# The Best Claude Prompts for Public Relations Professionals

> Source: <https://www.narracomm.com/the-best-claude-prompts-for-public-relations-professionals/>
> Published: 2026-07-18 06:26:43+00:00

# The Best Claude Prompts for Public Relations Professionals

**50 prompts for operators who bill for judgment, not volume.**

Most “AI prompts for PR” lists are written for people who want a press release in ten seconds. This one isn’t. It assumes you run communications for companies where a single sentence can move a valuation, that you already write well, and that your bottleneck is not typing speed — it’s the amount of thinking you can do per hour against the number of accounts, executives, and news cycles you’re responsible for.

Used correctly, Claude does not replace the senior practitioner. It multiplies the number of drafts, angles, counterarguments, and simulations that practitioner can run before committing. That’s the entire value proposition. The prompts below are built for that.

**Part I** covers the mental model — the handful of techniques that separate a prompt that produces usable strategic material from one that produces LinkedIn slop. Skipping it and jumping to the prompt library will cost you most of the value.

**Part II** is the 50 prompts, in ten categories.

**Part III** is the guardrails — the things that will end a career if you get them wrong.

# Part I — The Operator’s Mental Model

## The seven techniques that actually matter

### 1. Be explicit about context, audience, and stakes

The single most common failure in professional prompting is under-specification. Claude performs dramatically better when told *who the output is for, what happens next to it, and what would make it fail.* “Write a press release about our funding round” and “Write a press release announcing a $40M Series B, aimed at fintech trade reporters who have covered our category skeptically, that will be scrutinized by our general counsel and must not imply any regulatory approval we don’t have” produce work of entirely different quality.

Tell it the reader. Tell it the stakes. Tell it the constraints. Tell it what the output is *for*.

### 2. Use XML tags to structure your inputs

Claude is specifically responsive to XML-style tags. Use them to separate the different kinds of material you’re handing over — background, transcript, prior coverage, brand voice rules, the actual task. This prevents the model from confusing your instructions with your source material, which is the most common cause of weird output when you paste long documents.

```
<company_background>...</company_background>
<transcript>...</transcript>
<voice_rules>...</voice_rules>
<task>...</task>
```

This is not cosmetic. On long, mixed-material prompts it is the difference between a clean draft and a mess.

### 3. Give examples — this is how you buy voice

If you want output in a specific voice, showing beats describing. Two or three examples of the executive’s actual writing, or three headlines from the publication you’re targeting, will do more than a paragraph of adjectives about tone. This technique (multishot prompting) is the highest-leverage single move available to a communications professional, because voice is most of the job.

Build a reusable voice file for every executive you support (Prompt 30 does exactly this) and paste it into every drafting session.

### 4. Turn on extended thinking for genuinely hard problems

For strategy, crisis sequencing, positioning trade-offs, and anything where you’d want a smart colleague to actually think before answering, use extended thinking. It is wasted on a boilerplate rewrite and invaluable on “here are four stakeholder groups with conflicting interests, sequence our disclosure.”

### 5. Chain your prompts instead of asking for everything at once

Professional-grade output almost never comes from one prompt. Separate the stages the way you’d separate them with a human team:

**Research → Analysis → Draft → Critique → Revise**

Ask for the analysis first. Interrogate it. *Then* ask for the draft, referencing the analysis. Then, in a separate turn, ask Claude to critique its own draft against the brief as a hostile editor would. Quality improves noticeably at every break in that chain.

### 6. Put long documents at the top, your question at the bottom

When you’re pasting a 40-page transcript, an analyst report, or six months of coverage, lead with the material and close with the instruction. Long-context performance is measurably better this way.

### 7. Always give Claude an out

The most important sentence in any PR prompt: *“If the source material doesn’t support a claim, say so explicitly rather than filling the gap.”* Add it to everything. It is your primary defense against the failure mode that matters most in this profession (see Part III).

## Set up before you prompt

**Build a Project per client or executive.** Load it with the brand voice guide, the approved boilerplate, the messaging architecture, past coverage, product facts, and the list of claims legal has cleared. Every subsequent prompt inherits that context instead of you re-pasting it.

**Write a source-of-truth document.** One file per account containing only verified, approved facts and figures. Reference it explicitly in prompts: *“Use only figures from <approved_facts>.”*

**Pick the model to the job.** Reach for the most capable model for strategy, crisis, and positioning work — anything where the reasoning is the deliverable. Faster, lighter models are perfectly good for volume tasks like reformatting, first-pass summarization, and list processing. Matching the model to the task is a cost and latency decision, not a quality compromise.

**Turn on web search when currency matters.** Claude does not know this morning’s news, this quarter’s beat changes, or who just moved outlets. Any prompt touching current media landscape, live news cycles, or competitor activity should be run with search enabled — and the output still verified.

# Part II — The 50 Prompts

Placeholders appear in `[BRACKETS]`

. Every prompt assumes you’ve loaded relevant context into a Project or pasted it in tagged blocks.

## I. Strategy & Positioning

### 1. The Messaging Architecture Builder

*When: New client onboarding, repositioning, pre-launch.*

```
You are a senior strategist building a messaging architecture for [COMPANY].

<company_material>
[PASTE: product docs, founder interviews, existing site copy, sales deck]
</company_material>

<market_context>
[PASTE: category description, top 3 competitors, what they claim]
</market_context>

Build a messaging architecture with:
1. One positioning statement (25 words max) that a competitor could NOT credibly claim
2. Three message pillars, each with: the claim, the proof required, and the proof currently available
3. For each pillar, the strongest objection a skeptical journalist would raise
4. A "do not say" list — claims the material tempts us toward that we cannot substantiate

Flag every pillar where the proof is aspirational rather than existing. If the
source material doesn't support a pillar, say so rather than filling the gap.
```

**Operator note:** The “proof currently available” column is the one that earns your fee. Most messaging decks die because nobody audited whether the claims were provable.

