This is part 1 of a trilogy on how AI is changing PR and comms practice.
- The listening layer: how to build an AI radar for earned media (this post)
- The engagement engine: how to pitch journalists using AI without sounding like a bot
- The measurement stack: how to measure PR in the AI era
Your CEO forwards a competitor's article. The subject line is one word: "us?"
You read it. The reporter covers your category. The angle was yours to take. The piece ran Tuesday, and the news it reacted to broke the Friday before.
You found out 4 days late. So did your agency.
This happens a lot, and the frustrating part is that reporters want what you have. In Muck Rack's 2026 survey of 897 journalists, 86% said at least some of their stories start from a PR pitch. Nearly half said they seldom get one that matches what they cover. And 88% discard an off-beat pitch on sight.
So most pitches die before anyone reads a word. They go to the wrong person, or the right person a week too late.
Both are listening problems. And listening is the part of PR that AI agents do best.
Why start with listening?
Most PR teams have already adopted AI. They use it to edit and to write faster, inside the same workflow they had in 2019. Muck Rack's State of AI in PR report found 76% of PR pros use generative AI, and only 12% of those use agents.
I think the bigger gain comes from rebuilding the workflow around agents. That changes how you listen, how you engage and how you measure, so I've split it into 3 posts.
I'd start with listening because it's the safest place to begin. A radar reads and flags. It can't send anything, so it can't embarrass you.
Which PR tasks should AI agents do, and which stay human?
Agents read. Every article in your category and every target reporter's last 20 pieces, at 3am if that's when the story breaks.
Scoring is shared. The agent proposes which stories fit you, you correct it, and next week it's a little sharper.
Deciding whether your company should speak is yours.
1. Who should your PR radar listen for?
Before you pick a single keyword, name the people.
Which 2 or 3 audiences move your revenue? For most B2B tech companies it's the buyers in your ICP, plus the analysts and investors those buyers listen to.
Then ask where those people get informed. The answer is specific to your category. Muck Rack's analysis of 25 million links cited by AI models counted more than 20,000 distinct outlets, with no single publication dominating. So build your list by looking: which outlets do your buyers quote, and which ones appear when you ask an AI assistant their questions?
That list is the core of your radar.
Add the independents. Lenny's Newsletter passed 1 million subscribers in March 2025, and The Pragmatic Engineer reports more than 1.1 million readers. For a lot of B2B buyers, a handful of newsletters and podcasts like these are the trade press now.
Now write the brief. It has 3 parts:
- Topics you've earned the right to speak on.
- Spokespeople, and what each one can credibly say.
- Proof: numbers you can publish and customers who'll go on the record.
A radar with no brief is a keyword alert with better branding.
Last, set one objective per audience, with a number and a date. Something like "quoted in 4 of our 20 target outlets by March." You'll want a goal you can check when March arrives.
2. What should an always-on news feed track?
The feed runs on 5 watchlists.
Your category. The stories your buyers are reading this week.
Your keywords. Product and competitor names, plus the problem the way your buyers phrase it.
Journalists' beats. What each target reporter published in the last 90 days. The database label next to their name ("covers enterprise tech") is too coarse to pitch against.
Beat changes. Press Gazette tracked at least 3,444 journalism job cuts in the US and UK in 2025. Reporters move. Your radar should notice that your contact now covers something else, before you pitch them on the old beat.
Recurring questions. When reporters keep asking something your own data could answer, you've found a data story. Original research is one of the things journalists say they value most in a pitch.
The feed's job is to surface stories you can join. A story qualifies when it's live, a named reporter is on it, and you can add something the coverage lacks.
Two design choices matter more than the tooling.
Send instant alerts only for stories with a clock on them. Everything else goes into one daily digest.
And cap the volume. If the radar sends 40 alerts a day, people stop opening them by Thursday.
The same feed is your early warning. A critical thread picking up replies, say, or a reporter asking around about you. The agent flags it and a senior person decides what to do. Crisis response stays with humans.
3. How do you newsjack a story in under 2 hours?
Some stories have a short clock: a competitor's announcement, or a breach in your category. Think of July 2024, when a faulty CrowdStrike update crashed Windows machines around the world. For days, every reporter on that story needed someone who could explain what had happened.
A reactive comment is worth most in the first hours, while reporters are still writing. By day 3 the follow-ups are filed.
So the radar's job is to get you from signal to a pitch or a comment in under 2 hours. The steps run in this order:
- The agent detects the story and scores it against your brief.
- The agent proposes angles and a soundbite. A person picks.
- The agent finds the reporters already covering it.
- The agent drafts the pitch or comment.
- A person approves, and it goes.
People touch steps 2 and 5. Agents carry the rest.
