PERSONA Claude Prompt — The Full Guide (Why Specific Personas Beat Generic Ones)

What PERSONA actually does

PERSONA is one of the most underrated Claude prompt prefixes. Drop it in front of any question, follow it with a specific role description, and Claude responds as that person would — including their biases, their opinions, and the framing they'd actually use.

The catch: most people use it wrong. They write PERSONA: expert in databases and get a slightly more confident version of Claude's default answer. The trick isn't telling Claude to be an expert. The trick is giving Claude permission to have opinions by being absurdly specific about which expert.

The wrong way vs the right way

Here's the same question asked two ways. Try both yourself in Claude — the difference is dramatic.

Wrong way:

Act like a database expert. How do I make my Postgres query faster?

Claude gives you the textbook answer. EXPLAIN ANALYZE, check your indexes, look at query plans. Useful but generic. The kind of thing you'd find in any tutorial.

Right way:

PERSONA: Senior Postgres DBA, 15 years at Stripe, deeply skeptical of ORMs, has seen every kind of N+1 disaster. How do I make my query faster?

Now Claude doesn't just answer — it has opinions. It might tell you the ORM is hiding the real problem, that your join order is probably wrong because Postgres's planner is bad at multi-table joins above 8 tables, and that you should drop your ORM's lazy loading entirely for this hot path. Specific, opinionated, useful.

Why specificity unlocks better answers

By default, Claude is trained to be balanced. "On the one hand... on the other hand..." Generic personas don't override this. "Database expert" is too vague — Claude still hedges because there are many kinds of database experts and they disagree.

A specific persona breaks the symmetry. "Senior Postgres DBA at Stripe with 15 years of experience, cynical about ORMs" describes a single coherent worldview. Claude can model that worldview — and people with strong worldviews give clearer answers than committees.

The more concrete the persona, the more permission Claude has to commit to an opinion. Compare:

• ❌ PERSONA: marketing expert → vague, generic answer
• ⚠️ PERSONA: B2B SaaS marketer → slightly better
• ✅ PERSONA: Growth lead at a Series A B2B SaaS, struggled to scale outbound past $1M ARR, eventually figured out PLG works better for their ICP → opinionated, specific, actionable

The 4 ingredients of a good PERSONA prompt

Every PERSONA prompt that consistently works has these four components:

1. A specific role at a specific company (or company type)

Not "engineer" → "Staff engineer at a Series B fintech". Not "designer" → "Lead product designer at Linear". The named context anchors Claude in real constraints that company would actually face.

2. A duration of experience

"15 years" or "3 years" or "new in role" — each produces different answers. A 15-year Postgres DBA recommends battle-tested approaches. A 3-year one is more open to new tools. Both are useful, but for different questions.

3. A bias or strong opinion they'd hold

This is the part most people skip. Without it, the persona is just a job title. With it, the persona becomes a perspective.

• "Cynical about ORMs"
• "Convinced microservices were a mistake"
• "Believes in flat org charts"
• "Hates dependency injection"

Add one bias, get a different answer.

4. A specific scar — something they learned the hard way

This is the secret sauce. "Has seen every kind of N+1 disaster." "Lost a customer to a payment outage in 2022." "Watched their startup die because they over-hired." These specific scars produce specific advice.

Example PERSONAs that consistently produce great answers

PERSONA: Senior backend engineer at a fintech that had a 4-hour outage due to a bad migration last year. Now extremely cautious about schema changes. How should I approach renaming a column on a 50M-row table in production?

PERSONA: VC partner who has invested in 200 indie SaaS companies, cynical about TAM slides, only believes in revenue and retention. Look at my pitch deck.

PERSONA: Senior frontend engineer at Vercel, deep in the Next.js team, has seen every common mistake people make with the App Router. Review my routing structure.

PERSONA: Incident commander at AWS who has been on call for 8 years, has triaged hundreds of P0s, now obsessed with making runbooks idiot-proof. Help me write a runbook for our database failover.

When NOT to use PERSONA

PERSONA is one of the most powerful prefixes in the Claude toolkit, but it's not always the right tool.

Skip it for:

• Factual lookups — "What's the syntax for a Python list comprehension?" doesn't need a persona.
• Creative writing where you want range — A persona collapses the answer to one perspective; for brainstorming you often want multiple.
• Math, calculations, or anything with a single correct answer — Personas can introduce noise.
• Tasks where you don't actually want strong opinions — Sometimes you genuinely want balanced reasoning, and PERSONA undermines that.

The rule: use PERSONA when you want commitment in the answer. Skip it when you want completeness.

PERSONA combos that stack

PERSONA gets even more powerful when stacked with other prefixes:

• PERSONA + L99 → forces both a specific perspective AND maximum reasoning depth
• PERSONA + /nofilter → strips Claude's diplomatic instincts on top of the persona's biases
• PERSONA + /devmode → engineer-to-engineer tone, no beginner explanations
• PERSONA + WORSTCASE → asks the persona what could go wrong from their experience

The most useful combo for technical decisions: PERSONA: [specific senior engineer with relevant scar] + L99 + WORSTCASE. You get a confident, specific recommendation plus the failure modes that experience has taught them to expect.

A complete worked example

Question: Should I rewrite my Express backend in Go?

Default Claude answer: A balanced 5-paragraph essay listing pros and cons of both languages, ending with "it depends on your team's needs."

With PERSONA:

PERSONA: CTO who rewrote their Node.js backend in Go in 2024 and has strong feelings about it (both directions). 8 years building web infra. L99. Should I rewrite my Express backend in Go?

Claude now responds as a specific engineer who has actually done this. You get something like: "Don't rewrite. The Go performance gains are real but they only matter if you're CPU-bound, and most Express apps aren't — they're I/O-bound and Node handles I/O fine. The thing nobody warned us about: Go's error handling will slow your team down by 20-30% for the first 6 months. We saved about 40% on AWS bills, but we lost 4 months of feature velocity. Net negative for us. Here's when it's worth it: ..."

That's the answer you can act on.

Where to find more codes like PERSONA

PERSONA is one of about 120 Claude prompt prefixes the community has discovered. The first 11 are free at clskills.in/prompts — including L99, /ghost, OODA, /skeptic, and others.

If you want the full set with before/after examples for every code, when-NOT-to-use warnings, and the combos that stack into something more powerful than the sum of the parts, the Claude Code Cheat Sheet is a one-time $5-$10 purchase with lifetime updates.

TL;DR

PERSONA tells Claude who to be when answering. The trick is being absurdly specific about which expert — name the company, the years of experience, a specific bias, and a specific scar. Skip generic personas ("act like an expert"); they barely move the needle. Specific personas produce wildly better answers because they give Claude permission to commit to a worldview instead of hedging.

Use it for: technical decisions, advice, anything where you want a strong opinion.

Skip it for: factual lookups, brainstorming, anything where you want completeness over commitment.