Press AI -- Get Media Coverage Without a PR Agency - Vertical Agent Spec

One-line definition

An agent that researches relevant journalists, writes personalized pitches, and follows up on behalf of founders seeking earned media coverage without a PR agency.

The workflow it owns end-to-end

Trigger: Founder inputs a news hook (funding round, product launch, partnership, or study) and their company context.
Beat matching: Agent scans a journalist database (Muck Rack, Cision API, or a maintained proprietary index) to identify reporters who have covered comparable companies at comparable stages in the last 90 days, filtering by outlet tier and topic fit.
Asset generation: Agent drafts a press release, a 150-word personalized pitch per journalist (referencing their recent coverage), and a one-page media kit. Each pitch is distinct; they are not the same email with a name swapped.
Outreach sequencing: Agent sends the first pitch, schedules a soft follow-up at day four if no reply, and closes the thread after a second non-response. It logs every send, open, and reply.
Coverage logging: When a journalist replies or a story runs, the agent captures the outlet, the date, and the framing, then updates the founder's media history for future outreach context.

What it knows that a generic LLM doesn't

Journalist beat patterns: Which reporters at TechCrunch, Forbes, and sector trade pubs cover seed-stage B2B SaaS versus growth-stage consumer, and what their recent angles have been, not just their listed beat.
Pitch format norms by outlet tier: A wire-ready press release follows AP style with a dateline and a boilerplate; a direct pitch to a newsletter writer is three sentences and a link. These are not interchangeable, and most founders get this wrong.
News hook timing: Funding announcements have a 48-to-72-hour exclusivity window before broad distribution; pitching 14 days after a round closes is almost always too late. The agent knows the decay curve.
What kills a pitch: Subject lines over 60 characters, attachments in a cold email, pitching a consumer story to a B2B reporter, and leading with the company instead of the reader's audience. The agent applies these filters before sending.
Wire versus direct pitch tradeoffs: Paid wire distribution (PR Newswire, BusinessWire) produces SEO-indexed syndication but almost no journalist pickup. Direct pitching produces actual coverage but requires real personalization. The agent knows when each is appropriate and does not conflate them.
Editorial calendar patterns: Certain story types (year-in-review, funding trend roundups, sector surveys) have predictable publication windows. The agent surfaces these as hooks when a founder has relevant data.

What it explicitly declines

Guaranteed placement: the agent does not promise coverage and does not make any representation that outreach will result in a story. Journalists are not vendors.
Legal review of press release claims: any statement that touches financial performance, regulatory status, or pending litigation goes back to the founder for sign-off before sending.
Paid media or sponsored content placement: that is a different workflow with different ethics, and conflating it with earned media is how trust gets lost.
Outreach to journalists who have previously requested removal from a founder's list: the agent maintains a suppression list and does not override it.

Tools and integrations required

Muck Rack API or Cision API: journalist database with beat tagging, recent article history, and contact data; this is the single most expensive line item and without it the agent is just a writing tool.
SendGrid or Postmark: transactional email for outreach with open and click tracking; pitched from the founder's own domain, not a shared sender.
Hunter.io or Apollo.io: email verification to reduce bounce rate before sending; a high bounce rate on cold outreach can damage domain reputation.
Airtable or Notion API: persistent media contact and coverage log per founder; reused across campaigns.
PR Newswire or GlobeNewswire API (optional): for wire distribution on funding rounds where SEO syndication matters; founders opt in per release.
Slack or Telegram webhook: for real-time reply notifications so founders can respond to journalist interest within the same business day, which is when responses actually matter.

Trust escalation: when it pings a human

A journalist requests an on-record interview, a product demo, or a named executive quote: the agent flags this immediately and stops the automated thread. The founder responds directly.
Any pitch contains a statistic, percentage, or performance claim that the agent cannot verify from the founder's submitted materials: it surfaces the line and asks for a source before sending.
A recipient replies with a legal notice, a cease-and-desist reference, or a complaint about being contacted: the agent halts all outreach to that contact, logs the event, and notifies the founder.
The agent is about to send more than 30 pitches in a single campaign: it presents the full list for founder review first, because volume errors in press outreach are reputationally costly and not easily undone.

Pricing model

The honest pricing model here is per-campaign, not per-seat. A founder pays $149 per media campaign, which covers journalist matching (up to 25 contacts), asset generation (press release, media kit, personalized pitches), outreach sequencing over a 10-day window, and a coverage report at close. There is no monthly subscription by default, because the churn data on SaaS PR tools is unambiguous: users cancel after one campaign regardless of outcome. A subscription tier ($79 per month) makes sense only for companies with recurring news (funding rounds, quarterly product launches, research reports) and should be gated behind proof of that cadence during onboarding. The per-campaign model is honest about the episodic nature of the use case and stops fighting it.

Differentiation from a generic LLM wrapper

The writing is not the product. Any founder can paste their funding announcement into Claude and get a competent press release in 45 seconds, and the known risks section of this product's own brief acknowledges it. The agent is defensible only if it owns the distribution layer: a maintained, verified journalist database with beat-level precision, a sequenced outreach system that sends from the founder's own domain, and a coverage history that compounds over time. A founder using Claude gets a document. A founder using this agent gets 25 personalized pitches sent to journalists who have actually covered their sector in the last quarter, followed up at the right interval, with replies routed back in real time. That workflow takes a competent PR coordinator two to three hours to run manually. If the agent cannot deliver the journalist database and the outreach infrastructure, it should not be built as an agent; it should remain a template tool and price accordingly.