An Operator’s Playbook for Precision Media Targeting #
Companion to: Writing Bylines with Claude
Who This Is For #
High-intent operators who pitch media for a living or as a core function: agency leads running earned-media programs, in-house comms teams, and founders/executives doing their own outreach. The workflow assumes you’d rather send 8 pitches that land than 300 that get deleted.
The market context makes precision the whole game. Per Muck Rack’s State of Journalism 2026: 88% of journalists delete pitches that miss their beat, 43% say they seldom receive pitches matching what they cover, 71% reject pitches that feel promotional, and 50% are put off by anything that smells like a mass email. Meanwhile Cision’s 2026 State of the Media (1,899 journalists surveyed) finds 66% rely on PR-provided content for story ideas — the door is wide open — but 53% oppose AI-generated pitches.
Read those two findings together and you get this playbook’s thesis: use Claude for the research — the part where depth wins — and keep the pitch itself human. Claude’s job is to make you the one pitcher in fifty who demonstrably read the journalist’s last six months of work.
Operating assumptions:
- You have Claude with web search / Research mode (claude.ai) or Claude in Cowork.
- You may or may not have a paid media database (Muck Rack, Cision, Roxhill, Propel). This playbook works either way — Claude complements databases; it doesn’t fully replace their contact data.
- You’re playing a repeat game. Journalist research compounds: the dossier you build today is the relationship map you use for years.
What Claude Is — and Isn’t — For in Media Targeting #
Claude is excellent at: synthesizing a journalist’s recent coverage into an actual profile of how they think; finding the through-line across 20 articles; detecting beat drift (what they’ve started covering that their database bio doesn’t reflect yet); mapping competitive coverage; pressure-testing your angle against a specific journalist’s known patterns; and turning all of that into a targeting brief you can act on in minutes.
Claude is not: a contact database (email addresses it produces from memory are guesses — get contacts from databases, mastheads, or the journalist’s own pages); a substitute for reading the journalist’s actual work before you hit send; or a pitch-writing machine (journalists explicitly distrust AI-generated pitches, and templated-smelling outreach is the #1 delete trigger).
The verification rule carries over from every Claude research workflow: anything Claude tells you about a journalist from memory — their outlet, their beat, whether they still work there — is unverified until confirmed against a live source. Journalists change jobs constantly; media databases themselves struggle to keep up. Claude with web search ON and instructions to cite is your baseline mode for this entire playbook.
An ethics line worth stating once: everything in this playbook uses public, professional information — published work, public profiles, stated preferences. Building dossiers on journalists’ personal lives, non-professional social activity, or private details is both counterproductive (it reads as creepy the moment it leaks into a pitch) and outside professional practice. Research the reporter, not the person’s private life.
The Pipeline at a Glance #
| Stage | What Happens | Output |
|---|---|---|
| 0 | Infrastructure setup | Claude Project + dossier system |
| 1 | Build the target universe | Long list of candidate journalists |
| 2 | Deep-dive individual journalists | One-page dossiers |
| 3 | Angle-journalist fit analysis | Match matrix; kill bad fits |
| 4 | Personalization intelligence | The 2–3 details that earn the open |
| 5 | Timing and channel | When, where, how to reach out |
| 6 | Human-written pitch + tracking | Sent pitch; dossier updated with outcome |
Time cost: a rigorous 10-journalist target list through this pipeline takes 60–90 minutes with Claude, versus a half-day manually — and versus 5 minutes for the spray-and-pray export that produces the 88% delete rate.
Stage 0 — Build the Infrastructure Once #
Create a Claude Project per client or per beat area you regularly pitch (e.g., “Fintech media,” “Client X — enterprise IT press”).
0.1 Standing project instructions
You are my media-targeting research analyst for [CLIENT/COMPANY] in [SECTOR]. Standing rules: (1) Always use web search; never state a journalist’s current outlet, beat, or coverage from memory without a live citation — journalists change jobs constantly and stale attribution is the most damaging error in this work. (2) Distinguish clearly between VERIFIED (cited, dated) and INFERRED (your synthesis) in every profile. (3) We research journalists’ public professional output only — published work, public professional profiles, stated preferences. (4) Never draft the actual pitch unless I explicitly ask; your job is intelligence, not outreach copy. (5) When assessing fit, be ruthless — a journalist who is a poor fit for this story should be flagged as such even if they’re high-profile. Prestige is not fit.
0.2 The context corpus
Upload to the Project: the client/company one-pager, current news angles and assets (data, spokespeople, embargo material), past coverage of the company, and — critically — a running log of past pitch outcomes (who was pitched, on what, what happened). This last file turns Claude from a researcher into an institutional memory: it will stop you from re-pitching someone who asked to be removed, and spot that a journalist who ignored three product pitches responded to the one data-led pitch.
