A Production System for Volume Without Spam #
Part 3 of the series: Writing Bylines with Claude · Researching Journalists Before You Pitch
Read This First: The Honest Version of the Promise #
Fifteen pitches a day is trivially easy if you’re willing to send garbage. Claude can generate 15 pitches in ninety seconds; so can a mail merge. Both produce the outcomes the 2026 data documents: journalists get 50–150 pitches a week, respond to about 3.4% of them, delete 88% of off-beat pitches on sight, and — the number that governs this entire playbook — 53% oppose AI-generated pitches outright.
So the promise needs restating before it’s operational: 15 pitches a day means 15 genuinely individual pitches — each to a verified-fit journalist, each with a real hook, each in your voice — produced in a 3–4 hour block because Claude has compressed everything around the writing. The math that makes this legitimate:
- A well-targeted pitch with a strong hook converts at 8–18% for competent operators, versus the 3.4% baseline. Fifteen good pitches/day ≈ 1–2 journalist replies/day ≈ a steady coverage engine.
- The traditional constraint on volume was never typing speed — it was research, targeting, and personalization time (30–60 min/pitch done properly). That’s the part Claude compresses to 5–8 minutes. The final 150 words stay human.
- 15 bad pitches/day doesn’t produce 0 results; it produces negativeresults — blocked domains, burned beats, a sender reputation journalists remember.48% of journalists block PR people for aggressive behavior. Volume without the quality gate is a liability program.
The division of labor, stated once and enforced throughout: Claude does the research, targeting, fit-scoring, hook extraction, draft scaffolding, and QA. You write and own the pitch body. Where Claude touches draft text at all, it edits your words or assembles your pre-written voice components — it never generates a pitch from nothing. This line is what separates a production system from a spam cannon, and it’s also what keeps you honest when a journalist asks (as they now do) whether your pitch was AI-written.
Prerequisites. This playbook assumes the infrastructure from the first two guides in this series: a Claude Project per client/beat with standing instructions, a dossier system for journalists, an outcome log, and verified contact data (database or manual). If you don’t have those, build them first — a volume system on top of no research layer is just faster failure.
The Economics of 15 #
Why 15 and not 50? Because the constraint that matters is quality floor × relationship capital, and 15 is roughly the daily maximum one operator can hit while keeping every pitch above the floor:
| Daily volume | Per-pitch time | What it actually is | Expected outcome |
|---|---|---|---|
| 3–5 | 30–60 min | Traditional artisan pitching | High quality, starvation volume |
| 15 | 12–15 min | This system | Quality floor held; 1–2 replies/day |
| 50+ | <3 min | Mail merge with extra steps | 3.4% → declining; domain burned |
And why the floor is non-negotiable: pitching is a repeat game against a small population. The journalists on your beat number in the dozens, not thousands. Every pitch updates their prior about your domain name. A 5x volume increase that halves your reputation converts worse next quarter than the artisan approach — the spam cannon eats its own future.
The daily block that produces 15:
| Phase | Time | What happens |
|---|---|---|
| 1. Morning intelligence sweep | 30 min | Claude scans news, finds hooks, matches to standing angles + targets |
| 2. Target confirmation | 20 min | Fit-check today’s 15, verify freshness, kill collisions |
| 3. Hook + scaffold generation | 30 min | Claude produces per-journalist briefs (not pitches) |
| 4. Writing sprint | 75–90 min | You write 15 bodies from briefs, ~5–6 min each |
| 5. QA gate + send | 30 min | Claude red-teams; you fix and send |
| 6. Log + follow-up management | 15 min | Outcomes logged; today’s follow-up list executed |
Total: ~3.5 hours. The rest of your day still exists.
Stage 1 — The Morning Intelligence Sweep (Claude, 30 min) #
Volume pitching lives or dies on hooks. A pitch without a why now is a brochure. The sweep’s job is to end with 15 hook-angle-journalist triples, not 15 names.
