Why PropeLLM Works for PR Better Than Generic AI Models

I've spent the better part of two decades in public relations. I've sat in war rooms during product launches, fielded 2 a.m. crisis calls, watched the Rolodex evolve into the CRM, and seen more pitch decks than I'd care to count. In that time, I've also watched our industry get its collective head turned by every new technology that promised to "transform communications." Most of them didn't. A few did.

AI is one of the ones that actually will.

But I want to talk about a distinction that I think is getting lost in all the noise, a distinction that, frankly, I see PR leaders fumble in meetings every week. It's the difference between an AI that understands language and an AI that understands public relations. Those sound like the same thing. They're not. And the gap between them is where careers and campaigns are being quietly won and lost right now.

So let's talk about it honestly.

ChatGPT Is Brilliant. That Doesn't Mean It Understands PR.

Let me say this clearly up front: ChatGPT, Claude, Gemini, and the other large general-purpose models are extraordinary pieces of technology. I use them. My team uses them. If you're not using them at all, you're leaving real value on the table. They draft, summarize, brainstorm, translate, and explain at a level that would have seemed like science fiction five years ago.

But here's the thing I keep coming back to in conversations with comms leaders: PR is not a content-generation problem.

Writing is a part of PR. It is not PR.

If writing were the job, every English major with a thesaurus would be a great publicist. We all know that's not how it works. The press release is the easy part. Honestly, and any senior practitioner reading this is nodding right now, drafting a competent release has never been the hard part of our work. Nine times out of ten, the release is the thing the client argues about for three weeks while the actual PR work happens around it.

The hard part is everything else.

The hard part is knowing that the reporter you want to pitch just left her beat last month. It's knowing that this particular tech editor hates exclusives but loves data. It's knowing that Tuesday at 10:30 a.m. ET is dead air for your sector because of a competing earnings call. It's knowing that the angle your CEO wants to lead with is the same angle a competitor pitched six weeks ago and got crickets.

That kind of knowledge isn't on the public internet. And that means it isn't in the training data of a general-purpose model. It can't be. The model has no way to learn it.

What a Generic AI Actually Sees (and Doesn't)

A general-purpose AI has read enormous amounts of text. News articles, books, web pages, forum posts, technical documents. From all of that, it has learned the shape of public relations. It knows what a press release looks like. It knows the structure of a media pitch. It can mimic the tone of a thoughtful boilerplate or an executive bio.

What it has not seen is the back end of any of it.

It has never seen which of those pitches got opened. Which got replies. Which got coverage. Which got the dreaded "please remove me from this list." It hasn't seen the journalist who consistently responds to a personal subject line but ignores anything with the word "exclusive" in it. It hasn't seen the fact that the freelance writer with 4,000 LinkedIn followers actually drives more qualified traffic for B2B SaaS clients than the marquee outlet name everyone is chasing.

The signals that determine whether PR actually works, open rates, reply rates, coverage outcomes, follow-up patterns, beat shifts, declined pitches, editorial calendars, conference timing, embargo behavior, journalist preferences, those signals live in the operational data of comms teams. Not in published articles.

This is the distinction I want every PR leader to internalize: the public internet is downstream of PR work. It is the output. The interesting data is upstream.

A generic AI can write you a pitch. It cannot tell you what happens after you send it.

A generic AI can generate a media list. It cannot tell you which journalists on that list are actually likely to engage with your story this quarter.

A generic AI can rewrite a press release for the fifth time. It cannot tell you why the first four versions were getting ignored.

This isn't a flaw in those models. It's a scope question. They were built to be generalists. You wouldn't ask a brilliant generalist physician to read a complex cardiac MRI, not because they're not smart, but because that's not what they were trained for. You'd want a cardiologist. The radiologist who reads 200 of those scans a week sees patterns the GP simply cannot see, no matter how intelligent they are.

That's what specialized AI does for PR.

Why Propel Built PropeLLM

Propel has spent years building software exclusively for public relations professionals. Not communications-adjacent. Not marketing-with-a-PR-module. PR. We've worked with hundreds of communications teams across in-house and agency settings, and in doing so we've had a front-row seat to how modern PR actually gets done, what works, what doesn't, where the friction is, where the wins come from.

When we trained PropeLLM, we didn't start from scratch with public web data. We started with what we had been quietly compiling and learning from for years:

• 25 million PR pitches, the actual outreach communications that PR professionals send to journalists, with the language patterns, structures, subject lines, and approaches that real practitioners use in the field.
• 2 billion news articles, the published record of what journalists actually write, across beats, outlets, geographies, and timeframes. Not just headlines. Bylines, framings, sourcing patterns, story arcs.
• 500,000 journalist profiles beats, recent coverage, contact preferences, the things that make a reporter a real person rather than a name on a list.
• Years of PR workflow and engagement intelligence, what happens after a pitch goes out, how campaigns unfold over time, where outreach succeeds and where it stalls.

This dataset is proprietary, anonymized, and PR-specific. It exists nowhere else, because no one else has been building the underlying infrastructure for PR teams at this scale for this long. A general-purpose AI does not have access to this kind of data, and it is not going to get it by reading more of the open web. The information simply isn't there.

That is what we mean when we say PropeLLM is purpose-built. It is not a generic model with a PR system prompt taped to the front of it. It is a model whose understanding of the world is shaped, from the ground up, by the actual mechanics of how PR works.

