This is part 2 of a trilogy on how AI is changing PR and comms practice.

Reporters use AI all day. In Muck Rack's 2026 survey, 82% of journalists said they use at least one AI tool.

They'd still prefer you didn't use it on them. Cision asked about 1,900 journalists how they feel about AI-generated pitches and press releases. 53% are opposed, 25% are neutral and 21% are in favor.

Read what they object to, though. It's marketing language, unsourced claims and irrelevance. Reporters were complaining about all of that long before ChatGPT.

And the flood is coming either way. Anthropic's usage data shows sales and outreach automation at least doubled its share of API traffic in 3 months. Sending is now close to free, for you and for everyone else pitching the same reporter.

So I think relevance is the only thing left to compete on. The same agents that can flood an inbox can also do the homework most pitches skip.

Which parts of pitching should AI agents do?

Agents do the homework. They read a reporter's recent work and write the first draft.

Matching is shared. The agent proposes a target and shows its evidence, and you decide whether it holds.

The send is yours. So is every reply once a reporter writes back.

One weakness to design around: AI is poor at knowing what's new to a specific reporter, and it states wrong facts about people as smoothly as right ones.

1. How do you match a story to the right reporter?

Part 1 ended with a story worth joining. The next question is who should hear about it.

Journalists define relevance by their readers. In the same Muck Rack survey, 78% said a pitch feels relevant when it directly affects the community their audience belongs to. Nearly half said they seldom get a pitch that matches what they cover.

So match on 2 things: what a reporter published in the last 90 days, and who reads them.

The rule I'd enforce is simple. Every match cites the recent piece that justifies it. If the agent can't point to an article, there's no pitch.

The same test works beyond reporters:

TargetWhat a good match looks like
ReporterWrote on this topic recently, and your angle adds something their last piece lacked
Independent newsletterCovers your category for your buyer, takes outside material, and you can offer data or an exclusive
PodcastStill publishing, audience matches your buyer, books guests like your spokesperson
AwardThe category's criteria map to proof you can document
Conference stageThe call for papers fits a talk your speaker can give without a sales pitch

A few things I'd hold to.

One pitch goes to one journalist. If the story is big, the top outlet on your list gets an exclusive offer first.

"No match" is a fine result. Some stories should wait for a better moment.

For podcasts, choose on audience and ignore the reach claims. Buzzsprout's platform data puts the median episode at about 27 downloads in its first week, and the top 1% at 4,658 or more. A small show your buyers listen to beats a big one they don't. Plan to cut the episode into clips, since that's how most people will see it.

Planned news runs on the same engine. For a funding round or a launch, the agent builds the list and tracks who has been offered what, under which embargo. People negotiate the exclusive and run the briefings.

2. How do you write a pitch with AI that a reporter will read?

Keep it short. In a 2023 analysis of pitches sent through its platform, Propel found that pitches of 150 words or fewer got a 5.89% response rate, against 1.46% for long ones. The data is 3 years old, and I doubt reporters have grown more patient since.

Tailor it with one sentence. That sentence connects your angle to something the reporter wrote recently, and it has to be accurate. An invented "loved your piece on X" is the tell reporters spot first. It's worse when the piece argued the opposite.

Give the agent your spokesperson's real writing. The pitch should sound like a person your company employs.

Keep a facts register, which is a list of approved claims with their sources. When the brief is thin, agents fill the gaps with numbers that sound right. A made-up statistic in a reporter's inbox is very hard to take back.

Then comes the approval. It should take about a minute, and it's 4 questions:

  • Is this the right reporter?
  • Is the reference to their work accurate?
  • Is every fact ours?
  • Would I send this under my own name?

If the answer to any of them is no, the pitch goes back.

Send one follow-up a few days later, and leave it there.

What do you say when a reporter asks if AI wrote it?

Someone will ask, so decide your answer before they do.

Here's the one I'd give: agents help us research and draft, a named person approves every pitch, and that person is the one replying to you now.

It's short, and it puts a human on the hook.

PR people already lean this way. In Muck Rack's State of AI in PR survey, 9 in 10 said they aren't comfortable letting an agent send messages without human review.

3. What else can AI agents draft from your own material?

Pitches are one output. The same library feeds several more.

Start by putting your raw material in one place: case studies, metrics, customer quotes, bios and past submissions. Most companies have all of it, scattered across 4 drives and somebody's inbox.

Data stories. Point an agent at your product or customer data and ask what a reporter could use. Journalists in the Muck Rack survey named original research among the most useful things a pitch can carry. A person checks the method before any number leaves the building.

Awards and speaker abstracts. The agent maps an award's criteria to your evidence and tells you what's missing ("criterion 3 needs a customer metric you haven't supplied"). You add the story and check the claims. The tenth entry goes much faster than the first, because the agent reuses what worked.

4. Who should own the relationship with a reporter?

You should.

In the agency model, the record of who was pitched what, and who replied, sits in the agency's inbox. When your account lead moves on, the history goes with them. (It's the part of a retainer I'd question first.)

So keep your own log. Record every pitch and every reply, along with what each reporter prefers: wants data, say, or hates follow-ups. Write nothing about a reporter you wouldn't want them to read.

Be useful before you ask for anything. Answer a source request quickly. Send a data point with no pitch attached.

And when a reporter replies, a person takes the thread that day. The agent can prepare the brief beforehand (their last 5 pieces, the questions they're likely to ask) and log what you promised afterward. The conversation itself is yours.

After 6 months, that log shows which reporters respond to which angles. Feed it back into the radar from part 1.

Where does AI pitching go wrong?

I'd watch for 4 failures.

Fake personalization. The pitch praises an article the agent never read properly. Reporters forward these to each other.

Volume creep. Sending is cheap, so the list grows from 5 reporters to 50. The pitch that fit 5 of them now annoys the other 45.

The rubber stamp. If you approve 100% of drafts, you've stopped reading them.

Slow replies. A reporter who writes back at 4pm and hears from you on Thursday has already filed.

What should a person still decide?

The send.

A sent pitch can't be recalled. Anthropic studied how agents behave in real use and found that only 0.8% of their actions are irreversible. Its example of one is sending an email to a customer. A pitch to a reporter belongs in that 0.8%.

So a named person approves every one. That's how we built Newsbound: agents do the research and the drafting, and a person signs off before anything reaches a journalist.

Some work stays with senior people from start to finish. A crisis is the obvious case, and so is an acquisition.

What goes on the one-page pitch checklist?

StageCheck
MatchWhich piece from the last 90 days justifies this target?
MatchDo this reporter's readers include our buyers?
MatchHas anyone else been offered this story, and on what terms?
DraftIs it under 150 words?
DraftDoes every claim appear in the facts register?
SendIs this the right reporter?
SendIs the reference to their work accurate?
SendWould I send this under my own name?
AfterWho answers the reply, and by when?

Run that every week and the pitches go out one reporter at a time, each approved by a person.

Next in the trilogy: part 3, the measurement stack, on what to show your CFO.

Yoni Argaman is the founding CMO of Newsbound and a 5x tech CMO/VP Marketing.