PROMPT — Daily Sweep(run in the client Project, web search/Research mode ON, first thing)Run my daily pitch sweep for [CLIENT] in [SECTOR]. (1) NEWS SCAN: What broke in the last 24 hours — news, data releases, earnings, regulatory moves, viral discourse — that intersects with any of our standing angles [list 3–5 standing angles/assets: data set, spokesperson topics, contrarian positions]? (2) HOOK RANKING: For each intersection, rate the hook’s strength: STRONG (we add something the story is missing today), MEDIUM (relevant context), WEAK (adjacent at best) — and state what we specifically add that a journalist can’t get elsewhere. (3) TARGET MATCH: From our dossier files, which journalists are actively covering each hooked story or its beat this week? Check what each published in the last 14 days — flag collisions (they already wrote it) and flag anyone we pitched in the last 30 days per the outcome log. (4) Output: a ranked list of hook → angle → journalist triples, best 20 first, each with a one-line rationale and the citation for the hook. STRONG hooks only unless we can’t fill 20.
Operator judgment call on the output: cut the list to today’s 15. The cut criteria, in order: hook strength beats journalist prestige (a STRONG hook to a trade reporter outconverts a WEAK hook to the Journal); never two pitches into the same newsroom on the same story; and anything requiring an asset you don’t have ready today gets deferred, not fudged.
Standing angles are the flywheel here. Maintain 3–5 evergreen angles per client in the Project — the original data set, the spokesperson’s contrarian thesis, the customer-story bank. The daily sweep is cheap because Claude isn’t inventing angles each morning; it’s pattern-matching today’s news against a standing arsenal. Refresh the arsenal monthly.
Stage 2 — Target Confirmation (You + Claude, 20 min) #
Fifteen names survived the sweep. Before scaffolding, three fast checks — these exist because the sweep is only as fresh as its inputs:
PROMPT — Pre-Flight CheckFor today’s 15 targets: (1) Verify each is still at their listed outlet — search for evidence of moves; flag anything ambiguous rather than guessing. (2) Recency: anything they published in the last 48 hours that changes the pitch calculus — either a fresh collision or a fresh opening? (3) Preference check from dossiers: any stated no-pitch topics, format demands, or channel preferences that apply today? Output a GO / ADJUST / KILL flag per target with one line of reasoning.
Kill without sentiment. A killed target today is a better target next week; a mis-fired pitch is a degraded relationship for a year. Backfill kills from the ranked list’s positions 16–20.
Stage 3 — Hook + Scaffold Generation (Claude, 30 min) #
This is the stage most operators get wrong in both directions — either Claude writes the pitch (AI-cadence, delete) or the operator starts from a blank page 15 times (no volume). The middle path: Claude produces a pitch brief per journalist — everything around the words — and you write the words.
PROMPT — Pitch Briefs, BatchFor each GO target, produce a pitch brief — NOT a draft pitch. Format per journalist:
TO:name, outlet, channelHOOK (the why-now):one sentence, with the citationTHE OFFER:what we’re actually giving them — data point with the exact verified number, spokesperson + credential, exclusive/embargo status, asset linksTHEIR ANGLE:the version of this story THIS journalist would want to write, based on their dossier — their characteristic take, their story anatomy, who they’d want to quotePERSONALIZATION HOOK:one specific, cited connection to their recent work that does real work in the email (extends/answers/complicates something they wrote — no compliment-hooks)SUBJECT LINE RAW MATERIAL:the 3–5 words that must appear (the news peg + the offer) — not a finished subject lineLANDMINES:anything from the dossier or outcome log to avoid — past declines, stated dislikes, competitor relationshipsKeep each brief under 150 words. Where a fact isn’t verified in our files, write [VERIFY] rather than asserting it.
Fifteen briefs land in ~10 minutes of Claude time. Spend the remaining 20 reading them — every [VERIFY] flag gets resolved now, against the source, not post-send. A number you can’t verify in two minutes comes out of the offer.
Why briefs beat drafts, mechanically: a draft anchors you to Claude’s sentence rhythms — you edit toward its cadence, and the batch converges on uniform AI-flavored prose that journalists clock. A brief anchors you to facts and angles, and 15 humans-writing-from-briefs produce 15 different-sounding emails even on the same news hook. Uniformity is the tell; briefs prevent uniformity structurally.
