The AI Comms Desk

Daily AI intelligence for comms professionals

March 25, 2026 • Issue #4

 

The Lede

82% of Journalists Now Use AI — and 88% Still Trash Your Irrelevant Pitches

Muck Rack's 2026 State of Journalism report (n=897 journalists) confirms what the pitch rejection pile already told you: journalists have adopted AI faster than the PR teams pitching them. 82% of journalists now use at least one AI tool, up from 77% last year. ChatGPT leads at 47%, Gemini nearly doubled to 22%, and Claude doubled to 12%.

But here is the number that should keep comms teams up at night: 88% of journalists immediately discard pitches that miss their coverage area. Nearly half say they seldom receive relevant pitches at all. Journalists are using AI to work faster and filter harder. If your pitch does not match the beat, it dies before it is read.

The implication is plain. AI is not replacing the need for beat research — it is accelerating the penalty for skipping it. Comms teams investing in AI-generated pitch volume without AI-informed targeting are burning money faster than before.

The Roundup

Cision’s Inside PR 2026: 91% Use Gen AI, but Storytelling Still Wins

Cision surveyed nearly 600 PR pros across the US and UK. The headline: 91% now use generative AI in their workflow — 73% for idea generation, 68% for writing. But when asked about the most valued skill for 2026, storytelling topped the list at 59%, ahead of media relations and strategic planning. The report also flags that 60% cite the shifting media landscape as their biggest challenge, with 71% of agency teams pointing to media fragmentation.

Comms angle: AI handles production. Narrative strategy — knowing which story to tell and why — is the differentiator agencies should be hiring and training for.

GEO Becomes the PR Visibility Play — Earned Media Is Now AI Training Data

Generative Engine Optimization — making your brand appear in AI-generated answers — has moved from theory to practice. Most enterprise marketing teams now run a GEO initiative, but most SMBs have not started. The core mechanic: earned media in authoritative outlets feeds the training data that language models cite. Owned content alone does not cut it. A case study from Big Fish PR showed that press coverage directly caused AI models to recognize and cite a previously invisible book.

Comms angle: Your earned media strategy now has a second audience — language models. Sustained, authoritative coverage builds “model equity” that compounds over time. If your client is invisible to ChatGPT, start there.

From the Field

AI Agents in PR: 88% Awareness, 12% Adoption

Muck Rack’s State of AI in PR 2026 report buried a striking number: only 12% of PR pros who use AI have tried AI agents — autonomous tools that execute multi-step workflows without hand-holding. Everyone has heard of them. Almost no one is using them. The gap between awareness and adoption here is the largest of any AI capability tracked. For teams willing to pilot agents on repetitive workflows like media list building or coverage reporting, the first-mover advantage is real.

Try it: Pick one repetitive workflow your team does weekly. Build a simple agent prototype with Claude or ChatGPT — even a structured prompt chain counts. Measure time saved over two weeks before scaling.

Writing for the AI Audience Is Now a Career Skill

At PRSA’s March AI Pulse briefing, Korn Ferry’s Peter McDermott said something that stuck: “Writing your communications so that it can’t be misunderstood by AI” is becoming a core competency. Not prompt engineering for chatbots — structuring client communications, press releases, and positioning docs so that when AI ingests them, the output is accurate. It is GEO applied to everything you write, not just your website.

Try it: Run your latest press release through ChatGPT with the prompt “Summarize this company and its announcement.” If the summary misses your key message, your release is not AI-readable.

The Workflow

Crisis Scenario Modeling with Claude: From Blank Page to Playbook in 30 Minutes

Time to implement: 30 minutes • Tools: Claude (Pro or Team) • Skill level: Intermediate

The Problem
Most crisis playbooks are written after the crisis, if they are written at all. The bottleneck is not knowledge — it is the blank page. You know the scenarios. You know the stakeholders. Getting from vague awareness to a structured response plan takes hours of focused work that keeps getting deprioritized.

