Twelve copy-paste Claude Cowork prompts for the work that used to mean "open 12 tabs and read for 4 hours" — company research, teardowns, market sizing, lit review.
TL;DR. Twelve copy-paste prompts for the work that used to mean "open 12 tabs and read for 4 hours" — web research, document corpus synthesis, comparison, evidence collection. The "what would change the verdict" line in prompt #12 is a quiet superpower; it forces Cowork to admit what it doesn't know.
Research [company name] on the public web.
Produce /output/[company]-background.docx covering:
- Company shape (HQ, size, ownership, industry)
- Products/services (one paragraph)
- Recent moves (last 90 days)
- Known competitors (top 3, one line each)
Cite every claim. Mark inferred claims as such.
Length: 1 page.
Research [person name, role, company] from public sources.
Produce /output/[name]-profile.docx covering:
- Current role and tenure
- Career history (last 5 roles)
- Public commentary or talks (top 3 with links)
- Known interests or focus areas
Cite every claim. Avoid inferring personal details not publicly disclosed.
Research [competitor name].
Read everything in /inbox/competitor-collateral/[name]/.
Produce /output/teardown-[competitor].docx:
- Positioning (1 paragraph)
- Pricing (table — public vs inferred clearly marked)
- Last 90 days of moves
- Likely sales objections from prospects comparing to us
- Three places they win, three places we win
Cite every claim.
Search the web for news on [topic] in the last [N] days.
Produce /output/news-roundup-[date].md grouped by theme.
Each item: one-line summary + source link + date.
Skip news older than [N] days. No marketing or vendor blog posts.
Research [market segment] size and growth.
Produce /output/market-sizing-[segment].docx with:
- TAM / SAM / SOM estimates from at least 3 independent sources
- Methodology per estimate
- A 1-paragraph reconciliation of differences
- Confidence rating per estimate
Cite every source with publication date.
Read /inbox/corpus-[topic]/.
Answer this question: [question].
Output to /output/answer.md:
- Direct answer (3 sentences)
- Evidence (per source, with file + page)
- Counter-evidence or alternatives if the corpus contains them
- Gaps the corpus doesn't address
Read /inbox/papers/.
Produce /output/lit-review.docx structured as:
- Abstract (200 words)
- Themes (3–5)
- Per theme: a paragraph of synthesis, plus 3–5 cited supporting sources
- Open questions / where the field disagrees
House style: academic, hedged, no claims beyond what the corpus supports.
Research vendors: [vendor list].
Produce /output/vendor-comparison.xlsx with:
- Row per vendor, columns: positioning, pricing tier, key features, integrations, recent funding
- Confidence per row
- Source link per row
And /output/recommendation.md picking the best fit for /CLAUDE.md/use-case-spec.
Search for regulatory or policy changes affecting [industry / function] in [jurisdiction] in the last 12 months.
Produce /output/regulatory-scan.docx with:
- Per change: name, summary, effective date, applicability, source
- Confidence rating per item
- Items where Cowork is uncertain flagged for legal review
Conservative bias: when in doubt, flag.
Read /inbox/customer-feedback/ (CSVs, PDFs, transcripts).
Produce /output/feedback-synthesis-[period].docx:
- Top 5 themes by frequency
- Per theme: representative verbatim quote (with source), prevalence estimate, severity
- Top 3 candidate product changes
Verbatim quotes never paraphrased.
Read /inbox/conference-notes/ (slides, transcripts, notes).
Produce /output/event-recap.docx:
- Top 3 takeaways (1 paragraph each)
- 5 named people / orgs to follow up with, and why
- 3 ideas worth piloting at our company
- Closing one-line: was it worth attending?
Honest answer; no marketing.
Read /inbox/data-corpus/.
Hypothesis: [hypothesis].
Output to /output/hypothesis-test.md:
- Evidence supporting the hypothesis (with cites)
- Evidence against the hypothesis (with cites)
- Verdict: supported / mixed / contradicted, with confidence
- What additional evidence would change the verdict
The "what would change the verdict" line in prompt #12 is a quiet superpower. It forces Cowork to admit what it doesn't know — and gives the operator a roadmap for the next research session. We add similar what-would-change-this lines to research prompts in client engagements; they shift the output from "an answer" to "an answer plus the next question."
Book a 30-minute call. We'll ask where you are, what your team needs, and which systems Cowork should touch.