7 Ways PR Pros Should Be Raising the Bar with LLMs
When I got my first clearly AI-written piece of content from a client, my initial reaction was to worry. How soon would it be before my job was obsolete?
Then I read the piece and thought, okay, I have some time. I can make this a lot better.
Now, after receiving many AI-written materials from clients, diving headfirst with AI myself (and with my team), and sharing many documents with clients that started with an AI-baseline, I have a new perspective.
AI can obviously help us move faster. It can also help us collaborate, express ourselves, and understand each other’s thinking. It’s also a great partner to build out strategic ideas. But not if we’re passive participants, sit back, and let AI happen all over us.
A new divide is emerging between those who use AI intentionally and those who just copy/paste whatever comes out.
Machines can get us pretty far. But they can get everyone pretty far. And that’s pretty far toward average because AI is the average of what’s out there. So, to stand out, you have to be part of the few communications professionals who take AI in PR further—who raise the bar. Here's how, by training your AI models and more:
Perfecting the Use of AI in PR
1. Protect Your Knowledge Before You Do Anything
Before you even start prompting, know this: most LLMs store and train on what you input unless you actively opt out. That means anything sensitive (marketing roadmaps, non-public sales data, proprietary client information, even draft messaging) could become part of the model.
We’ve seen the horror stories. Don’t let this be you. Instead:
Ensure privacy. Avoid plugging proprietary brand language or embargoed info into unrestricted tools. Use “private” or enterprise versions like ChatGPT Teams.
Check your settings: Make sure you aren’t “improving the model for everyone.” While you’re at it, bookmark the data controls on your LLM of choice and update settings regularly. Updates happen, and settings can shift without warning.
Remember, if you’re working with clients, assume their trust in your process includes how you use AI behind the scenes.
2. Train Your AI Models
Want better output? Feed your AI assistant better inputs. LLMs don’t know your brand’s tone, competitive differentiators, or messaging pillars until you explicitly tell them. Give them the tools and feedback they need to improve in the moment.
Here’s how to do that well:
Train your AI model like a new team member. Upload brand voice guides, case studies, earned media highlights, and executive POVs. These “anchor documents” become your bowling lane bumpers—they won’t write for you, but they’ll keep your output out of the gutter.
Use built-in memory features to retain context. ChatGPT’s custom instructions and Claude’s long-term memory let you embed brand identity over time. I save each client’s key materials to their own folder and treat that as the persistent knowledge base I can draw from.
Prime every session. Even with memory turned on, I like to start each new conversation with a quick brand and audience refresher—just a few bullet points on voice, goals, and target persona.
PRO TIP: Keep a boilerplate prompt handy. Something like, “You are a communications strategist for [Brand Name]. The brand voice is [tone/attributes], and the target audience is [audience]. I’ll share some background materials now. Don’t do anything until I say go.”
When you do this consistently, a generalist model becomes a strategic comms co-pilot that actually gets your brand or client.
3. Train Better Prompters, Not Just Faster Writers
Most people type a topic into an LLM and copy-paste whatever comes out. But good communicators know the best content starts with the best questions and context.
Prompts are a creative and strategic skill. And when AI can churn out a thousand versions of “fine,” your edge comes from asking questions that don’t lead to generic answers. I’d argue that the ability to ask thoughtful, high-context questions—whether you're interviewing a subject matter expert for a ghostwriting assignment or collaborating with your favorite AI assistant—will be one of the defining skills for communicators moving forward.
Here’s how to raise the bar in prompting:
Build prompt libraries. Instead of starting from scratch every time, create and maintain a shared library of high-performing prompts that your team can repurpose, remix, and evolve. Think of them as internal assets that get updated over time. These are some sample prompts we love at Carve:
⭐️ Targeting: Here is this press release that explains the story we want to tell. List 10 top-tier reporters who would be a good target for an exclusive.
⭐️ Reframing: Here is one POV we’ve already put out on this topic. What other angles would you suggest we explore tied to the topic? Or what follow-up questions for our executive would help us dig deeper?
⭐️ Media Scanning: What are the five biggest trends in [INDUSTRY] that our brand should be commenting on based on the past content we’ve worked on together? Provide links to all sources.
