This is part 3 of a trilogy on how AI is changing PR and comms practice.

Once a month, a PDF arrives from the PR agency. It has a page of logos and a line like "potential reach: 412 million."

Nobody forwards it, because nobody believes the number.

Then budget season comes. Your CFO asks what the company got for $15,000 a month, and the PDF is no help.

The pressure is real. BCG's 2026 survey of communications leaders found that 37% are considering cutting agency spend by more than 5% in the next 12 months.

And here's the odd part. Measurement is where PR people use AI least. In Muck Rack's State of AI in PR survey, 86% of those who use AI use it for editing. Only 19% use it for measurement and reporting.

I think that's backwards. Counting and comparing is the work agents do well, and it's the work people put off.

So the last piece of the workflow is a report written for finance. It shows what went in, what came out, what it cost and where it's heading.

Which parts of PR measurement should AI agents do?

Agents collect and count. They gather and code the coverage, and they re-run the same AI questions on a schedule.

Scoring is shared. When an agent judges whether an article carried your message, a person scores a sample of 20 and compares. If the two disagree often, the instructions need fixing.

Meaning is yours. What the numbers say, and what you'll do differently next month, is a human call.

1. What should you measure PR against?

An objective, written down before the month starts.

AMEC, the industry's measurement body, puts this first in its Barcelona Principles. It's also the step I'd guess most teams skip.

A usable objective names one audience, one change, a number and a date. For example: "Be named in AI answers for 8 of our 15 buyer questions by June, up from 3."

(If you took the baseline in part 1, you already have the "up from.")

Then sort every number you report into one of 3 layers. The Barcelona Principles ask for all three:

LayerWhat it measuresExamples
OutputsWhat you produced and what appearedPitches sent, placements, share of voice, AI answer visibility
OutcomesWhat your audience now knows or doesTarget accounts that can name what you do, branded search, inbound that mentions coverage
ImpactWhat it did for the businessPipeline and deals where earned media played a part

One rule keeps you credible with finance. An output is never presented as an outcome. Share of voice tells you what was published. Whether anyone changed their mind is a separate question with separate evidence.

2. Which PR metrics should you stop reporting?

Start with advertising value equivalency. It prices an article as if it were an ad of the same size. The Barcelona Principles are blunt about it: "Invalid measures such as advertising value equivalents (AVEs) should not be used."

Site-level reach is next. Of an outlet's 40 million monthly visitors, maybe a few thousand read your article.

Raw clip counts go too. They score a one-line mention in a roundup the same as a feature.

And be careful with podcast audience claims. Buzzsprout's platform data puts the median episode at about 27 downloads in its first week.

These numbers lasted this long because they're easy to produce and they only ever go up.

3. Which PR metrics hold up in front of a CFO?

I'd report 5.

The pitch funnel. How many pitches went out, how many got a reply, how many landed. Add the cost per landed placement, since that's the number finance will work out anyway.

Target-outlet hit rate. Of your placements, what share ran in the outlets you named in advance? Ten placements in outlets your buyers never read is a worse month than two in outlets they do.

Share of voice. Your mentions against 3 to 5 named competitors, inside a fixed set of outlets. Choose the set once and keep it, or the trend means nothing.

Message pull-through. Did the piece carry your key message and quote your spokesperson? An agent can score this across every article. It tells you what was said. What readers believed is another matter.

Speed. Hours from a story breaking to your pitch going out. It's the direct test of the radar from part 1.

4. How do you measure what AI assistants say about you?

It belongs in a PR report because earned media feeds the answers. Muck Rack analyzed 25 million links cited by AI models and found about 84% came from earned media.

In May 2026, AMEC published principles for measuring this. They split the job into 3 areas:

AreaWhat you track
Upstream reputation signalsEarned coverage, third-party commentary, reviews, expert content
Search and content readinessWhether your own site is credible, accessible and structured for AI systems to read
Downstream AI outputsHow you appear in answers: citations, framing, omissions

The method is plain. Take a fixed set of 10 to 20 questions your buyers ask. Run them on the same assistants, on a fixed schedule.

Run each question many times. SparkToro and Gumshoe ran 2,961 prompts and found that ChatGPT and Google's AI returned the same list of brands less than 1% of the time. So track how often you appear across many runs, and ignore any single screenshot.

AMEC's standard for evidence is "repeatable prompts, documented methods, transparent assumptions and clear limitations." It says to treat AI outputs as "directional evidence rather than absolute truth," and it warns against leaning on any single tool's score.

Be careful about cause. Report the sequence: "In the 3 weeks after the article ran, we appeared in answers to 4 more of our 15 questions." Leave out the word "because." Your competitors published that month too, and the models changed.

I wouldn't promise a percentage to anyone.

5. How do you connect PR to revenue?

Partly, and I'd say so to the CFO before they ask.

Some outcomes you can measure. Run a short survey of target accounts each quarter ("can you name what we do?"). Watch branded search after coverage lands. Add a "how did you hear about us" question to your demo form.

Some impact you can log. Put 2 fields on every deal: where the buyer first heard of you, and whether they mentioned an article or an AI assistant. It takes a rep 10 seconds, and in a year you'll probably have data most of your competitors don't.

What I haven't seen anyone show is a clean line from one placement to one closed deal. If a vendor shows you one, ask how they got it.

Finance is used to this. In McKinsey's 2026 global AI survey, only 37% of companies could attribute any profit impact to AI at all, and they keep investing. Partial attribution, stated plainly, is something a CFO can work with.

6. What goes on the one-page PR report?

BlockLayerWhat it shows
Spend and unit costInputTotal PR spend, cost per pitch, cost per landed placement
The funnelOutputPitched, replied, landed, against last month
Share of voiceOutputYou against named competitors, 6-month trend
Message pull-throughOutputShare of placements carrying the key message
AI answer visibilityOutputQuestions where you appear, against the baseline
Audience and business signalsOutcome and impactSurvey result, branded search, deal mentions
Learned and nextJudgmentOne thing that worked, one that didn't, next month's bets

Two rules make it work. Report the same metrics every month, and show each one against the baseline.

An agent can fill every block except the last.

Where does PR measurement go wrong?

I'd watch for 4 failures.

Small numbers. Three placements in a month can't show a trend. Report rolling quarters until the volume is there.

Moving the goalposts. Change the question set or the outlet list in July and you've thrown away January's baseline.

Reading the noise. AI answers shift from run to run. Checking weekly will convince you of things that aren't happening.

The agent grading itself. If the same system writes the pitches and scores the results, a person has to check a sample.

What should a person still decide?

The last block of the report.

"Learned and next" is the only part that needs an opinion, and I'd bet it's the part your CFO reads first. It's where you say what you'll stop doing.

Anything that leaves the building as a claim about results gets a human read too. At Newsbound, visibility in AI answers is one of the things we measure, and we report it as what we observed.

How does measurement feed the radar?

This is where the trilogy closes its loop.

Next month's report tells the radar what to watch. A reporter who replied twice goes onto the watchlist. An angle that never landed comes off the brief. An outlet that keeps appearing in AI answers moves up your short list.

Listen, engage, measure, and then listen better.

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