### 2. The Positioning Stress Test

*When: Before you present. Always before you present.*

```
Act as three hostile reviewers of the positioning below, in sequence.

<positioning>
[PASTE]
</positioning>

Reviewer 1 — A skeptical tier-one reporter: What's the story here, and why
would you not write it? What claim would you fact-check first?

Reviewer 2 — The competitor's head of comms: How would you neutralize this
positioning in your own messaging? What's the counter-narrative?

Reviewer 3 — The client's CFO: What does this cost, what does it return,
and what's the argument against spending anything on it?

For each, write in that reviewer's voice and be genuinely adversarial. Then
give me the three revisions that would most improve survival odds.
```

**Operator note:** Claude will default to agreeable. Explicitly demanding adversarial roleplay is how you get real pressure-testing. Do this before the client does it to you.

### 3. The Narrative Whitespace Map

*When: You need an angle nobody owns.*

```
<competitor_messaging>
[PASTE: homepage copy, recent press releases, exec quotes from 5 competitors]
</competitor_messaging>

<our_capabilities>
[PASTE]
</our_capabilities>

Map the narrative territory of this category:
1. What claims are saturated (3+ competitors making them)? List them.
2. What claims are contested (2 competitors fighting over them)?
3. What territory is unclaimed AND credible for us given our capabilities?
4. What territory is unclaimed because it's a bad idea? Explain why.

Rank the unclaimed-and-credible territories by: defensibility, media interest,
and buyer relevance. Show the ranking logic.
```

**Operator note:** Section 4 is the guardrail. Empty territory is often empty for a reason, and a strategist who can’t tell the difference is expensive.

### 4. The Proof Point Audit

*When: The client says “we’re the leader in X.”*

```
<claims>
[PASTE: every claim from the client's marketing and messaging material]
</claims>

<evidence>
[PASTE: customer data, case studies, third-party validation, metrics]
</evidence>

For each claim, classify it as:
- SUBSTANTIATED (evidence directly supports it — cite which)
- PARTIAL (evidence supports a narrower version — write the narrower version)
- UNSUPPORTED (no evidence provided — state what evidence would be needed)
- RISKY (may create legal, regulatory, or credibility exposure — explain)

Output as a table. Do not assume evidence exists that isn't in the material.
```

**Operator note:** This is the single most useful defensive prompt in PR. Run it on every new account in week one. The RISKY column has saved campaigns.

### 5. The Audience Ladder

*When: One announcement, five stakeholder groups, five different anxieties.*

```
<core_message>
[PASTE]
</core_message>

<stakeholders>
[LIST: e.g. enterprise buyers, regulators, employees, investors, existing
customers, the trade press]
</stakeholders>

For each stakeholder group, produce:
1. What they actually care about in this announcement (their question, not ours)
2. What they fear it means
3. The message translated for them — same truth, their frame
4. The one thing we must NOT say to this group, and why
5. The channel and messenger best suited to reach them

Keep every version factually identical. Flag any translation that shades meaning.
```

**Operator note:** Point 5 in the instruction is the integrity check. Message tailoring becomes spin the moment the versions stop being reconcilable.

## II. Media Relations & Pitching

### 6. The Reporter Beat Brief

*When: Before any pitch to a journalist who matters. Run with web search on.*

```
Research [REPORTER NAME] at [OUTLET] and build a pitch brief.

Cover:
1. What they've actually covered in the last 90 days — themes, not just titles
2. Their apparent thesis about [CATEGORY] — what do they seem to believe?
3. Story structures they favor (data-led, profile, contrarian, investigative)
4. Companies like ours they've covered, and whether coverage was favorable
5. What they appear to be tired of
6. Three angles from us that fit their demonstrated interests
7. Three angles we should NOT pitch them, with reasons

Cite sources for every claim. Where you're inferring rather than observing,
label it as inference.
```

**Operator note:** Verify before you act. Journalists change beats and outlets constantly, and a pitch built on stale intelligence is worse than no pitch. Treat the output as a research head start, not a dossier.

### 7. The Pitch Angle Generator

*When: You have news and need to know what’s actually newsworthy about it.*

```
<announcement>
[PASTE the raw facts — no spin]
</announcement>

<context>
[PASTE: what's happening in the category, recent related news]
</context>

Generate 10 distinct pitch angles. For each:
- The headline a journalist would write
- Why this is news NOW (the timeliness hook)
- Which outlet tier and beat it suits
- The strongest reason an editor would kill it

Then rank all 10 by genuine newsworthiness — not by what the client wants
to hear. Be direct about which angles are not newsworthy at all.
```

**Operator note:** The kill-reason column is the point. If Claude can generate a compelling reason to spike your angle in one line, so can an editor in one second.

### 8. The Cold Pitch Writer

*When: You know the angle and the reporter. Now earn the open.*

```
<reporter_brief>
[PASTE output of Prompt 6]
</reporter_brief>

<angle>
[PASTE the chosen angle from Prompt 7]
</angle>

<proof>
[PASTE: the data, access, or exclusivity we can offer]
</proof>

Write a pitch email under 150 words that:
- Opens with the news or the tension, never with a greeting about their work
- Makes the relevance to their beat obvious in sentence one
- States plainly what we're offering (interview, data, exclusive, embargo)
- Ends with one specific, low-friction ask

Then give me 5 subject lines: 2 straight-news, 2 curiosity-driven, 1 contrarian.
No adjectives that the reporter would have to take on faith. No "hope you're
doing well."
```

**Operator note:** The word limit is doing real work. Remove it and you get 400 words of throat-clearing.

### 9. The Rejection Post-Mortem

*When: The pitch failed. Learn something from it.*

```
<pitch_sent>
[PASTE]
</pitch_sent>

<outcome>
[Describe: no response / declined / passed to another reporter / covered by
a competitor instead]
</outcome>

<what_ran_instead>
[PASTE the story that ran in that slot, if any]
</what_ran_instead>

Diagnose the failure. Consider: angle weakness, timing, reporter fit, proof
insufficiency, subject line, competitive crowding, or a story that was never
newsworthy. Rank the likely causes by probability and explain your reasoning.

Then tell me what would have to be true — what additional proof or news —
for this angle to work on the next attempt.
```

**Operator note:** Most agencies never do this. Systematic post-mortems are how a pitch operation compounds instead of repeating.