⭐️ Look-a-likes: [PERSON] is an excellent fit for our influencer program. Can you share 10 others like her? They must be US-based and consistently active on LinkedIn.
⭐️ Insight Distillation: Can you share the three most salient takeaways from this webinar transcript?
Treat it like an interview. Don’t expect brilliance from a single input. The best outputs come from a back-and-forth conversation.
Encourage curiosity, not dependency. If the prompt is shallow, the output will be too. Use AI in PR to pressure-test ideas, explore counterpoints, and clarify thinking. One of my favorite go-to prompts when something sounds too buzzwordy: “How would this come across to someone skeptical—or totally unfamiliar with the space?” It’s a fast way to de-jargon and refocus on clarity and intent.
4. Edit Out the Average
Another defining skill for communications moving forward? Strong editing. Besides the baseline editorial questions tied to flow, word choice, and more, your job now includes spotting what feels average. If you're not stepping away from your drafts and pushing for differentiated angles and emotional precision, you're blending in with the noise.
As Katie Parrot said in Every: “LLMs will flatten your weirdness. And weirdness is what people remember.”
Here are some tips for keeping your weird:
Work in draft mode. There’s a nice split-screen in ChatGPT that allows you to keep prompting on the left and drafting on the right. It’s like writing, researching, and brainstorming all at once. If you did that with another person, it could be overwhelming. I always ask, “Let’s move to draft mode,” if it’s not offered.
Challenge the output. Ask: Is it actually saying something—or just summarizing what’s been said before? Would I click this? Would my audience?
Watch out for repetition. This is one of my biggest beefs with LLMs: they repeat ideas endlessly. Ruthlessly edit drafts for repetition. If you have subheads in a piece, make sure each section has a distinct point and you’re not saying the same thing from six paragraphs ago. I also haven’t found LLMs to be particularly good at responding to prompts like: “Remove the repetition from this paragraph.” They tend to look for words, and you want to focus on ideas instead to get to tighter writing.
Spot the patterns. Once you see the patterns of AI writing, you can’t unsee them, so get really good at spotting them. The em dash took the initial hit (I loved you first!), but there’s also the “it’s not just X, it’s Y” structure, the three single words in a row cadence, or the overuse of the word “sharper.” I could go on.
Move the bold idea up. You’re not obligated to follow the machine’s order of operations.
Push for story and specificity. Look for where you can add proof that makes it feel human. Insert stories and leave comments tied to specific places where more color is needed.
PRO TIP: If you include sources or links to help inform the piece, remind your AI assistant to keep them in the draft. AI often deletes them and, if you're not paying attention, they might disappear for good. Also, fact-check everything. AI lies to me daily.
5. Deep Research Your Targets
LLMs can be powerful tools for reporter targeting. Instead of relying on static databases or old media lists, use AI in PR to surface who to target and what they actually care about based on their recent work.
Summarize coverage. Drop a reporter’s last 5–10 headlines into your AI assistant and ask for the dominant themes, tones, and beats they cover most often. Do they love product-led angles? Do they favor founders or case studies?
Extract phrasing. Ask the LLM to identify common language or framing a journalist uses in their articles. Then, match that tone in your pitch.
Write with precision. Use AI to simulate how a pitch might land with the reporter. Try: “How would [Reporter Name] respond to this pitch?” or “Rewrite this email in a way that would appeal to [Reporter Name] based on their recent articles.”
Build beat maps. Ask your AI tool to cluster reporters covering a topic by niche. For example, fintech through a regulatory lens vs. fintech through a consumer trend lens. This helps you segment smarter.
PRO TIP: “Fact check everything” applies here, too. I’ve had moments where I found what I thought was a perfect target for a big initiative, and then realized they were laid off with the last restructure.
The era of sending out generic pitches is officially over. It should’ve been decades ago, but somehow we still end up with reporters on social media calling out PR pros for irrelevant pitches. There is no excuse anymore.
6. Learn How to Show Up in LLMs
LLMs don’t crawl the live web like Google. They pull from what’s been published, linked, and reinforced over time. That means if you want your brand to show up when someone asks ChatGPT, "Which vendors should I consider?" you need to be feeding the internet the right information consistently.
