Market-level teardowns won't help you win an account. The four prompts that produce intel a deal team can actually use.
The competitive teardown most marketers run is a market-level exercise. Five competitors, a positioning matrix, a white-space hypothesis. It's useful if you're a CMO setting a category bet. It's almost useless if you're trying to win a specific B2B account.
B2B competitive intelligence is a different job. The buying committee is five to twelve people. The decision cycle is months. You're not competing on positioning, you're competing on which champion at the prospect's company is willing to fight for which vendor. The intel you need is account-shaped, not market-shaped.
Most marketers run the market-shaped prompt on a B2B deal, get a useful-looking deck, and walk into a deal review unable to answer the question that actually matters: who at the prospect is leaning toward whom, and why.
A generic competitive teardown on a specific B2B deal produces a page of inferences dressed up as findings. The output looks authoritative. It will not survive contact with the deal team, because nobody at the prospect's actual company shows up in it.
The teardown answered the wrong question, sharply. Sharp answers to wrong questions are the second-most expensive mistake in sales enablement. Confident-sounding wrong answers are the first.
Two questions, in this order. The order matters.
The first tells you the buying patterns the competitor's pitch works on. The second tells you the patterns it does not. Together they tell you which accounts in your pipeline are "competitor's natural fit" versus "yours to lose." That is a sales conversation, not a marketing deck.
You cannot answer either question from market-level inputs. You need: public customer logos, case studies, G2 and Capterra reviews (especially the three-star and below ones), LinkedIn signal from accounts that switched away, and ex-customer conversations if you can get them. Real source material is everything in B2B intel because the inference temptation with thin B2B data is exactly the failure mode that gets posts written about how "AI doesn't work for B2B."
Inputs: the competitor's public customer list, their case studies, the testimonials they feature on their site.
Job: extract the pattern. Industry concentrations, company sizes, named use cases, which buying personas show up as the quoted champion (CMO, RevOps, IT, etc.), which integration partners co-marketed.
Refuse-condition: do not infer a pattern from fewer than three case studies in the same dimension. If the data is thin, say so explicitly.
Output spec: a one-page pattern profile with confidence ratings (high if five or more instances, medium if three to four, low if two or fewer, flag if one).
What you get: the prospect lookalike profile. If your current deal looks like this profile, the competitor's pitch will land. You need to be specifically different on something the prospect cares about, or your deal is structurally uphill.
Inputs: G2 and Capterra reviews under four stars, "alternatives to X" content from third-party sites, LinkedIn signal from your own won deals where the prospect previously used the competitor.
Job: extract the loss reasons by category. Pricing, integration, support, specific feature gaps, vendor concentration concerns.
Refuse-condition: do not declare a loss reason from a single review. Cross-reference at least two independent sources before naming it as a pattern.
Output spec: a ranked list of loss reasons with example evidence inline for each.
What you get: the objection map. When you walk into a deal where the competitor is incumbent, this is the list of dissatisfactions to surface in your discovery. Not as attack lines. As genuine questions.
Inputs: the LinkedIn profiles of the named champions in the competitor's case studies, plus any public talks or interviews they have given.
Job: infer what the competitor's pitch actually said to them. What problem framing did it use? What metric did it promise to move? What internal political battle was the champion likely fighting when they bought?
Refuse-condition: speculation must be marked [INFERRED] and grounded in the public source it came from. Pure speculation, even confident speculation, gets refused.
Output spec: a profile of the competitor's natural-fit champion: title, seniority, problem framing, metric, internal politics.
What you get: a clear picture of who in your target account the competitor will already be talking to. You either out-pitch them with the same champion, or you find a different champion the competitor's framing leaves cold.
Inputs: outputs from prompts 1, 2, 3 above, plus the specific prospect account.
Job: synthesise into a one-page deal-review brief. Here is what the competitor will be selling them. Here is the champion they will be selling to. Here are the three objections we can credibly raise. Here is the one risk to our deal that comes out of this analysis.
Skip-condition: if the inputs from prompts 1 to 3 do not support a defensible synthesis, say so and recommend deferring the deal review until better intel exists. Better to delay the brief than ship one built on hunches.
What you get: a brief the deal team can actually use in the next pipeline call.
The market-shaped teardown asks Claude to infer positioning patterns from polished public material that vendors specifically wrote to look good. The output is roughly as honest as the input, which means: useful for understanding market narrative, useless for understanding the actual deal.
The four prompts above force Claude to work from messier, more honest data. Third-party reviews, where customers complain. LinkedIn signal, where switches show up. The specific accounts your competitor named, with the specific objections that came up. The output is grounded in what the competitor's customers actually experienced, not what the competitor's marketing department wants them to have experienced.
That is the lever. B2B intel works when the source material is real enough to disagree with the vendor's own narrative. It fails when you ask Claude to refine the vendor's narrative back at you.
This is the prompt set where Claude is most useful and most dangerous. The data is always thin. The inference temptation is huge. Without refuse-conditions and explicit confidence ratings, you will ship a deal-review brief built on three reviews and a hunch.
The refuse-conditions are not optional decoration. Better to ship a one-line brief that says "the intel is too thin to draw conclusions, here is what we would need to know" than a polished page of confident bullshit that loses you the deal.
B2B intel that helps you win deals is intel that names the deal. Anything more general is a deck for someone else's quarter.
The free Competitive Landscape Teardown handles the market layer (positioning matrix, white-space hypotheses, kill criteria). This post is the account-level extension. The two work in sequence: market first to choose the bet, account-level when a specific deal needs the brief.
Get the free competitive teardown →