### 10. The Exclusive & Embargo Strategy Memo

*When: The news is big enough that distribution strategy matters more than the release.*

```
<news>
[PASTE]
</news>

<target_outlets>
[LIST with tiers]
</target_outlets>

<business_objective>
```

[e.g. enterprise buyer awareness / investor signal / recruiting / category

leadership]

</business_objective> Build a distribution strategy comparing three approaches: (a) tier-one exclusive, (b) embargoed multi-outlet briefing, (c) open announcement. For each: expected reach vs depth trade-off, relationship cost, risk of break, sequencing, and fit against the stated business objective. Recommend one, with the reasoning stated as an argument a skeptical CMO could attack. Include the failure scenario for your recommendation and the contingency if the embargo breaks.

## III. Press Materials & Announcements

### 11. The Announcement Hierarchy

*When: The client thinks everything is the headline.*

```
<raw_information>
[PASTE everything the client wants included]
</raw_information>

<audience>
[WHO this must land with]
</audience>

Rank every element by news value to that specific audience. Then:
1. Identify the single lede — the one fact that makes this a story
2. Identify what belongs in paragraph two vs the boilerplate vs nowhere
3. List every element the client wants that has zero news value, and explain
   plainly why a journalist would not care
4. Tell me if there is no story here. Say it directly if so.

Do not soften point 4.
```

**Operator note:** Point 4 is why this prompt exists. The most valuable thing a senior counselor does is tell a client they don’t have news — and having the analysis written down makes that conversation survivable.

### 12. The Press Release Drafter

*When: You’ve earned the right to write it — the news is real.*

```
<approved_facts>
[PASTE only verified, cleared facts and figures]
</approved_facts>

<lede>
[PASTE from Prompt 11]
</lede>

<style_examples>
[PASTE 2 press releases whose tone we want to match]
</style_examples>

Draft a press release. Constraints:
- Inverted pyramid; the news is complete in the first sentence
- No adjectives that cannot be substantiated from <approved_facts>
- Banned phrases: "delighted to announce," "leading provider," "revolutionary,"
  "game-changing," "next-generation," "seamlessly," "robust," "cutting-edge"
- Every number traceable to <approved_facts>
- Quotes marked [PENDING APPROVAL] — draft them as placeholders that the named
  person must review, approve, and edit before use
- Under 400 words before boilerplate

If a required element is missing from the facts provided, insert
[MISSING: description] rather than inventing it.
```

**Operator note:** The banned-phrase list is the highest-ROI line in the prompt. Extend it with your own house peeves. And note the quote handling — see Part III.

### 13. The Quote Architect

*When: You need spokesperson quotes that survive both legal and the editor’s cut.*

```
<speaker_profile>
[PASTE: their role, their actual speech patterns from transcripts, what they
personally care about]
</speaker_profile>

<news>
[PASTE]
</news>

Draft 5 candidate quotes for [NAME] to review. Each must:
- Say something only this person, in this role, could credibly say
- Advance the story rather than restate the lede
- Survive being cut to one sentence
- Contain no unsubstantiated claims

Vary the register: one visionary, one operational, one customer-centric,
one candid about difficulty, one forward-looking.

These are drafts for the named individual to approve or rewrite in their own
words. Label them clearly as such.
```

**Operator note:** Ghost-drafting quotes for approval is standard practice. Publishing a quote the named person never saw or approved is not. Keep that line bright.

### 14. The Fact Sheet & Backgrounder Builder

*When: Reporters need context you don’t want them getting from competitors.*

```
<company_material>
[PASTE]
</company_material>

Build a press backgrounder containing:
1. Company at a glance (founded, size, funding, footprint — facts only)
2. What the company actually does, in language a generalist reporter can
   reproduce accurately without a briefing
3. Category context: what problem exists, who else works on it
4. Milestone timeline
5. Leadership bios (100 words each)
6. Three data points a journalist could build a story around
7. Anticipated questions with factual answers

Where the material provided is insufficient, mark [NEEDS INPUT] with a
specific description of what's missing. Do not estimate or infer figures.
```

### 15. The Boilerplate Rewrite

*When: The boilerplate hasn’t changed since Series A and it shows.*

```
<current_boilerplate>
[PASTE]
</current_boilerplate>

<what_changed>
[PASTE: new funding, positioning, scale, markets]
</what_changed>

Write 3 boilerplate options at 60, 80, and 100 words. Each must:
- Be accurate as of today, not aspirational
- Be comprehensible to a reporter with no category knowledge
- Contain no claim requiring a footnote
- Survive translation and syndication without losing meaning

Flag any phrase in the current boilerplate that is now inaccurate or
overstated.
```

## IV. Thought Leadership & Bylines

### 16. The Byline Thesis Miner

*When: The executive has ideas and no time, and you have a transcript.*

```
<transcript>
[PASTE: 60-90 min interview, podcast, internal talk, or Q&A]
</transcript>

Extract every distinct argument this person made — not topics, arguments
(a claim plus a reason). For each:
1. State the argument in one sentence
2. Note how strongly they held it (hedged / stated / insisted)
3. Assess: is this a genuine POV, or a category platitude anyone would say?
4. If genuine, what would make it publishable — what proof or specificity
   is missing?

Rank by how contrarian and defensible each argument is. Be honest about which
ones are neither.
```

**Operator note:** This is the best byline-sourcing method there is. Executives produce far better thinking in conversation than in writing, and mining transcripts beats asking them to draft.