Axios’ Eleanor Hawkins said it like this: “Generative engines pull from what's visible, authoritative, and structured. If you're not showing up in those places, you're not showing up.”
Her post references a recent Muckrack study that analyzed over one million links cited in LLM AI responses. More than 95% of those links were from non-paid media, and a full 89% were from earned media.
For communications pros, this is great news because media coverage matters more than ever. But it’s worth noting what sort of coverage and content tracks so that you can pull levers to influence it. Enter the generative engine optimization (GEO) era. (You may have also heard it as AI search optimization or AIO, but the mess of acronyms is a whole other article).
Here are some tips to get you started:
Publish on high-authority sites. Outlets like Reuters, CNN, and Financial Times are often cited across LLMs. You can see some differences when you look across ChatGPT, Claude, and Gemini.
Target high-authority niche sites, too. Muckrack reports that “AI systems tend to select niche-appropriate outlets frequently when queried about specific industries.” They even go so far as to break out industries, including energy, government, retail, and technology. We see this, too, when we’re auditing our client’s citations. For example, for one digital experience provider, CMSWire and DemandGen Report show up often.
Be recent and consistent. The MuckRack report found that AI systems—especially OpenAI’s—strongly favor media coverage published within the last 12 months. (I can’t confirm the same thing about research studies.) That means outdated press hits, even if impressive, may never surface in AI recommendations. To maintain visibility, brands must create a steady drumbeat of new, high-authority earned content. This reinforces the need for a sustainable PR model built on a consistent stream of useful, trend-driven, and POV-driven stories that shape perception over time.
Make your content easy for humans and AI to read. On-page structure and readability are important to AI search optimization. Organize content with clear headlines, publish dates, and author names, and use Q&A format and TL/DRs when appropriate.
There’s also the backend to consider, and for this, I asked Holistic Marketing’s Krystine Monnett, since she works with us to keep our Carve website in top technical shape:
“A tool that is useful in helping AI Models understand your content is structured data utilizing schema markup. This allows important information like content type (event, press release, blog, etc) to be easily understood and include valuable attributes (author, publish date, location, etc). Your website platform may automatically deploy a level of structured data already, but you can check within your Google Search Console account or by testing URLs individually here.”
The good thing is that you may have already been doing a lot of this since SEO hinges on these best practices.
Keep in mind, anyone who says they’ve got GEO all figured out is overly confident. The truth is that this is changing every day along with AI. But we can spot the patterns now of what will matter for the near future. Your best bet is to find the industry experts you like and follow them so you learn (and apply what you learn) in the moment.
PRO TIP: One expert I love is SparkToro’s Rand Fishkin, and he said this in a recent video on his LinkedIn feed: “Your job is to do PR and content marketing across all platforms your audiences pay attention to.”
7. Audit Yourself When You Do Show Up
If customers, investors, analysts, and reporters are using tools like ChatGPT, Claude, and Perplexity to learn about your brand, you need to know what they’re seeing. This opens a new chapter for PR reporting where we’re not just measuring impressions, but influence on AI-powered perception.
Use AI in PR to track which brand messages, stories, or spokespersons appear in summaries and auto-generated content. Try:
Run brand audits regularly. Pick a few key prompts and run them at set intervals to see where your brand stands and how it might be shifting as coverage and owned content is published. Ask about your brand, but also about your executives. Assume the role of a potential customer and make sure you’re querying LLMs the way they do. And, yes, this might involve some audience research.
Create before-and-after snapshots. Use these AI search optimization audits to measure the impact of campaigns. What changed in your AI footprint after a major launch or media push? This is a new attribution signal.
Distill campaign impact. Try prompts like: “Summarize how [Brand] has been covered in the media over the last six months.” If the model reflects your desired themes, the messaging is landing. If not, it's an opportunity to refine.
Compare across platforms. LLMs are trained on different datasets. ChatGPT may reflect a different brand perception than Perplexity or Claude. Running the same query across tools can highlight gaps or wins in visibility.
LLMs are a new layer of influence and comms professionals are best equipped to work with it. Our work shapes what gets learned, repeated, and reinforced by the tools billions rely on. Now’s the time to bring critical thinking, lead with strategy, and own the narrative in the places AI reaches first. If we don’t raise the bar, AI will lower it for us.