Stage 4 — The Writing Sprint (You, 75–90 min) #
Fifteen pitch bodies, ~5–6 minutes each. This is fast because the brief killed every reason to : no researching mid-sentence, no hunting for the stat, no wondering what their angle is.
The form, per the 2026 behavioral data: under 150 words converts at 5.89% vs 1.46% for 500+; under 200 words is the stated preference of 69% of journalists; subject line under 60 characters, peg + offer, no clickbait. The skeleton that fits those constraints:
Line 1 — the hook doing work: their world + your why-now, fused. Not “I hope this finds you well.” Not “I loved your piece.” The personalization hook from the brief, deployed as substance:*“Your Tuesday piece argued X can’t scale — our new data from 400 deployments shows exactly where it breaks.”*Lines 2–4 — the offer, concrete: the number, the source, the spokesperson, the exclusive. What they get, not what you want.Line 5 — the ask, singular and small:“15 minutes with [spokesperson] this week?” / “Want the full data set under embargo?” One ask. Never “let me know if you’d like to learn more.”Sign-off — human. Name, one line of who you are if they don’t know you. No paragraph of company boilerplate.
Sprint mechanics that hold the pace: write all 15 line-1s first, in one pass — openers are the hardest sentence and batching them keeps you in hook-mode; then complete each pitch top to bottom; don’t polish mid-sprint. Voice check: read each aloud once. If two adjacent pitches sound identical, one of them gets rewritten — same-sounding is the failure signature.
Claude’s only role in this stage (optional, per pitch, 30 seconds): “Here’s my draft to [NAME] — against their brief: is the hook doing work or decorating? Is anything inaccurate vs. the verified offer? Cut me to 140 words without losing the ask.” Compression and accuracy checks on your words. If you catch yourself asking Claude to “punch up” or rephrase for style, you’ve crossed the line back into generation — stop.
Stage 5 — The QA Gate (Claude + You, 30 min) #
Every pitch passes four checks before send. Batch it:
PROMPT — Send Gate, BatchHere are today’s 15 final pitches with their briefs. For each, check: (1) ACCURACY — every number, name, title, and claim matches the verified brief; flag any drift (numbers mutate during writing). (2) FIT — does the pitch actually match this journalist’s beat and the brief’s angle, or did the writing wander? (3) PROMO SCAN — flag any sentence that reads as advertising; 71% of journalists reject overly promotional pitches. (4) TELL SCAN — flag AI-cadence patterns, template smell, or any two pitches in the batch that resemble each other too closely. (5) MECHANICS — under 200 words, subject under 60 chars, one ask, correct name spelling (wrong-name is the #1 unforced error in volume pitching). Output: PASS or a specific fix per pitch. Do not soften findings.
You fix, you send — 1:1 emails, before noon in the journalist’s time zone where feasible, never a visible batch. Send-time discipline matters less than fit, but the data says mornings win and the 8 a.m.–noon window is the follow-up sweet spot too.
Stage 6 — The Log and the Follow-Up Machine (15 min) #
The log
Every send gets logged in the Project’s outcome file: date, journalist, hook, angle, subject line, response. At 15/day this log becomes statistically meaningful within weeks — 300+ pitches/month is a real data set, which is the hidden compounding advantage of systematic volume:
PROMPT — Weekly Conversion Retro(Fridays, 10 min)Analyze this week’s ~75 pitches against outcomes. (1) Which hook types, subject patterns, offer types, and journalist tiers got replies? (2) Which briefs predicted well vs. badly — where did fit-scoring miss? (3) Openers: any pattern between line-1 styles and response? (4) What should change in next week’s sweep criteria, standing angles, or brief format? Give me the three highest-leverage adjustments, not a report.
Follow-ups
The data is unambiguous: one follow-up, once, 2–7 days later — 62% of reporters say once only, and repeated chasing gets you blocked. Operationally: each morning, Claude pulls from the log everyone pitched 3–4 business days ago with no response; the follow-up is two sentences (one new element — a fresh data point, an updated peg — plus the original ask), human-written, 15 seconds each. A follow-up that adds nothing is a nag; skip it.