### 17. The Contrarian Take Generator

*When: You need a POV that earns attention without embarrassing anyone.*

```
<category_consensus>
[PASTE: what everyone in this space currently claims to believe]
</category_consensus>

<our_evidence>
[PASTE: data, experience, customer patterns we can cite]
</our_evidence>

Generate 5 contrarian positions our executive could credibly take. For each:
- The claim
- The evidence from <our_evidence> that supports it
- The strongest counterargument
- Who would be angered, and whether that's a problem
- The reputational downside if this position ages badly

Exclude any position we cannot support with evidence provided. Rank by
(defensibility × attention) and flag any that are contrarian purely for
provocation's sake.
```

**Operator note:** The “ages badly” column matters more than people think. Positions taken for engagement have a way of resurfacing in due diligence.

### 18. The Op-Ed Drafter in Executive Voice

*When: Thesis approved, publication targeted.*

```
<voice_profile>
[PASTE output of Prompt 30]
</voice_profile>

<writing_samples>
[PASTE 2-3 things this person actually wrote or said verbatim]
</writing_samples>

<thesis>
[PASTE the approved argument]
</thesis>

<publication>
[NAME] — [PASTE 2 recent op-eds from this outlet to match format and register]
</publication>

Draft an 800-word op-ed. Requirements:
- Open with a specific scene, number, or tension — never with "As the CEO of"
- One argument, defended, not a survey of the topic
- Concede the strongest counterargument explicitly and answer it
- Every claim either evidenced or clearly framed as opinion
- Close with an implication, not a product mention

Match the voice in <writing_samples>, not generic executive register.
```

### 19. The POV Series Builder

*When: One good interview should produce a quarter of social content.*

```
<source_material>
[PASTE transcript or byline]
</source_material>

<voice_profile>
[PASTE]
</voice_profile>

Build 12 LinkedIn posts from this material. Rules:
- Each makes ONE point; no listicles of everything
- Open with the claim, not with context-setting
- 80-150 words
- No engagement-bait questions at the end
- Vary structure: observation, data point, story, correction of a common
  belief, prediction, lesson from failure

Mark which 3 are strongest and explain why. Flag any post that overstates
what the source material actually said.
```

### 20. The Award Submission Builder

*When: Awards are worth more than they should be, and the deadline is Friday.*

```
<award_criteria>
[PASTE the actual judging criteria and categories]
</award_criteria>

<our_material>
[PASTE: results, metrics, testimonials, campaign details]
</our_material>

Map our material against each judging criterion. For each:
1. Our strongest evidence
2. How strong that evidence is relative to what likely wins this award
3. Gaps — what we cannot currently evidence

Then draft the submission narrative, leading with the results a judge scans
for first. Where we lack evidence for a criterion, tell me plainly rather
than writing around the gap. Recommend whether we should enter at all.
```

## V. Crisis & Issues Management

### 21. The Pre-Mortem Risk Register

*When: Before launch, before the funding announcement, before the campaign.*

```
It is [6 MONTHS FROM NOW] and [INITIATIVE] has become a reputational problem
for [COMPANY].

<initiative>
[PASTE the full plan]
</initiative>

<company_context>
[PASTE: sector, regulatory exposure, past controversies, stakeholder map]
</company_context>

Working backwards from failure, produce a risk register:
1. What went wrong (10 distinct scenarios, from most to least likely)
2. The earliest observable signal of each
3. Which stakeholder raises it first
4. Whether it's preventable now, mitigable later, or neither
5. The single decision today that most reduces total risk

Include at least three scenarios where the problem is caused by something
we said rather than something we did.
```

**Operator note:** Run this with extended thinking. The last instruction is deliberate — communications-caused crises are the ones comms teams are worst at anticipating.

### 22. The Tiered Holding Statement Generator

*When: Something is happening and you have twenty minutes.*

```
<situation>
[PASTE what is known — separate confirmed facts from unconfirmed reports]
</situation>

<constraints>
[PASTE: legal posture, what cannot be said, regulatory obligations]
</constraints>

Draft holding statements at three severity tiers:

TIER 1 — Acknowledged, investigating, no admissions
TIER 2 — Confirmed issue, action underway, accountability signaled
TIER 3 — Confirmed serious issue, full accountability, remediation committed

For each: the statement (under 100 words), what it commits us to, what it
forecloses, and the follow-up question it invites.

Then tell me which tier the known facts currently justify, and what new
fact would move us up a tier.
```

**Operator note:** Pre-drafting all three tiers before you need them is the actual best practice. The commitment analysis is what prevents a Tier 1 statement that accidentally admits Tier 3 liability.

### 23. The Hostile Q&A Simulator

*When: Your executive is about to face someone who is not friendly.*

```
You are an investigative reporter who believes [COMPANY] is [SPECIFIC
ACCUSATION]. You have [PASTE: what's publicly known].

<our_position>
[PASTE our factual position and approved messaging]
</our_position>

Conduct a hostile interview. Rules:
- Ask one question at a time and wait for my answer
- Follow up on evasion, hedging, or non-answers — escalate when I dodge
- Use silence, restatement, and "so you're saying..." traps
- Do not accept a talking point as an answer to a factual question
- After 10 exchanges, break character and assess: where did I lose credibility,
  which answers created new liability, and what would you have written?
```

**Operator note:** Turn-by-turn roleplay is the single most underused capability in PR. Media training that used to require a booked consultant now runs at 11pm before the interview.