Before any of it, run a go/no-go test:
- Do we add something new? Fresh data, or an expert who'll go on the record.
- Is the window still open?
- Would we be comfortable seeing our name next to this story?
That third question kills anything touching a tragedy. Good.
Two hours is only possible if you prepared on a quiet day. That means pre-approved soundbites on your standing topics, a spokesperson who picks up within 30 minutes, and legal sign-off on those topics in advance.
Expect most alerts to end in "no." A radar makes it easy to comment on everything, and a company that comments on everything gets ignored.
4. How do you track competitors' press coverage with AI?
Your CEO's "us?" email is a competitor watch, run by hand, with a 4-day lag.
An agent can do it continuously. For each competitor, it maps the coverage by outlet, by angle, by spokesperson and by format.
The map is most useful for what's missing from it.
Look for outlets that cover your category and haven't quoted any competitor. Look for angles a competitor used to own and dropped. Those are open seats.
Look at the reporters, too. Someone who has covered a competitor 3 times is covering the category. They're a target for you, and you already know what they find interesting.
Then check the AI answers. Ask ChatGPT, Gemini and Perplexity the questions your buyers ask. Write down who gets named and which outlets get cited. (Part 3 of this series turns that into a metric.)
One rule: public information only.
5. How do you stop missing award and speaking deadlines?
You've probably found the perfect award a week after it closed.
Awards, speaking slots, calls for papers, podcast guest spots and editorial calendars all have deadlines, and all of them are published somewhere. Finding them is reading work. Give it to an agent.
For each opportunity, the agent scores fit on 3 things: audience match, the organizer's credibility, and cost. Have it flag pay-to-play awards as pay-to-play, so nobody mistakes a purchase for a win. (Some paid awards are fine. Just know which is which.)
Then plan backwards. A deadline on March 1 means a customer sign-off by February 10, a metrics pull by February 17 and a draft by February 22. The agent creates those milestones the day it finds the deadline.
Put 2 more things on the same calendar.
Your own news. Funding rounds and launches sit next to the external dates, so a reactive pitch doesn't land on top of your announcement.
And refresh dates. Muck Rack's May 2026 report found that AI citation rates peak in the first month after a piece is published, then fall steeply through month 6. Of the journalism citations with a known date, 57% were published in the last 12 months. Coverage ages out. When your best article in a key outlet nears its first birthday, the calendar should tell you it needs a successor.
6. What baseline should you record before you start?
Do this in week 1, before the radar changes anything. It's probably an afternoon of work.
Record your share of voice: your mentions against 3 to 5 named competitors, inside your short list of outlets, over the last 90 days.
Then record what AI says. Take 10 to 20 questions your buyers ask and run them through the main engines. Save the answers with the date. Note whether you're named, who else is, and which outlets are cited.
Six months from now, someone will ask whether any of this worked. The baseline is how you'll answer.
Where does AI media monitoring go wrong?
I'd watch for 4 failures.
Alert fatigue. Too many flags and the radar becomes one more muted channel.
Keywords with no context. A fintech company tracking "Apple" learns a lot about fruit.
Chasing everything. The radar makes commenting cheap, and reporters still remember who wasted their time.
Confident mistakes. This is the one to design against. AI gets things wrong in the same polished voice it uses when it's right. In a Harvard and BCG experiment with 758 consultants, people using AI produced 40% higher-quality work on tasks the AI handled well. On a task outside its range, they got the right answer 60 to 70% of the time. Consultants working without AI got it right 84% of the time.
The fix is cheap. Every alert links to its source, so a person can check the claim in one click. If the radar says a reporter covers payments, you should be able to see the 3 articles that prove it.
What should a person still decide?
Nothing a radar does is irreversible. It reads and it flags. So you don't need to approve each step. Watch it, and correct it when it's off.
One decision stays yours: which stories your company joins.
That's the rule we built Newsbound on. Agents do the reading and the drafting, and a person signs off before anything reaches a journalist.
What goes in a one-page radar spec?
| Part | What to write down |
|---|---|
| Audiences | The 2 or 3 groups whose opinion moves revenue |
| Short list | The outlets, newsletters and podcasts they trust |
| The brief | Topics, spokespeople, proof |
| Watchlists | Category, keywords, beats, beat changes, recurring questions |
| Scoring | Relevance, timing, right to comment, risk |
| Routing | Who gets instant alerts, who gets the digest, and the daily cap |
| Calendar | Deadlines, your own news, refresh dates |
| Baseline | Share of voice and AI answers, dated |
Fill that in and you know the moment and the reporter.
Next in the trilogy: part 2, the engagement engine, on what you send them.
Yoni Argaman is the founding CMO of Newsbound and a 5x tech CMO/VP Marketing.