Role-Based Prompt Checklist by Job Function

A working set of prompt patterns organized by job function, meant to be scanned for your own role rather than read start to finish.

How to Use This Checklist

Jump straight to the section matching your role; the patterns are grouped so you do not need the whole page.
Treat each item as a template to adapt, not a script to copy word for word.
Borrow patterns from other roles freely; many of these transfer well across job functions.
Revisit this list after a few weeks of real use and note which patterns you actually reach for, since that is a better guide than this list alone.

Writers

State the audience explicitly: "Draft this for [specific reader], who already knows [X] but not [Y]" instead of leaving the audience implied.
Name the point of view and tense up front: "Write this in first person, past tense" prevents Claude from guessing and getting it wrong mid-draft.
Ask for a scoped edit, not a rewrite: "Tighten paragraph three, leave the rest as-is" keeps a revision from drifting the whole piece.
Request options at decision points: "Give me three opening lines, ranked by how strong the hook is" works better than accepting the first line offered.
Set a hard length constraint: "Cut this to 300 words without losing the closing point" gives Claude something concrete to check against.
Ask for a voice check against a reference: "Does this match the tone of the attached sample chapter?" catches drift before it compounds.

Business Analysts

Ask for the headline first: "Lead with the single most important number, then the detail" keeps a summary from burying the point.
Separate fact from interpretation: "Mark anything that's a guess about cause, not just what the data shows" prevents speculation from reading as established fact.
Request a source check: "Which column or figure is this claim based on?" is a fast way to catch a misread number before it reaches a stakeholder.
Specify the comparison baseline: "Compare this week against the same week last month, not just last week" avoids an unintentionally misleading comparison.
Ask for the format the audience needs: "As a one-page summary with a table by region" instead of leaving structure to guesswork.
Flag anomalies explicitly: "Call out anything that looks unusual compared to the prior period" surfaces what a quick read might miss.

Marketers

Ask for distinct angles before variations: "Give me four different strategic angles first, then variations within each" avoids a batch of samey copy.
State banned language directly: "Never use [specific phrase or claim] anywhere in this" is more reliable than a general tone description alone.
Request channel-specific adaptation explicitly: "Adapt this for email vs. social, keep the core message identical" prevents an inconsistent core claim across channels.
Ask for a rationale alongside the copy: "Explain briefly why this angle should work for this audience" helps you evaluate options, not just read them.
Set a concrete constraint per channel: "Under 100 characters for the ad headline, under 280 for the social post" gives Claude something to check its own output against.
Request a scoped revision on a batch: "Keep angle two, cut angle four, sharpen the urgency in the rest" instead of regenerating everything each round.

Customer Support

Give the ticket's specifics up front: "Customer's issue: [specific problem]. Draft a reply." keeps the draft grounded in the actual complaint.
State the required structure: "Acknowledge the issue, explain briefly, give one clear next step" keeps replies consistent without re-describing the format every time.
Name hard constraints directly: "Never promise a refund or a specific fix date" belongs as an explicit rule, not an assumed norm.
Ask for a summarization grouping: "Group this batch of tickets by issue type, and flag which are new this week" turns a pile of tickets into something a team can act on.
Request an escalation flag: "Note if any of these tickets look serious enough to escalate" adds a decision-relevant layer to a summary.
Ask for tone matching a difficult situation: "Keep this warm but direct, the customer is already frustrated" sets expectations beyond a generic tone word.

Applying the Checklist in Order

Start with your own role's section (1-6): get comfortable with the patterns for your day-to-day work before borrowing from elsewhere.
Layer in cross-role patterns once comfortable: many items, like requesting distinct options or stating hard constraints explicitly, transfer well regardless of job function.
Revisit the constraint-setting patterns most often: vague requests are the most common source of disappointing output across every role listed here.

FAQs

Do these prompt patterns require a Project to work?

No, they work in a plain chat as well. A Project makes them faster to reuse since the recurring parts, like audience or tone, can live in the standing instructions instead of being retyped each time.

Why are writer and marketer patterns so similar?

Both roles produce content that benefits from consistent voice and scoped revision. The main difference is scale: marketers more often need distinct variations at volume, writers more often revise one continuous piece.

Which pattern on this list matters most if I only try one?

Stating hard constraints explicitly, such as a length limit or a banned phrase, rather than relying on Claude to infer them from a general tone description. This single habit prevents more disappointing output than any other change.

Can support and analyst patterns be combined for one task?

Yes. Summarizing a batch of support tickets by theme borrows directly from the analyst's grouping and headline-first patterns, since both are about turning volume into a scannable structure.

Should I use these exact phrasings, or adapt them?

Adapt them. These are templates showing the shape of a good request, not scripts meant to be copied verbatim into every prompt.

What's the most common mistake across all four roles?

Leaving structure or constraints implied instead of stated. A vague request like "make this better" produces a vaguer result than "tighten the second paragraph, keep the ending as-is."

Why does the marketer section emphasize asking for angles before variations?

Without a real strategic distinction to work from, a large batch of variations tends to blur together. Naming distinct angles first gives Claude an actual structure to differentiate against.

How do I know if a prompt pattern from another role applies to mine?

If the underlying need matches, like wanting distinct options rather than one answer, or wanting a hard constraint respected, the pattern usually transfers even if the specific example doesn't match your role's content.

Should I put these patterns into my Project's custom instructions?

The recurring ones, like required structure or hard constraints, work well there. Patterns specific to a single task, like a one-time comparison request, are better left in the individual conversation.

Why does the support section separate reply drafting from ticket summarization?

They're different tasks with different goals: one drafts a single customer-facing reply, the other produces an internal, pattern-focused document for the team. Mixing the two in one request tends to produce a weaker result for both.

Is asking for multiple options always better than asking for one answer?

Not always, but for genuinely open decisions, like an opening line or a campaign angle, seeing distinct options is usually more useful than evaluating a single first attempt in isolation.

How often should I revisit which patterns I'm actually using?

Every few weeks is reasonable. Patterns that sound useful on paper do not always match how you actually work day to day, and it's worth pruning the list to what you genuinely reach for.

Role-Specific Use Cases Basics - the broader starting point this checklist supports.
Claude for Writers: Drafting and Editing With Projects - a full walkthrough behind the writer prompt patterns above.
Claude for Marketers: Campaign Ideation and Content at Scale - the reasoning behind the angle-first marketing patterns.
Choosing Projects vs Artifacts by Role - deciding where these patterns should live once you have a routine.

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.