### 24. The Stakeholder Sequencing Plan

*When: Who hears it first determines how it goes.*

```
<news>
[PASTE: the disclosure, restructuring, incident, or departure]
</news>

<stakeholders>
[LIST: employees, customers, investors, regulators, partners, press, community]
</stakeholders>

<constraints>
[PASTE: legal disclosure obligations, market rules, contractual notice terms]
</constraints>

Build a sequencing plan: who is told what, in what order, through which
channel and messenger, within what time window.

For each group state: the risk of them hearing it from someone else first,
what they'll do with the information, and the leak probability once informed.

Flag every legally mandated disclosure timing you can identify from the
constraints — and flag where I need counsel to confirm rather than guessing.
```

**Operator note:** That last line matters. Claude should surface where legal input is required, not simulate it.

### 25. The Apology Audit

*When: The statement is drafted and someone needs to check whether it apologizes.*

```
<draft_statement>
[PASTE]
</draft_statement>

Audit this against the anatomy of a credible apology:
1. Does it name what happened specifically, or hide behind abstraction?
2. Does it accept responsibility, or use agentless constructions ("mistakes
   were made," "the situation occurred")?
3. Does it acknowledge harm to specific people, or only "concerns"?
4. Does it commit to specific remediation with a timeline, or gesture at
   "doing better"?
5. Does it contain a "but" or "if" that negates the apology?
6. Would the harmed party read this as an apology?

Quote the exact phrases that fail each test. Then rewrite the weakest section.
Be blunt — a non-apology that reads as an apology internally is the standard
failure mode here.
```

## VI. Executive Communications & Prep

### 26. The Interview Prep Dossier

*When: 30 minutes before the call and the executive hasn’t read anything.*

```
<interview_details>
Reporter: [NAME], [OUTLET]. Topic: [TOPIC]. Format: [live/recorded/written].
</interview_details>

<reporter_brief>
[PASTE output of Prompt 6]
</reporter_brief>

<our_messaging>
[PASTE approved messaging and facts]
</our_messaging>

Build a one-page prep dossier:
1. The three messages to land, in priority order
2. The 8 questions most likely to be asked, with a 2-sentence answer each
3. The 3 questions we least want and how to answer them honestly without
   creating a headline
4. Bridging language that doesn't sound like bridging
5. Specific words and phrases to avoid, with the reason
6. The quote we want them to pull

Keep it scannable. An executive will read this in the car.
```

### 27. The Media Training Sparring Partner

*When: Building the reflex, not just the answer.*

```
You are conducting a media training session with [ROLE]. Topic: [TOPIC].

Run three rounds, one at a time:
Round 1 — Friendly trade reporter, softball questions
Round 2 — Well-prepared business reporter, informed skepticism
Round 3 — Adversarial reporter with a thesis and a deadline

After each round, break character and give feedback on: message discipline,
answers that created new exposure, verbal tics, over-explanation, and the
single strongest and weakest answer.

Escalate difficulty regardless of how well I perform.
```

### 28. The Internal Message Test

*When: The all-hands message will be screenshotted and sent to a reporter.*

```
<internal_message>
[PASTE]
</internal_message>

<context>
[PASTE: what's happening — layoffs, restructuring, acquisition, controversy]
</context>

Analyze as three readers:

1. A worried employee: What do they hear? What's the first question they ask
   a colleague? What's left conspicuously unsaid?
2. A departing employee with a grievance: What here would they screenshot and
   send to a journalist, and what would the resulting headline be?
3. A reporter who receives the full text: What's the story?

Then rewrite the passages that create the most exposure — without removing
honesty. If the message is evasive, say so; evasion is what gets leaked.
```

**Operator note:** Assume every internal message is external. This prompt operationalizes that assumption.

### 29. The Keynote Narrative Architect

*When: The executive has 25 minutes and a deck of 60 slides.*

```
<raw_content>
[PASTE: everything they want to cover]
</raw_content>

<audience>
[WHO is in the room and what they came for]
</audience>

<objective>
[What should change in their minds]
</objective>

Build a talk structure:
1. The single argument this talk makes (one sentence)
2. The opening 90 seconds — a specific story, number, or tension, no agenda slide
3. Three movements, each with its claim, evidence, and transition
4. The one idea they'll repeat to a colleague afterward
5. What to cut from <raw_content> and why

Then flag every place where the narrative depends on a claim we haven't
evidenced.
```

### 30. The Executive Voice Profile

*When: Once per executive. Then reuse forever.*

```
<samples>
[PASTE 5+ verbatim samples: transcripts, emails they wrote, unedited posts,
Q&A answers]
</samples>

Build a reusable voice profile documenting:
1. Sentence rhythm and typical length
2. Vocabulary they reach for — and words they never use
3. How they open and close arguments
4. Their relationship to hedging, certainty, and admission of doubt
5. Recurring metaphors and reference domains
6. Their humor, if any, and its register
7. Rhetorical tells that make writing recognizably theirs
8. What a bad impersonation of them gets wrong

Output as a reference document I can paste into future drafting prompts.
Quote specific examples for each characteristic.
```

**Operator note:** This is the highest-compounding prompt in the guide. Build it once per principal, store it in the Project, and every draft afterward starts closer to their voice. Point 8 is the one that actually prevents parody.

