Tear Down Competitors With Verified Claims

Workflow area: Ideation & Critique · Tier: Core · Hands-on time: ~40 minutes

The deep-research modes in Claude, Perplexity and Gemini will happily assemble a competitor analysis in minutes — and that analysis will confidently describe features that shipped, features that were announced and cancelled, and features that exist only in a 2023 blog post. Competitor products change monthly; training data and indexed articles lag by quarters. So the workflow is a pincer: AI research for breadth and market context, your own screenshots and walkthroughs for ground truth, and a claims table that refuses to let the two blur together.

When to use — and when not to

• Use it entering a redesign or new-market pushyou need the landscape mapped this month, not as you remember it from last year's teardown.
• Use it when stakeholders cite competitors at you"Competitor X does this" deserves a verified claims table, because half the time Competitor X does not, or does it badly.
• Use it on a cadence for your top three competitorsquarterly refreshes are cheap once the structure exists, and drift between refreshes is itself a finding.
• Skip it for strategy you should owna teardown tells you where others are, not where you should go; copying verified features is still copying.
• Skip the deep-research pass for niche B2B tools with no public footprintif the product hides behind a sales demo, AI research will hallucinate the gap; go straight to trials and demos.

The recipe

1. 1.

   Define the comparison spine first. Pick the five to eight dimensions that matter for your decision — onboarding flow, pricing model, collaboration features, AI capabilities, accessibility posture, platform coverage. A teardown without a spine becomes a feature-list swamp.

2. 2.

   Run the deep-research pass for breadth. In Perplexity, Gemini Deep Research, or Claude with web search:

   Copy

   Research [COMPETITOR A, B, C] for a competitive UX teardown.
   For each, along these dimensions: [YOUR SPINE]
   
   Rules:
   - Every claim must carry a source URL and the date of the source.
   - Distinguish SHIPPED (in product today, per source) from ANNOUNCED
     (marketing/roadmap) from REPORTED (third-party claims). Label each.
   - Flag any claim where sources conflict or are older than 12 months
     as LOW CONFIDENCE.
   - Include pricing pages, changelogs, release notes and recent
     reviews as preferred sources over blog commentary.
   - Say "no reliable source found" rather than inferring.
   

3. 3.

   Do your own multimodal walkthrough for ground truth. Sign up for trials of the top two competitors. Screenshot the flows on your spine — onboarding, the core task, settings, empty states. Then run each set through Claude or Gemini: "Here are screenshots of Competitor A's onboarding, captured today. Describe the flow structure, required steps, and moments of friction. What patterns are notable versus common practice?" Your capture date becomes the freshness stamp no article can beat.

4. 4.

   Merge into the claims table — the deliverable. One row per claim:

   Copy

   DIM        CLAIM                       COMP  STATUS    SOURCE         DATE     VERIFIED
   Onboarding No signup required for      A     SHIPPED   own trial      2026-08  HAND ✓
              first project
   AI         "AI layout suggestions"     B     ANNOUNCED changelog URL  2026-05  —
   Pricing    Free tier capped at 3       A     SHIPPED   pricing URL    2026-08  HAND ✓
              projects
   Collab     Real-time multiplayer       C     REPORTED  review URL     2025-11  LOW CONF
   

5. 5.

   Hand-verify the claims your decisions will lean on. Rank claims by decision weight and verify the top ones yourself — open the pricing page, run the flow, click the feature. Anything you could not verify keeps its LOW CONFIDENCE label in every downstream deck. Then write the one-page synthesis: where competitors converge (table stakes), where they diverge (open territory), and the two or three verified gaps your design could own.

Make it yours

• Perplexity for the fast quarterly refresh: re-run the same spine prompt quarterly and diff against last quarter's table — the diff, not the table, is what you present.
• Gemini for video teardowns: competitor demo videos and YouTube reviews are walkthroughs someone else recorded; have Gemini summarize the flows shown, then verify key moments at the timestamps yourself.
• Claude Project as the living teardown: keep the claims table, screenshots and synthesis per competitor in one Project; when anyone asks "does X do this?", the answer comes with a source and a date or it does not come at all.

Quality check

• Every claim has a URL and a datean unsourced claim in a teardown is an opinion in a costume.
• Shipped, announced and reported never blurthe most expensive teardown errors are announced features graded as shipped reality.
• Freshness is visible per claim, not per documenta table "updated August 2026" containing 2024 sources is laundering staleness.
• Decision-bearing claims got hand verificationif the redesign bets on a competitor gap, you personally confirmed the gap exists this month.
• The synthesis names territory, not just featuresa teardown that ends at "they have X, we don't" produced an inventory, not an insight.