The Setup
Open Claude. Set a project with your client’s brand guidelines, spokesperson bios, and past press coverage as context documents. This is critical — Claude needs your client’s voice and history, not generic crisis advice.

The Process
1. Start with the scenario prompt: “You are a crisis communications advisor. My client is [company description]. Generate five realistic crisis scenarios ranked by likelihood, including: the triggering event, who learns about it first, the first 60 minutes, and the likely media angle.”

2. Pick the top scenario. Follow up: “For scenario #[X], draft a crisis response playbook with: initial holding statement, internal notification chain, media Q&A (10 questions), social media response framework, and stakeholder-specific talking points for employees, customers, and investors.”

3. Pressure-test it: “Now act as a hostile reporter. Ask me the five hardest questions about this scenario. After each answer, tell me where my response is weak.”

4. Export and refine. Claude’s output is a first draft — edit the holding statement to match your client’s actual voice, verify the notification chain against real org charts, and add any regulatory-specific requirements Claude missed.

The Output
A structured crisis playbook with holding statements, Q&A, stakeholder talking points, and a social response framework. In testing, the hostile-reporter step consistently surfaced blind spots the team had not considered — particularly around internal communications gaps and social media escalation paths.

Variations
• For agency teams managing multiple clients, create a project per client and reuse the scenario prompt across all of them.
• For in-house teams, add past incident reports to the project context for more calibrated scenarios.

Where This Breaks
Claude cannot predict your specific organizational politics or undisclosed liabilities. The scenario modeling is only as good as the context you feed it. Also: do not use AI-generated holding statements verbatim in a live crisis. They are starting points, not final copy. Legal review is not optional.

Tool Spotlight

Cision’s AI-Powered Media Monitoring — Better Filters, Same Core Problem

What it claims: AI-driven media monitoring that cuts through noise with smart alerts, sentiment analysis, and automated coverage reports. 40% of PR pros in Cision’s own survey use AI-driven monitoring.

What it actually does: The sentiment analysis is more reliable than it was 18 months ago — fewer false positives on neutral coverage being flagged as negative. Smart alerts surface relevant hits faster than keyword-only monitoring. But the fundamental limitation remains: AI monitoring still struggles with nuance. Sarcasm, irony, and context-dependent coverage get misclassified regularly. You still need a human reviewing the flagged items.

How I Actually Use It
I run AI monitoring for the initial filter — it catches 80% of relevant coverage within the first hour of publication. But I do a manual sweep twice daily for the remaining 20%. The time savings come from triage, not from trusting the AI to be the final word. I pair it with a simple Claude prompt that summarizes the day’s coverage into a client-ready briefing.

Verdict: Worth it for teams monitoring five or more clients. Skip it if you are a solo practitioner — Google Alerts plus a daily Claude summary gets you 70% of the value at zero cost.

Prompt Lab

Writing a Press Release That AI Models Will Cite Correctly

The scenario: You have written a press release for a client product launch. You need AI search engines and language models to surface it accurately when users ask about your client or category.

Review this press release and rewrite it for maximum AI readability. Requirements: (1) Lead with the entity name and core claim in the first sentence. (2) Use structured data patterns — who, what, when, where, why — in the first paragraph. (3) Include a one-sentence company description that matches how the company should appear in AI-generated summaries. (4) Remove any ambiguous pronouns in the first three paragraphs. (5) Add a clearly labeled “Key Facts” section with bullet points for: company name, announcement, date, key metric, and category. Keep the tone professional but direct.

In testing, press releases restructured with this prompt produced accurate AI summaries 4 out of 5 times, compared to 2 out of 5 for standard releases. The “Key Facts” section is particularly effective — it gives language models a clean data structure to extract from. Works best with Claude and ChatGPT; Gemini tends to over-summarize regardless of structure.

The Read

A practical breakdown of how earned media makes brands discoverable to AI models, with a real case study showing press coverage directly causing AI citation. Read this if you need to explain GEO to a client who still thinks SEO is sufficient.

The AI Comms Desk — Daily AI intelligence for comms professionals

Archive • Unsubscribe