## VII. Measurement, Analysis & Reporting

### 31. The Coverage Quality Scorer

*When: You need to report on coverage without resorting to AVE.*

```
<coverage>
[PASTE: articles, with outlet and date]
</coverage>

<objectives>
[PASTE: what the campaign was trying to achieve]
</objectives>

Score each piece 1-5 on:
- Message pull-through (did our key messages survive?)
- Prominence (headline/lede vs passing mention)
- Audience fit (does this outlet reach our buyer?)
- Sentiment and framing (including subtle framing against us)
- Spokesperson positioning (authority vs vendor quote)
- Competitive context (did we appear alongside or beneath competitors?)

Output a table with scores and one-line justifications. Then: what does the
pattern say about our pitching, and which two pieces genuinely moved the
objective versus which merely accumulated?
```

**Operator note:** Define the rubric before the campaign, not after. Retroactive rubrics are how agencies grade their own homework.

### 32. The Share of Voice Analysis

*When: The client asks how they’re doing versus the competition.*

```
<our_coverage>
[PASTE: 90 days]
</our_coverage>

<competitor_coverage>
[PASTE: same period, per competitor]
</competitor_coverage>

Analyze:
1. Volume share — but weight by outlet reach and relevance, not raw count
2. Message share — which narratives is each company associated with?
3. Quality gap — who gets features vs mentions vs quotes?
4. Territory — which topics does each company own in coverage?
5. Where we're winning that we didn't intend to
6. Where a competitor owns territory we claim in our messaging

State clearly where the data provided is insufficient for a conclusion.
```

### 33. The Client Report Narrative

*When: The numbers are what they are and the story must still be honest.*

```
<results>
[PASTE: coverage, metrics, activities]
</results>

<goals>
[PASTE what we committed to]
</goals>

Write the monthly report narrative. Structure:
1. What we set out to do
2. What happened — including what didn't work
3. What we learned that changes next month's approach
4. What we need from the client to do better
5. Next month's priorities

Rules: lead with outcomes not activities; do not describe effort as if it
were result; state misses plainly before explaining them. If the month was
weak, the report should say so in the first paragraph.
```

**Operator note:** Clients forgive weak months. They do not forgive discovering the weak month themselves, three months later.

### 34. The Message Pull-Through Audit

*When: You want to know if the messaging architecture is actually working.*

```
<messaging_architecture>
[PASTE the intended messages]
</messaging_architecture>

<coverage>
[PASTE 90 days of articles]
</coverage>

For each intended message, calculate:
1. How often it appeared, in what form (verbatim / paraphrased / distorted /absent)
2. Which spokespeople delivered it successfully
3. Which outlets consistently reframed it, and how
4. Whether journalists are using our language or the category's default language

Then: which messages are not landing, and is the cause the message itself,
the proof, the spokesperson, or the targeting?
```

### 35. The Campaign Post-Mortem

*When: It’s over. Extract the institutional knowledge before everyone forgets.*

```
<campaign_plan>
[PASTE original plan and objectives]
</campaign_plan>

<what_happened>
[PASTE results, coverage, timeline, obstacles]
</what_happened>

Conduct a post-mortem:
1. Objectives met, partially met, missed — with evidence
2. Which of our assumptions turned out to be wrong?
3. What worked that we did not predict?
4. What would we do differently with the same budget?
5. What would we need (budget, access, proof, time) to do materially better?
6. What should become standard practice from this campaign?

Be specific and avoid diplomatic hedging. This document is for us, not the client.
```

## VIII. Competitive & Market Intelligence

### 36. The Competitor Narrative Teardown

*When: Know the other side’s messaging better than they do.*

```
<competitor_material>
[PASTE: site copy, releases, exec interviews, social, job posts]
</competitor_material>

Reverse-engineer their communications strategy:
1. Their positioning statement, as they'd write it internally
2. Their message pillars and the proof they lean on
3. Their intended audience — who is this actually written for?
4. What they're conspicuously not saying, and what that implies
5. Where their messaging exceeds their evidence
6. What their job postings and executive hires suggest is coming next
7. The single strongest attack on their positioning that we could credibly make

Distinguish observation from inference throughout.
```

**Operator note:** Point 6 is underrated intelligence. Hiring patterns telegraph strategy months before announcements do.

### 37. The Newsjacking Radar

*When: Something big just happened and you have two hours to be relevant.*

```
<breaking_news>
[PASTE the story]
</breaking_news>

<our_client>
[PASTE: expertise, credible authority, spokespeople available]
</our_client>

Assess in order:
1. Is our client genuinely qualified to comment, or would this be opportunism?
   Answer honestly — if the answer is no, say so and stop.
2. If yes: what's the specific angle only they can provide?
3. What's the reputational risk of commenting on this story?
4. Which reporters are covering it and need an expert source right now?
5. Draft a 60-word reactive comment and a 40-word pitch note.

Flag any scenario where commenting would look exploitative — particularly if
the news involves harm to people.
```

**Operator note:** The stop condition in step 1 is not decoration. Most newsjacking failures are qualification failures, and tragedy-adjacent newsjacking is a career event.

### 38. The Regulatory Calendar Builder

*When: You want scheduled at-bats instead of reactive scrambling. Search on.*

```
Build a 6-month communications calendar for [SECTOR], covering regulatory
milestones, policy decisions, court dates, standards deadlines, major industry
events, and recurring data releases.

For each entry:
1. Date (or expected window)
2. What will happen
3. Why journalists will need expert comment
4. What [CLIENT] can credibly say about it
5. What we'd need to prepare in advance to comment same-day

Prioritize by likely media volume and our credibility to comment. Cite sources
and flag any date that is projected rather than confirmed.
```

**Operator note:** This turns a reactive comms function into an inventory-managed one. The pre-drafted positions are what let you comment in hours instead of days.