0.3 The dossier template
Standardize one format so dossiers are comparable and reusable (save as a Project doc):
JOURNALIST DOSSIER — [Name]
Outlet + role (verified date):
Beat, as evidenced by last 90 days of coverage (not their bio):
Coverage patterns: story formats they favor / sources they quote / angles they take
Recent relevant pieces (3–5, linked, dated):
What they have NOT covered that they logically should (the gap):
Competitor coverage: has covered [rivals]? how framed?
Stated pitch preferences (if publicly stated anywhere):
Active channels: email / LinkedIn / Bluesky / X (with activity level)
Personalization hooks (2–3 specifics, cited):
Fit score for [current angle]: High / Medium / Low + one-line why
Relationship history with us:
Last verified: [date]
Stage 1 — Build the Target Universe #
Start wide, then cut hard. The failure mode here is anchoring on the same 10 names every agency pitches. Claude’s search-and-synthesize loop is built for surfacing the second ring — trade reporters, newsletter writers, and beat-adjacent journalists with better response rates than the A-list.
PROMPT — Universe Sweep(web search / Research mode ON)I’m preparing to pitch this story: [2–3 sentence summary of the story/angle, not the company boilerplate]. Build me a candidate list of journalists who have covered this specific topic area in the last 6 months. For each: name, outlet, a linked recent article on the topic with date, and one line on their angle. Cast wide: national business press, trade publications in [SECTOR], major newsletters (Substack/Beehiiv), and podcast hosts who do news-driven interviews. Explicitly include the non-obvious tier — trade and newsletter writers — not just the top-10 outlets everyone pitches. Exclude anyone whose most recent relevant piece is older than 6 months. Aim for 20–30 candidates. Flag any where you couldn’t verify current employment.
Two refinement sweeps that consistently surface names databases miss:
PROMPT — Competitor Coverage MiningSearch for coverage of [COMPETITOR 1], [COMPETITOR 2], and [COMPETITOR 3] in the last 6 months. Which journalists wrote those pieces? These reporters have demonstrated appetite for exactly this category. For each: link the piece, note how they framed the competitor (favorable/skeptical/analytical), and whether they’ve ever covered us.
PROMPT — Citation-Chain MiningHere are 3 recent articles that are exactly the kind of coverage we want [paste URLs]. For each, identify: the journalist, other journalists whose work is linked or referenced within the pieces, and journalists who wrote follow-on or reaction coverage. This chain is my highest-probability target set.
Database note: if you run Muck Rack, Cision, Roxhill, or Propel, export your database search for the same topic and paste it in — then ask Claude to reconcile: “Compare my database export against your sweep. Who did each miss? Whose beat has visibly drifted from their database classification, judging by their actual last 90 days of output?” Beat drift is the single most valuable thing this reconciliation catches: databases classify journalists by historical tags; Claude reads what they published last month.
Stage 2 — The Deep-Dive Dossier #
Take the top 8–12 candidates from Stage 1 into individual deep-dives. This is the stage that produces the intelligence a mass-pitcher never has.
PROMPT — Journalist Deep-Dive(web search ON; one journalist per run)Build a dossier on [NAME], currently at [OUTLET] (verify this is still true — flag immediately if you find evidence of a move). Research their last 6 months of published work and fill in this template: [paste dossier template]. I care most about: (1) the through-line — what does this person’s body of work suggest they believe about [TOPIC]? What’s their characteristic take: skeptic, enthusiast, methodologist, industry-critic? (2) Story anatomy — do their pieces start from data, from a human anecdote, from a company announcement, from a contrarian claim? (3) Sourcing pattern — who do they quote: executives, academics, analysts, practitioners? (4) The gap — what’s the story in their coverage area they haven’t written yet that our material could fill? Cite every claim with a link. Mark anything you inferred rather than found as INFERRED.
The four numbered questions are the difference between a database record and a dossier. A database tells you Jane Doe covers enterprise AI. The dossier tells you Jane Doe covers enterprise AI as a skeptic who opens with practitioner anecdotes, quotes academics over vendors, and hasn’t yet written the procurement-failure story your customer data supports. That last clause is a pitch.
PROMPT — Preference and Channel CheckFor [NAME]: search for any publicly stated pitch preferences — many journalists state them in their outlet bio, Muck Rack profile, personal site, newsletter, or pinned social posts (things like “no PR pitches,” “email only,” “I don’t cover funding rounds”). Also check where they’re professionally active: LinkedIn, Bluesky, X — and how recently they’ve posted on each. Cite everything. If they have stated a preference, that preference is law.
