PromptBro

Claude Prompt Generator — Build Expert Prompts with PromptBro

PromptBro builds structured prompts specifically optimised for Claude — using the XML tag structure, long-context handling, and literal instruction-following that makes Claude models perform at their best.

What Makes a Great Claude Prompt?

Claude responds exceptionally well to XML tag structure. Wrapping distinct parts of your prompt in semantic tags — <task>, <context>, <instructions>, <format> — dramatically improves Claude's ability to parse what you want and follow it precisely. Claude follows instructions more literally than most models, which is a strength: be specific about what you want, because Claude will do exactly what you ask rather than inferring intent. If you're ambiguous, Claude will ask for clarification rather than guess — which is helpful but slows you down if you haven't specified constraints upfront.

Claude handles extremely long documents and contexts better than any other model available today — feeding it 50,000 words of source material and asking for synthesis is a genuine use case where Claude excels. For complex reasoning tasks, ask Claude to think before answering using <thinking> tags — this gives the model space to reason through the problem before committing to an answer. Always include an explicit output format specification; Claude will respect it precisely. Multi-step reasoning, structured document analysis, and careful instruction-following are where Claude consistently outperforms alternatives.

Example Claude Prompts

Document analysis — XML structure

<task>
Analyse the following contract and identify all clauses that create financial liability for our company.
</task>

<context>
We are a Series A SaaS startup. We are reviewing a vendor contract before signing. Our legal counsel is unavailable and we need a preliminary risk assessment before Monday.
</context>

<instructions>
1. Read the full contract text below.
2. Identify every clause that could expose us to financial liability — indemnification, penalties, auto-renewal, payment terms, termination fees.
3. For each clause, note the section number, quote the relevant language, and rate the risk: Low / Medium / High.
4. Flag any clause that is non-standard or unusual for a SaaS vendor agreement.
5. Do not summarise the contract generally — focus only on financial exposure.
</instructions>

<format>
Return a structured list. Each item:
- Section: [number]
- Clause type: [type]
- Quoted language: "[quote]"
- Risk level: [Low/Medium/High]
- Why it matters: [1–2 sentences]

End with a "Top 3 concerns" block.
</format>

<contract>
[paste contract text here]
</contract>

Complex reasoning — explicit thinking step

<task>
Determine the optimal pricing strategy for a B2B SaaS product entering a market with two established competitors.
</task>

<context>
Our product: Project management tool for architecture firms (50–500 employees).
Competitor A: Broad horizontal tool, $25/user/month, dominant market share.
Competitor B: Vertical niche tool, $65/user/month, strong reviews but limited integrations.
Our differentiator: Native BIM file integration and compliance reporting built-in.
Target customer: IT decision-maker at mid-size architecture firm, budget-conscious.
</context>

<instructions>
Think through this carefully before writing your answer. Consider:
- What pricing signals quality in this market segment?
- How does our differentiation affect price sensitivity?
- What are the risks of pricing above or below each competitor?
- What pricing structure (per seat, flat, tiered) fits this buyer's procurement process?

After reasoning through these, give a concrete recommendation with a rationale.
</instructions>

<format>
<thinking>
[reason through the problem here before your answer]
</thinking>

Then write your recommendation as:
- Recommended price point: [specific number or range]
- Pricing structure: [structure with rationale]
- Positioning: [1 sentence on how to frame price vs. competitors]
- Key risk: [1 sentence on the biggest pricing risk to monitor]
</format>

Writing task — tone and format specified

<task>
Write a cold outreach email from our CEO to the Head of Engineering at a target company.
</task>

<context>
Our company: Depot — a dev infrastructure startup that speeds up CI/CD pipelines by 40–70%.
Target: Head of Engineering at a 200-person fintech company that recently posted 3 DevOps engineer job listings (signal: they're scaling and struggling with build times).
Sender: Our CEO, who previously led engineering at Stripe.
</context>

<instructions>
- Length: 120–150 words maximum. Not a word more.
- Tone: Peer-to-peer, direct. Not salesy. Assume both are engineers.
- Reference the job listings as the opening hook — show we did our research.
- One specific claim about what Depot does (40–70% faster builds), not a generic pitch.
- One clear, low-friction CTA: a 20-minute call, not a demo.
- No subject line in your output — just the email body.
- Do not use "I hope this email finds you well" or any similar opener.
</instructions>

<format>
Output the email body only. No commentary before or after.
</format>

Long-context synthesis — multi-document

<task>
Synthesise the attached research papers into a single coherent literature review section on the topic of transformer attention mechanisms and their computational scaling properties.
</task>

<context>
Audience: PhD-level ML researchers. Assume familiarity with transformers, attention, and standard complexity notation. This will be used as the related work section of a conference paper.
Word limit: 600–800 words.
</context>

<instructions>
1. Identify the 3–4 core themes across the papers (e.g. sparse attention, linear attention approximations, hardware-aware implementations).
2. Group papers by theme rather than summarising each sequentially.
3. Note where papers agree, where they contradict, and where the literature has open questions.
4. Use proper academic citation format: (Author et al., Year).
5. Do not introduce any claims not supported by the provided papers.
</instructions>

<format>
Write in standard academic prose. Use subheadings for each theme. End with a 2-sentence paragraph identifying the most significant open problem the reviewed literature leaves unresolved.
</format>

<papers>
[paste paper texts or abstracts here]
</papers>

Build your own Claude prompt with PromptBro

PromptBro's 6-step guided flow builds prompts optimised for Claude's specific strengths — XML tag structure, literal instruction-following, long-context analysis, and explicit format control. Just describe your goal, and PromptBro assembles the full prompt structure Claude responds best to.

Try PromptBro free — build your first prompt in 60 seconds →

Related generators