### 39. The Analyst & Influencer Map

*When: The people who shape the category aren’t all journalists.*

```
<category>
[DESCRIBE]
</category>

Map who actually shapes opinion in this category beyond traditional media:
industry analysts, newsletter authors, podcast hosts, academic voices,
practitioner communities, and standards bodies.

For each: their apparent thesis about the category, their audience, their
influence on buying decisions, whether they're accessible, and what would
earn their attention (data, access, briefing, exclusivity).

Rank by influence-per-unit-effort. Cite sources; flag inference.
```

### 40. The Category Language Tracker

*When: The words the market uses are changing and you’re using last year’s.*

```
<coverage_sample>
[PASTE 40+ recent articles about the category]
</coverage_sample>

<our_language>
[PASTE our messaging vocabulary]
</our_language>

Analyze the language of the category:
1. What terms do journalists use to describe this space?
2. What terms have fallen out of use in the last 12 months?
3. Where does our vocabulary differ from the market's — and does that make
   us distinctive or unintelligible?
4. Which competitor's language is being adopted by journalists?
5. What terminology is emerging that we should adopt early or resist?

Recommend specific vocabulary changes with reasoning for each.
```

**Operator note:** Point 4 identifies who is winning the narrative war, since the surest sign of message dominance is journalists borrowing a company’s vocabulary unprompted.

## IX. Agency Operations & New Business

### 41. The RFP Response Architect

*When: The response is due and half the questions are procurement boilerplate.*

```
<rfp>
[PASTE the full RFP]
</rfp>

<our_capabilities>
[PASTE: case studies, team, results, differentiators]
</our_capabilities>

First, analyze the RFP itself:
1. What does this document reveal about their real problem versus the stated one?
2. Which requirements are genuine priorities versus procurement filler?
3. What are they anxious about, based on what they over-specify?
4. Is there an incumbent, and what does the RFP suggest went wrong?
5. Should we bid at all? Give me the honest answer.

Then, if we should bid: the response structure, the two proof points to lead
with, and the one question we should ask that demonstrates we understand the
problem better than the competition.
```

### 42. The New Business Diagnostic

*When: You want to walk into the pitch already knowing more than they expect.*

```
<prospect>
[COMPANY] — [PASTE: their site copy, recent coverage, exec social, job posts,
funding history]
</prospect>

Build a diagnostic:
1. Their current narrative — what are they trying to be known for?
2. The gap between that narrative and their actual public footprint
3. Three specific, verifiable observations about their communications that
   they probably haven't noticed
4. What their business stage implies they need from comms right now
5. Who internally likely owns this problem and what their incentive is
6. The one insight that would make them want a second meeting

Ground every observation in the material provided; mark inference as inference.
```

**Operator note:** Point 3 is the pitch. A specific, verifiable observation about the prospect’s own footprint outperforms any credentials slide.

### 43. The Scope & Pricing Rationale

*When: The client wants to know why it costs what it costs.*

```
<scope>
[PASTE what we're proposing to do]
</scope>

<client_context>
[PASTE: their size, stage, category, internal resources]
</client_context>

Build the rationale:
1. Translate each scope item into the business outcome it serves
2. Estimate senior vs junior hours realistically
3. Identify which items are essential versus valuable-but-optional
4. Build three tiers (essential / recommended / comprehensive) with a clear
   logic for what moves between them
5. For each tier, what the client should expect NOT to get
6. The strongest argument a procurement officer would make against this pricing,
   and the honest answer

Do not pad. Flag anything in the scope that we cannot clearly connect to
an outcome.
```

### 44. The Onboarding Brain Dump Converter

*When: Two hours of kickoff call, and someone has to turn it into a brief.*

```
<kickoff_transcript>
[PASTE]
</kickoff_transcript>

Convert into a structured brief:
1. Stated objectives versus implied objectives (what they said versus what
   they seem to actually want)
2. Success as they define it — quote them
3. Constraints: legal, cultural, political, resource
4. Internal stakeholders, their positions, and where they disagree with each other
5. Landmines mentioned in passing that we should take seriously
6. Contradictions between what different people said
7. Questions we must resolve before we can start

Point 6 matters most — surface every place where stakeholders want different
things.
```

**Operator note:** Undetected stakeholder disagreement is the leading cause of dead campaigns. Surfacing it in week one is worth the entire engagement.

### 45. The Difficult Conversation Drafter

*When: The client is wrong, the relationship matters, and the email must go today.*

```
<situation>
[PASTE what happened and what the client is asking for]
</situation>

<our_position>
[PASTE why we disagree, with evidence]
</our_position>

<relationship_context>
[PASTE: tenure, value, temperature, history]
</relationship_context>

Draft a response that:
- Acknowledges their concern accurately before disagreeing
- States our position with evidence, not defensiveness
- Offers a concrete alternative path
- Preserves the relationship without conceding on the substance
- Never uses "as we discussed" as a weapon

Then give me: the sentence most likely to be misread, the likely response,
and a version that is 30% warmer in case I'm misjudging the temperature.
```

## X. The AI-Era Surface: GEO, AEO & Answer Engines

*This category is where the discipline is changing fastest. Your clients’ buyers increasingly ask an AI assistant before they ask a search engine — and answer engines assemble their responses from what they can retrieve and verify. That makes AI visibility a communications problem, not an IT one.*

### 46. The AI Answer Audit

*When: Quarterly, for every client. Search on.*

```
I want to audit how AI answer engines represent [COMPANY].

For each of these questions, research what is publicly retrievable and
assess what an AI assistant would likely answer:
1. "What is [COMPANY]?"
2. "Is [COMPANY] reputable?"
3. "Best companies for [CATEGORY]"
4. "[COMPANY] vs [COMPETITOR]"
5. "[COMPANY] controversy" / "problems with [COMPANY]"
[ADD 5 buyer-specific questions]

For each: what sources dominate the retrievable record, whether they're owned,
earned, or third-party-critical, and what's conspicuously missing that the
company could supply.

Then: rank the gaps by how cheaply they could be filled with owned material.
```

**Operator note:** Run this against the client’s competitors too. The delta is the pitch. And note the framing — the deliverable is a list of *gaps you can fill*, not a complaint about coverage.

