AI for PR Professionals in 2026: Pitches, Embargoes and Disclosure

AI can clean media lists, draft pitches and releases, summarize coverage, build monitoring queries and prepare crisis holding statements. It can't earn a reporter's trust, decide what your client says in a crisis or carry your ethical duties. For PR professionals in 2026, the safe gains are in research and first drafts.

Speed was never the hard part of pitching. The hard part is that someone on the other end reads your email and remembers it. Below is the work from list to coverage report, the tools built for each stage, and the embargo, confidentiality and disclosure lines that a faster draft doesn't move.

Why faster pitching can backfire with journalists

More pitches won't fix a relevance problem, and relevance is what reporters complain about. In Cision's 2026 survey of 1,899 journalists, 72% said fewer than a quarter of the pitches they receive are relevant, and 53% opposed AI-generated pitches, citing accuracy and personalization. The same survey found 66% use PR-provided content for story ideas, so the relationship is worth protecting.

PR teams feel the same strain from their side. Muck Rack's State of PR 2026, as reported by Bizcommunity, found 71% of respondents name low journalist response rates as the biggest obstacle to earned media. Nearly half pitch more than 20 journalists per announcement. AI makes that number easy to raise, and raising it is the wrong use of the tool.

Adoption isn't the question anymore. Muck Rack's State of AI in PR 2026 puts generative AI use at 76%, with 91% saying they always edit AI text. What separates teams now is which tasks they hand over.

The PR tasks AI handles, with tools built for them

AI helps most with PR work that starts from something you can check: a beat list, a clip, a transcript, an approved message document. The Occupational Information Network (O*NET) lists press releases, media requests, spokesperson coaching and public opinion research among the job's tasks. Each example below is taken from the vendor's site, and the order isn't a ranking. Muck Rack, Cision and Meltwater don't list prices on these pages, so expect a sales call.

Media lists: use AI to prune before you add

Pruning is a good first job for AI on a media list, because every suggestion can be checked against a byline. Give an assistant your list and the last few bylines for each reporter, then ask who has moved beats. Check every suggested change against the reporter's recent work, because a confident wrong answer here turns into a misdirected pitch. Muck Rack's own guidance says agents should support decisions rather than make them.

Here are 40 reporters with their last five headlines.
For the story below, sort them into: covers this now, used to cover this, never covered this.
Quote the headline behind each call. If you can't tell, say so. Don't guess a beat.

Story: [one sentence, public information only]

Pitches: research goes to AI, the sentence that matters doesn't

The opening line of a pitch should show you read the reporter's work, and AI can't honestly claim to have read anything on your behalf. Let AI summarize their last ten stories and test three angles against them. Then write the first two sentences yourself, from something you actually read. Cision's survey says journalists most value original studies, access to experts and embargoed information, and a language model can't supply any of the three on your behalf.

Also assume machines read your release on the other side. Muck Rack's journalist survey puts AI use at 82%, and AP's July 2026 standards approve AI for summarizing documents, while leaving verification with its journalists. Put the news, the number and the named source plainly in the first lines, where a summary will find them. If you send the same pitch widely, read how Gmail mail merge handles names and limits first, then ask whether the list should be that long.

Quotes: never let AI write words for a real person unchecked

An executive quote drafted by AI is fine only after that executive has read and approved it. Ars Technica retracted a February 2026 story after AI tools turned a source's blog post into quotes the source never said, and the reporter who hadn't checked them later lost his job. The same failure on your side means a client quoted saying something they didn't, on a wire. Compare every quote and statistic in an AI-assisted release to its original, one at a time.

Coverage reports: drop AVEs, keep the commentary

AI can sort clips and draft the summary, but it shouldn't pick the measure of success. Version 4.0 of the Barcelona Principles from the measurement body AMEC rejects advertising value equivalents and adds ethical governance for AI and automation, so a tool's automatic "media value" figure doesn't belong on your summary page. Sentiment scores need a spot check too: read 20 clips yourself before you quote a percentage to a client.

One new line may earn a place in the report. Muck Rack's May 2026 analysis of more than 25 million links found 84% of AI citations come from earned media. Muck Rack sells media tools, so weigh that, but it's a fair reason to check how assistants describe your client each quarter.

Crisis: AI prepares the binder before the bad day

The best crisis use of AI happens before the crisis: scenario drafts, holding statements and question lists that counsel has already reviewed. In the first hour, AI can summarize what's circulating, but it can't confirm facts or decide whether your client speaks. Debevoise's guidance on corporate deepfakes recommends pre-drafted statements, takedown notices, named approvers, a way to prove content is fake, and a tabletop test. Add a fake video of your CEO to the next drill.

Embargoes, confidences and disclosure: the rules AI doesn't change

Anything under embargo, anything a client told you in confidence and anything that could move a share price needs a decision about tools before it needs a prompt. The Public Relations Society of America (PRSA), the Federal Trade Commission (FTC) and the Securities and Exchange Commission (SEC) all reach this work. This is general information, so take specific cases to counsel.

Embargoes deserve one more habit. Keep the unreleased version of a story, and the time it lifts, out of any automated send. A scheduled workflow doesn't know that the launch moved to Thursday, and an embargo broken by your own tool is still broken by you.

Where Merlin fits for PR professionals

Merlin helps with the journalist relationships a media database can't hold: what you last said to a reporter, what you promised them and who has gone cold. This Mac assistant needs a connected Gmail and Google Calendar account, so a team on Microsoft mail isn't a fit today. Merlin isn't a media database, a monitoring service or a wire, and phone outreach stays with you. Keep the tools above for that work.

• One record per reporter. Open a journalist in the People panel for a short Right now summary and a timeline of your emails and meetings. Add a note such as "prefers data, hates Friday pitches" and ask What did we last talk about? before you write.
• Relationships cooling off. A reporter lands under Gone quiet after a run of one-to-one emails, at least three, followed by two months of silence in both directions. Messages to more than four people don't count, so your blast list stays out of it.
• Replies you owe. Needs you picks up direct emails that still expect an answer, from two days old until about a month old. That window misses today's deadline request, which is closer to what the built-in Inbox Watcher is for: it flags urgent mail as it arrives, though nothing guarantees it catches every query.
• Prep before a briefing. Shortly before an interview on your calendar, a Meeting Brief lists attendee backgrounds, prior discussions and commitments.
• Media training and client calls. Recording on your Mac is free, and transcripts start on the $29 Pro plan. On a paid plan, audio goes to a transcription service, so decide first whether a confidential call belongs there. Other recorders are compared in this note-taking guide.
• Follow-ups you sign off. Tell Merlin in chat to write to the reporter you owe a data point. Merlin shows the email on a card, and it goes only after you confirm. Routines work differently: they run unattended and can send or post with no card, which is why embargoed sends should stay out of them.

Try it on one announcement, not on the whole practice

One real announcement gives you something to compare with how you normally work, which a platform demo can't.

1. Write the rule first. List what never goes into which tool: embargoed news, client confidences, unreleased financials. Muck Rack found 51% of respondents have an AI use case policy, so if your team doesn't have one, this list can start it.
2. Prune one list. Run the prompt above on your next media list, check each change by hand and note how many names go.
3. Draft the release, then fact-check it line by line. Mark every number and quote against its source. Count the errors, and let that count decide how much you trust the next draft.
4. Hand-write the first two lines of each pitch. Track replies against your last campaign, on a shorter list.
5. Rebuild one coverage report. Replace the AVE column with outcomes such as key message pickup, target-outlet coverage or inbound inquiries, have AI draft the commentary, and see what the client asks about.