Channel intelligence matters more than it used to. The 2026 data shows a fragmented landscape: LinkedIn is journalists’ most-used and most-trusted professional platform (62% use it professionally; 33% rank it most valuable), while activity has migrated unevenly between X and Bluesky — 76% of journalists maintain X accounts but only 42% had posted within a month, versus 81% of the 25% on Bluesky. Translation: check where this specific journalist is actually alive, not where journalists in general are.
Batch mode for volume operators
Running 10+ dossiers, don’t do them one conversation at a time by hand. Options, in ascending order of automation: (a) run each deep-dive prompt in sequence in one chat, one message per journalist — Claude keeps template consistency; (b) in Cowork/Claude Code, feed a CSV of names and have Claude work through them, writing each dossier to a file and building a summary matrix at the end; (c) for standing programs, schedule a recurring refresh (see Stage 6.3). Whatever the mode, spot-verify: click 2–3 cited links per dossier before trusting it.
Stage 3 — Angle-Journalist Fit (Where Most Pitches Should Die) #
You now have dossiers. Most operators skip from “list” to “send.” The professionals run the fit gate, because 70% of journalists say the #1 thing they want is alignment with their coverage — and misses get deleted, remembered, and held against your domain.
PROMPT — Fit MatrixHere is our story/angle: [summary + available assets: data, spokesperson, exclusive, embargo]. Against each dossier in this Project, score fit High/Medium/Low on four axes: (1) Beat match — is this literally what they cover
now, per their last 90 days? (2) Angle match — does our framing fit how they characteristically approach stories, or would they have to write against their own grain? (3) Asset match — do we have what their story anatomy needs (they open with data → do we have original data? they quote practitioners → can we offer one)? (4) Timing match — does this collide with or complement what they just published? Then be brutal: which journalists should we NOT pitch despite good surface fit, and why? Output as a ranked table with a one-line pitch rationale per High.
Then, for each High-fit target, the single most valuable exercise in this playbook:
PROMPT — Pre-Mortem the PitchRoleplay as [NAME], based strictly on the dossier: their demonstrated skepticisms, their beat, the 100+ pitches they get weekly. I’m about to pitch them [angle]. As them: What makes you delete this in 5 seconds? What would you need to see in the first two sentences to keep reading? What’s the version of this story YOU would actually want to write — which may differ from the version we want you to write? Answer in their voice, grounded in cited patterns from their actual work — no generic journalist clichés.
That last question — the version they’d want to write — is where operators consistently find the reframe that gets the reply. You are not pitching your story; you are offering raw material for theirs.
What journalists say they want the material to contain, per the 2026 surveys: original data (40% in Muck Rack’s survey prioritize it; Cision’s respondents rank original research and credible data at the top), access to credible sources/experts (58%), and ready-to-use assets. If your fit matrix shows a High-fit journalist but you lack the asset their story anatomy needs, that’s not a pitch yet — it’s a brief for what to build first.
Stage 4 — Personalization Intelligence (Not Personalization Theater) #
“I loved your recent piece on X” is personalization theater; every journalist reads it as the mail-merge token it is. Real personalization demonstrates you understood their work, not that you found their byline.
PROMPT — Hook ExtractionFrom [NAME]’s dossier, give me 3 personalization hooks, each of which proves I actually read and thought about their work: (1) a specific argument, question, or observation from a recent piece (quote it exactly, link it) that our story extends, answers, or complicates; (2) a connection between two of their pieces they may not have had spelled out to them; (3) the gap — the story their coverage arc points toward but hasn’t reached. For each hook, one sentence on how it would naturally open an email without sounding like flattery. Reject any hook that’s generic praise.
Rules of engagement for using them:
One hook per pitch. Two reads as research showing off; three reads as surveillance.The hook must do work. It should connect to the reason you’re writing, not sit decoratively at the top. “Your piece argued X; our data shows the counterintuitive case where X breaks” is a hook. “Great piece on X!” is spam with homework.Professional output only. A reference to their published work or public professional commentary is standard practice. A reference to anything personal is a violation of Stage 0’s ethics line and, practically, an auto-delete.
Stage 5 — Timing and Channel #
PROMPT — Timing Scan(web search ON)For my top 5 targets: (1) What has each published in the last 2 weeks — and does anything create a natural hook or a collision (they JUST wrote our story without us)? (2) Any relevant editorial calendar moments, industry events, earnings dates, or report releases in the next 3 weeks that make this pitch timely — or that will bury it? (3) For each journalist, recommend pitch timing and channel based on their dossier: email is the default; note if their stated preferences or platform activity suggest otherwise.
Baseline mechanics from the 2026 preference data, absent journalist-specific intelligence: 1:1 email, under 200 words, sent before noon in the journalist’s time zone, one timely follow-up — and never a cold mass blast. Journalist-specific intelligence always overrides the baseline.
The collision check in (1) is underrated. Pitching a journalist the story they published nine days ago is more damaging than no pitch: it proves you don’t read them, which is the exact signal 88% of them cite as the delete trigger.