Silence is data, not an invitation. No response after one follow-up = log it, move on, and let the retro decide whether the miss was hook, fit, or timing.
Scaling Beyond Yourself #
Two-operator model (30/day): one operator owns Stages 1–3 and 5 (the Claude-heavy analyst work) for both; both write their own Stage 4 sprints. The analyst/writer split beats two soloists because the sweep and QA batch efficiently.
Cowork/Claude Code model: for standing programs, the morning sweep runs as a scheduled task before you sit down — news scan, target match, and pre-flight drafted into a daily brief file by 7 a.m. Your day starts at Stage 2 judgment instead of Stage 1 collection, which buys back ~40 minutes. (Ask and I’ll set this up as a scheduled task.)
What does not scale: the writing sprint. The moment pitch bodies are generated to hit a number, you’ve rebuilt the mail merge with better research. If the business needs 50/day, the answer is more writers or fewer, better pitches — not deeper automation of the words. The 2026 preference data gives no indication journalists are becoming more tolerant of automated outreach; every survey wave shows the opposite.
Failure Modes at Volume #
Quota poisoning. It’s 4 p.m., you’ve sent 11, and pitches 12–15 get force-fitted to weak hooks to hit the number. Fix: 15 is a ceiling, not a quota. An 11-pitch day with 11 STRONG hooks beats a 15-pitch day with 4 fillers — the fillers are the ones that get you deleted-and-remembered.
Batch convergence. Week 3: your pitches have quietly standardized — same opener rhythm, same ask phrasing. Journalists who get two of them compare notes more often than you think. Fix: the QA gate’s resemblance check, plus rotating your line-1 batching order; convergence is entropy and needs active resistance.
Brief-trust decay. Volume breeds complacency: you stop resolving [VERIFY] flags and stop clicking dossier links because “Claude’s been right all month.” The fabricated-number pitch ships in week 6. Fix: [VERIFY] resolution is a hard gate, and Friday’s retro includes spot-auditing 5 random briefs against sources.
Same-newsroom fratricide. Two pitches, same outlet, same story, same morning — both journalists notice, both decline. Fix: the sweep’s one-per-newsroom-per-story rule is enforced at cut time, not hoped for.
Follow-up creep. The log grows, follow-up discipline slips, and someone gets chased three times. At volume, one blocked domain taxes every future pitch. Fix: follow-ups come only from the log query, capped at one, ever, per pitch.
The AI-authorship question. A journalist replies: “Did an AI write this?” If your Stage 4 line held, the true answer is “I wrote it; I use Claude for research and fact-checking” — which is an answer that builds trust in 2026 rather than ending the relationship. If you can’t give that answer truthfully, the system has already failed upstream of the question.
The Daily Checklist #
Standing (from playbooks 1–2): Project per client · standing angles current (refreshed monthly) · dossiers alive · outcome log running · contact data verified
Every day:
- [ ] Morning sweep run; hook-angle-journalist triples ranked; cut to ≤15 on hook strength
- [ ] Pre-flight: employment verified, 48-hr collision scan, preferences checked; kills backfilled
- [ ] Briefs generated; every [VERIFY] resolved against source
- [ ] Writing sprint: 15 human-written bodies, <200 words, one ask each, read-aloud pass
- [ ] QA gate: accuracy, fit, promo, tell, mechanics — all PASS
- [ ] Sent 1:1, morning-weighted; all logged
- [ ] Today’s follow-up list (3–4 days old, one only, adds something) executed
- [ ] Friday only: conversion retro; three adjustments into next week
Sources #
Response-rate and pitch-form benchmarks: Propel Media Barometer · Martech Zone — Media Barometer response data · Propel Q2 2024 Barometer PDF (length vs. response) · EPR/5W Pitch Response Rate Study 2026
Follow-up data: Sword and the Script — follow-up data roundup (2026) · Cision — following up on a pitch
Journalist preferences: Muck Rack — State of Journalism 2026 · Cision — 2026 State of the Media · PRLab — PR statistics 2026