The Skills a marketer doing competitive intel should actually build. Four of them, what goes in each, a worked SKILL.md, and the one rule that separates real intel from confident bullshit.
Competitive research is the discipline where Claude is most likely to make you look stupid. Not because it is bad at it. Because it is fluent. Ask it to compare you against three competitors and it will hand back a clean, confident teardown in seconds, complete with positioning verdicts and pricing tiers, half of which it inferred from nothing. The output reads like intelligence. Some of it is just well-formatted guessing.
That is exactly why this is the discipline where Skills earn their keep. A Skill is not a smarter prompt. It is a standing instruction that loads automatically and enforces the same discipline every time, including the discipline to shut up when the data is thin. For competitive work that matters more than for anything else a marketer does. This post covers the four Skills worth building, what goes in each, and a refusal rule that should be in all of them.
A Skill is a folder with a SKILL.md file that Claude loads on its own whenever the description matches what you are doing. You write it once, it travels across every conversation and Project. If the format is new to you, the 20-minute build walkthrough covers the YAML and install, and the complete Skills guide covers the why.
Resist the urge to build one "Competitive Research" Skill that does everything. It will load on every conversation and produce mush. Build four narrow ones. Each describes its trigger (when it loads), what it knows (the standing rules), and what it refuses (the guardrails). The refuses are where the value is.
Trigger: any request to compare, analyse, or tear down competitors, landscapes, or rivals. Words like "compare," "teardown," "competitive landscape," "stack us against."
Knows: the exact structure every teardown should follow, so output is never freeform. Positioning, pricing, messaging, proof, channel, and the gap. Same six headers every time, so two teardowns done a month apart are actually comparable.
Refuses: to skip a section because data is missing. If pricing is not public, the Skill writes "pricing: not disclosed, requires sales contact" rather than inventing a number. A blank that says blank is intelligence. A filled blank that is a guess is sabotage.
Trigger: any competitive analysis where the source material is light. One screenshot, a homepage, a single press release.
Knows: the minimum evidence bar for each kind of claim. A positioning verdict needs the actual homepage copy. A pricing claim needs a pricing page or a quote. A "they are winning on X" claim needs more than vibes.
Refuses: to draw a conclusion the evidence does not support. This is the most important Skill on the list and it gets its own section below.
Trigger: requests that move from "what are they doing" to "what should we do about it." Words like "where is the gap," "how do we position against," "what is our wedge."
Knows: that a positioning move is a hypothesis, not a fact, and should be labelled as one. It frames every recommendation as "if X is true, then Y," names the assumption, and says what would prove it wrong.
Refuses: to present a positioning recommendation as a finished answer. It hands you a testable bet with its assumptions on the table, not a verdict you have to take on faith.
Trigger: B2B work where the unit of analysis is one named account, not a whole market. "What is Acme running," "how do we displace the incumbent at this account."
Knows: that account-level intel is a different job from market-level scanning. It works one named buyer at a time: their stack, their renewal timing, the incumbent, the internal champion. Granular, not panoramic.
Refuses: to generalise from one account to the whole segment, or to treat a single sales call note as a market trend. The B2B competitive intelligence post goes deep on this one.
Here is the teardown-structure Skill, written out in full. This is the actual file. Drop it in a folder, add your competitor notes as a reference file, install it. The shape is the same one from the build walkthrough: YAML up top, then knows, does, refuses.
--- name: competitive-teardown description: Use this Skill whenever the user is comparing, analysing, or tearing down competitors, rivals, or a competitive landscape. Triggers on "compare," "teardown," "competitive landscape," "stack us against," "how do we look next to." Enforces the six-section teardown structure and refuses to fill any section with inferred data. --- # Competitive Teardown ## What this Skill knows Every teardown uses the same six sections, in order: 1. Positioning: the one sentence they lead with 2. Pricing: tiers and numbers, or "not disclosed" 3. Messaging: the three claims they repeat most 4. Proof: what they cite (logos, numbers, case studies) 5. Channel: where they actually show up 6. The gap: what they do NOT say, claim, or serve Reference notes for tracked competitors are in competitors.md (this Skill's folder). ## What this Skill does - Fill all six sections for every competitor named - Mark the evidence source for each line in brackets (e.g. "[homepage]", "[pricing page]", "[inferred]") - Flag any line marked [inferred] for the user to verify - End with one "biggest open question" the data cannot answer ## What this Skill refuses - Do not invent pricing. If not public, write "not disclosed, requires sales contact." - Do not assign a positioning verdict from a logo or a tagline alone. Quote the actual copy or say you cannot. - Do not present [inferred] lines as fact. Label them. - If asked for a verdict the evidence does not support, say what is missing instead of guessing.
The bracketed evidence tags are the trick. Once Claude has to mark every line as [homepage] or [inferred], the guessing stops hiding. You can scan a teardown and see in two seconds which lines are load-bearing and which are Claude filling space. That single habit, baked into a Skill so you never have to ask for it, is worth more than any prompt-library trick.
Here is the one to get right. Competitive research is the place where the gap between confident output and confident bullshit is widest, and the hardest to spot, because both look identical on the page. A real finding and a confident hallucination wear the same font.
Left alone, Claude will tell you a competitor "is clearly targeting enterprise" on the strength of one screenshot of a homepage. It is not lying. It is pattern-matching, and the pattern is thin. The output sounds like a conclusion. It is a coin flip in a suit.
The fix is a refusal rule, in every competitive Skill you build:
Do not draw a conclusion the evidence cannot support. State what you know, mark what you inferred, and name what you would need to be sure. When data is insufficient, say "insufficient data to conclude X" and stop. A correct "I cannot tell from this" beats a confident wrong answer every time, because the confident wrong answer is the one that ends up in the deck.
This feels like it makes Claude less useful. It makes it dramatically more useful. An honest teardown with four solid findings and three flagged unknowns is something you can act on. A glossy teardown where you cannot tell the findings from the guesses is something you have to redo by hand, which means it saved you nothing. The whole point of competitive intel is to be the person in the room who knows what is real. A Skill that refuses to guess is what makes you that person.
One more thing this rule buys you: it scales across a team. A junior marketer running the teardown Skill gets the same refusal discipline a senior analyst would apply by instinct. The standard stops living in one person's head and starts living in the file. That is the actual reason to build these as Skills and not just prompts.
A Skill cannot do the research for you. It enforces structure and refusal discipline, but it works on what you feed it. If you paste in one homepage and nothing else, even a perfect Skill can only tell you it has one homepage to work with. The Skill is the standard. The inputs are still your job: pull the pricing pages, the recent posts, the review-site quotes, the job listings. Then let the Skill keep you honest about what they add up to.
The fastest way to see the structure in action is to run the prompt version first. The free competitive landscape teardown prompt produces exactly the six-section output the Skill enforces, with no email gate. There is a real worked teardown if you want to see the output before you run anything. Once you like the shape, wrap it in a Skill so you never paste it again.
The Competitive Teardown is one of the four free Claude Skills on the Hub, with the YAML, the six-section structure, and the refuse-thin-data rule already wired in. Voice Extractor, Competitive Teardown, SERP-Informed Brief, Performance Readout. CC BY 4.0. Drag into Claude Desktop or Claude Code. No email gate.
Get the 4 free Skills