Stage 6 — The Pitch (Human), the Log, and the Flywheel #
6.1 The pitch is yours
Claude got you here; you write the email. Two reasons. First, the trust data: 53% of journalists oppose AI-generated pitches, and journalists are professionally trained pattern-detectors — AI cadence in a pitch reads instantly. Second, the pitch is the relationship artifact; it should sound like a specific human who will be on the other end of the reply.
Legitimate Claude assists that stop short of generation: “Here’s my draft pitch — against [NAME]’s dossier, is anything off-key for how they think? Is my subject line accurate to what I’m offering? Cut it to under 150 words without losing the hook.” Editing and pressure-testing your words, not producing them.
6.2 Log the outcome
After every pitch: update the Project’s outcome log — pitched date, angle, response (reply/silence/decline/story), and anything learned. Thirty seconds per pitch. This is what makes pitch #40 to this beat smarter than pitch #1, and it’s the raw material for the most useful quarterly prompt you’ll run:
PROMPT — Program RetroAnalyze the outcome log. What patterns distinguish pitches that got replies from those that didn’t — angle type, asset type, journalist tier, outlet type, day/time, hook style? Which journalists engaged but never wrote — and what might convert them? Which dossier predictions proved wrong, and what should we change about how we score fit?
6.3 Keep dossiers alive
A dossier decays in about a quarter — journalists move, beats drift. For standing targets, refresh before every campaign, or schedule it: a recurring monthly task that re-verifies employment and appends the last 30 days of coverage to each top-tier dossier means you start every cycle current instead of researching from zero. (In Cowork, this is a scheduled task; ask and I’ll set one up.)
The Tooling Map — Where Claude Sits #
| Layer | Tools | What it’s for |
|---|---|---|
| Contact data + monitoring | ||
Claude(Projects, Research mode, Cowork)Budget note for solo operators: this playbook plus a free/cheap contact layer (Qwoted’s free database, outlet mastheads, journalists’ own sites) replaces a surprising fraction of what a $10K+/yr database subscription does for targeting — what it doesn’t replace is monitoring/alerting at scale and bulk verified contact data.
Failure Modes #
The stale-employment embarrassment. Pitching someone at an outlet they left in March. Databases make this mistake; Claude-from-memory makes it worse; only live verification prevents it. Fix: employment re-verified within days of send, every time — it’s one search.
Prestige capture. Burning your best angle on a top-10-outlet journalist at Low fit while the High-fit trade reporter — who would have written it, and whose piece the nationals would have followed — goes unpitched. Fix: the fit matrix outranks the outlet’s prestige. Trade → national is a ladder; deleted-at-national is a dead end.
Personalization theater. Hook that name-checks a headline without engaging it. Journalists see a hundred of these a week; it now signals automation rather than effort. Fix: the hook must connect to your ask or it gets cut.
AI-cadence pitch. You let Claude draft “just this once” under deadline; the journalist clocks it; per the survey data, half of them are already hostile to exactly this. Fix: Stage 6.1 is a bright line. Claude edits; you write.
Dossier worship. Treating a three-month-old dossier as truth, or treating INFERRED lines as VERIFIED ones. Fix: refresh cadence + the verified/inferred discipline from Stage 0.
The collision pitch. Pitching the story they just wrote. Fix: Stage 5’s two-week recency scan is mandatory, not optional.
Creep drift. Research sliding from professional output into personal territory. Fix: if a detail didn’t come from their published work or public professional presence, it doesn’t exist.
The Operator’s One-Page Checklist #
Setup (once per client/beat): Project created · standing instructions set · context corpus + outcome log uploaded · dossier template saved
Per campaign:
- [ ] Universe sweep run (topic + competitor + citation-chain mining); database export reconciled if available
- [ ] Top 8–12 deep-dived into dossiers; employment verified fresh; 2–3 cited links spot-clicked per dossier
- [ ] Preference + channel check done; stated preferences treated as law
- [ ] Fit matrix run; Low-fit names killed regardless of prestige
- [ ] Pre-mortem roleplay run on every High-fit target
- [ ] One working personalization hook selected per target
- [ ] Two-week recency/collision scan done; timing set
- [ ] Pitch written by a human, under 200 words, pressure-tested against the dossier
- [ ] Outcome logged; dossier updated
Sources #
Journalist preference data: Muck Rack — The State of Journalism 2026 · Muck Rack 2026 report announcement (82% AI adoption) · Cision — 2026 State of the Media Report · Cision press release — PR as primary source · PR Daily — What journalists want in 2026
Platform activity: PRWeek — Journalists more active on Bluesky than X
Tooling landscape: BuzzStream — media databases reviewed · Roxhill — media database comparison