### 47. The Entity Consistency Check

*When: AI engines hedge on companies whose basic facts don’t match across sources.*

```
<sources>
[PASTE the company's description from: their site, LinkedIn, Crunchbase, press
releases, executive bios, directory listings]
</sources>

Audit for entity consistency:
1. Where do the company description, founding date, headcount, location,
   leadership titles, and category descriptors disagree across sources?
2. Which version appears most often, and which is authoritative?
3. What ambiguities would cause an AI system to hedge or omit the company?
4. Are executives consistently associated with the company across sources?

Output a correction list: every source that needs updating, what it should say,
and the priority order.
```

**Operator note:** Unglamorous, cheap, and one of the highest-yield tasks in modern comms. Inconsistent entities get hedged or dropped from generated answers.

### 48. The Answer-First Rewrite

*When: Existing content is good and completely unciteable.*

```
<existing_content>
[PASTE]
</existing_content>

<target_question>
[The question a buyer would actually type or ask an assistant]
</target_question>

Restructure this to be citable by answer engines without degrading it for
human readers:
1. Open with a 40-60 word direct answer to <target_question>
2. Restructure the body into scannable, extractable claims — each self-contained
3. Add a table or numbered process where the content contains comparable or
   sequential information
4. Draft 5 FAQ entries with direct answers, suitable for FAQ schema
5. Flag every claim that needs a citation to be credible to a machine
6. Suggest an H1 and headings that match how the question is actually asked

Preserve voice and nuance. Answer-first is a structural change, not a
simplification.
```

### 49. The Citable Announcement Structure

*When: You want the release to feed the machines as well as the reporters.*

```
<announcement_facts>
[PASTE approved facts]
</announcement_facts>

Draft an announcement optimized for both journalists and AI retrieval:
1. A one-sentence factual summary that a machine could extract verbatim
2. The standard release (inverted pyramid, human-readable)
3. A structured facts block: entities, dates, figures, relationships — each
   stated as a discrete, unambiguous claim
4. Three FAQ entries answering what people will ask about this news
5. Explicit entity definitions (who the company is, what the product does) so
   the release is self-contained for a system with no prior context
6. Suggested schema types for the page

Never sacrifice factual precision for retrievability. Every claim must trace
to <announcement_facts>.
```

### 50. The GEO Measurement Framework

*When: The client asks you to prove the AI work is working.*

```
Build a measurement framework for AI answer-engine visibility for [COMPANY].

Include:
1. A tracked query set (20 questions spanning brand, category, comparison,
   and reputation) with the rationale for each
2. What to record per query, per engine (mention, sentiment, sources cited,
   whether owned material is cited, position among alternatives)
3. A baseline scoring method that produces a single trackable number
4. Cadence, and who runs it
5. Which metrics are leading indicators versus lagging
6. Honest limitations: what this method cannot tell us, and where results
   will vary between runs

Design it so a non-technical account team can execute it in under two hours
a month.
```

**Operator note:** Point 6 protects you. AI answers vary between sessions and engines, so a measurement framework that oversells its precision will be discredited the first time a client runs the query themselves and sees something different.

# Part III — The Guardrails

Five rules. The first one is not negotiable.

### 1. Never publish a fabricated quote or statistic

Claude will happily generate a plausible-sounding quote from a named executive and a plausible-sounding market statistic. Both are catastrophic if published unverified. Attributed quotes must be reviewed, approved, and ideally edited by the person named. Statistics must trace to a real, cited source you have personally checked. “The AI wrote it” is not a defense that has ever worked, in any profession.

Build the habit into your prompts: *“Mark all quotes [PENDING APPROVAL]”* and *“Use only figures from <approved_facts>.”*

### 2. Verify anything current before you act on it

Media landscapes change weekly. Reporters move, beats shift, outlets fold. Treat any output about who covers what as a research head start requiring verification, not a source of truth. Run with web search on when currency matters, and check the citations.

### 3. Know your confidentiality position before pasting client material

Understand your organization’s data policies and your clients’ contractual restrictions before putting embargoed news, unreleased financials, or privileged material into any AI tool. Enterprise arrangements differ from consumer ones. This is a question to resolve with your legal and IT teams once, in advance — not per-document, under deadline.

### 4. Ask for the case against — every time

Claude tends toward agreeableness. If you ask whether your strategy is good, you will often be told it is. The prompts in this guide that explicitly demand adversarial review (2, 21, 23, 41) are structured that way for exactly this reason. Adopt the habit generally: *“Give me the strongest argument that this is the wrong approach.”*

### 5. Keep the judgment where it belongs

Claude drafts, analyzes, simulates, and stress-tests. It does not decide whether to disclose, when to apologize, whether a client should take a public position, or how much risk a company can absorb. Those are judgment calls that carry professional and sometimes legal weight, and they belong to the practitioner and the lawyers. The tool that makes you faster at everything else is what gives you more time for them.

## How to actually adopt this

Don’t try all fifty. Pick the three that map to your biggest weekly bottleneck — for most senior operators that’s **30** (voice profiles), **21** (pre-mortems), and **46** (AI audits) — and run them until they’re reflexive. Then build your own.

The prompts in this guide are scaffolding. The moment you start editing them to fit how you actually work, adding your own banned-phrase lists, your own rubrics, your own house standards, they stop being someone else’s prompts and start being institutional knowledge that compounds.

That’s the point. The prompt isn’t the asset. The judgment encoded in it is.