ask questions about accounts and deals and retrieve insights, and the Agent Index lists ask_account, ask_deal, generate_brief and get_transcript. Use get_transcript to confirm any line before it goes on a card, because a summary can drift from what was said.

Gong shows how a competitor comes up in your calls. Its Help Center describes the Competition tracker, which detects competitive topics with AI, plus custom trackers, initiative boards and win/loss analytics, all measuring mentions and deal outcomes inside your own conversations. It has no view of what the competitor changed or what its customers say about it publicly.

That outside half is Linkeddit's. compete_signals_get returns the competitor's moves with a priority, category, source link and week, and its reviews view returns stored reviews from G2, Capterra, TrustRadius and Trustpilot, with site, rating, date and link. Both come with one plan at $60 a month. A buyer's objection that matches a recurring public complaint is a stronger line than either alone.

Only if you start it. Klue's post argues that a one-off Claude or ChatGPT prompt has no standing connection to your calls, cannot notice a new competitor mention, and needs someone to rerun it. For Claude that is accurate: it acts when you start it, so this workflow is a saved prompt you rerun, not a feed.

What helps is that Linkeddit's side runs on its own clock. On the test project the competitor signals carried the week of 2026-10-26, so a Monday rerun reads fresh data. If you need every closed deal to refresh the card for a large team without anyone starting anything, Klue describes a tool built for that. If a reviewed card each month is enough, a saved prompt does the job.

A call quote tells you one buyer said it. A pattern across low-rated public reviews tells you whether it is likely to come up again. In the test run, two of the four stored reviews were one-star and both complained that a dashboard or subscription was locked or changed after payment, which is the kind of theme a rep can probe for. Four reviews is a small sample, so Claude should call it a lead to test, not a finding.

Claude Academy's guide says to verify specific facts such as competitor pricing and launch dates before sharing, and to ask Claude for citations. The prompt on this page does that by forcing a link beside every claim. A product marketer's complaint on Reddit shows why the refresh matters: "I can't even refresh our collaterals once a month consistently." A dated card with a refresh date at least shows how old it is.

Leave out customer names and anything you could not confirm in a transcript. Gong transcripts name real buyers, and a battlecard is shared widely, so Claude should describe the buyer by role and keep the call reference for the person who needs to check it.

Also leave out claims that rest on a failed fetch. The run showed Trustpilot pages for Simple Analytics and Umami with status error, which means the site did not load, not that nobody reviewed them. A card that treats an error as silence can tell a rep that a competitor has no complaints. The prompt asks Claude to mark anything it could not confirm, so a gap appears as a gap.