Where Generic AI Falls Short in Real PR Workflows

Let me get concrete. Here are the places where I have personally watched generic AI break down when comms teams try to use it for real work, and where domain-specific intelligence makes a clear difference.

Identifying the right journalists for a story. A generic AI can pull together a list of reporters who seem relevant based on what's published. But "wrote about fintech once in 2022" is not the same as "actively covers fintech this quarter and replies to pitches under 150 words from founders, not agencies." Targeting in PR is about velocity and recency and fit, not keyword matching. A model that's only seen finished articles cannot understand outreach receptivity.

Understanding what makes something newsworthy. This is one of the most misunderstood skills in our craft. Newsworthiness isn't a feature of a story, it's a relationship between a story, a moment, a beat, and an audience. A generic model trained on news will tell you what has been newsworthy historically. It cannot tell you what is likely to land now, this week, against the current news cycle, for the specific reporter you're approaching.

Suggesting angles journalists actually care about. Every veteran publicist has had the experience of pitching the angle the client demanded and watching it die, then pitching a side angle and getting a feature. That instinct comes from pattern recognition across hundreds of outreach attempts. It's exactly the kind of pattern that lives in PR engagement data, not in published copy.

Personalizing pitches. Real personalization is not "Hi [FirstName], I saw your recent article on [Topic]." Real personalization is knowing that this reporter publishes a weekly newsletter on Thursdays, prefers data-driven stories over thought leadership, and just moved from a national outlet to a vertical publication where the editorial appetite is different. Generic AI cannot personalize at that level, because it doesn't know any of that.

Building media lists that work. Anyone can generate a list of 200 names. The art is generating a list of 30 names where 12 of them might actually engage. Quality of list is downstream of quality of signal, and the signal isn't public.

Developing campaign strategy. A campaign is not a press release plus a list. It's a sequence of moves played over weeks, against a competitive backdrop, with multiple stakeholders, surprise events, and shifting media priorities. Strategic PR thinking requires understanding the operational rhythms of media, not just the language of it.

Competitive media analysis. Who is winning share of voice in your category? Which competitors are quietly building relationships with the journalists you should know? Which outlets are over-indexed for your competitor and under-indexed for you? These are the questions a comms leader actually asks. They require structured, comparative analysis grounded in real coverage and outreach data.

Crisis communications planning. In a real crisis, you have minutes, not hours. You need to know which reporters are likely to write about this within 24 hours, which outlets have covered similar issues before, what tone they took, and how the conversation evolved. A generalist model can give you a framework. A specialist can help you act.

Story ideation and positioning. Where does this announcement fit in the broader narrative your sector is having this quarter? What's the white space? What angles are over-pitched right now and which are wide open? That's a market-aware question, and the market is the press.

In each of these, a generic AI can produce something reasonable and well-written. It's the judgment that lags, because judgment in PR is built on signals the open internet doesn't carry.

What Generic AI Doesn't See

‍

I want to spend a moment on this, because I think it's the most important point in this whole piece.

PR success is shaped by context that is not, and cannot be, present in public internet content. Specifically:

• Journalist preferences. What length of pitch does this reporter actually respond to? What format? What time of day? Do they prefer being approached cold or via a warm intro? Are they on a sabbatical right now?
• Outreach history. Have we (or our client) pitched this person before? How did it go? Are we in a cooling-off period? Are we already in conversation about a different story?
• Media engagement patterns. Across millions of interactions, what kinds of subject lines actually get opened? What pitch structures lead to replies? What follow-up cadence converts versus annoys?
• Pitch performance trends. What's working this month across our industry? The PR climate shifts constantly, what landed in Q1 will not necessarily land in Q3.
• Coverage outcomes. Did the pitch actually produce coverage? What kind? Was it the framing the brand wanted? Did it drive the downstream business outcomes?
• Industry-specific communications knowledge. Healthcare PR is different from fintech PR is different from consumer lifestyle PR. Each has its own outlets, rhythms, regulatory considerations, and reporter cultures.

These signals exist within Propel's ecosystem. They've been observed, measured, and learned from across years of real PR activity. PropeLLM's understanding of public relations is shaped by this data.

A generic model, by definition, cannot see any of it. It is reading the finished newspaper. We are watching the newsroom.

The Future of AI Isn't Bigger Models. It's Smarter Ones.

There's an idea that has dominated the AI conversation for the last several years: that bigger is better. More parameters, more data, more compute, more capability. And to be fair, that has been broadly true at the frontier. The big general-purpose models have gotten remarkable.

But I think we're entering a different phase, and you can already see it across professional industries.

In legal, firms are increasingly relying on AI systems trained specifically on case law, contracts, and litigation patterns. A generalist model can summarize a contract. A legal-specific model can flag the unusual indemnification clause and tell you why it matters in this jurisdiction.

In healthcare, AI tools trained on clinical notes, imaging, and biomedical literature are outperforming general models on diagnostic and decision-support tasks. Not because the general models are bad, but because medicine is a domain with its own language, its own evidence standards, and its own consequences.

In financial services, specialist models trained on filings, market data, and transaction patterns are doing work that a general chatbot fundamentally can't, risk modeling, fraud detection, regulatory analysis.

And in PR, the same pattern is emerging. The most effective AI for our work will not be the largest model. It will be the model that has actually learned how PR works, what gets pitched, what gets covered, what gets ignored, what gets results.

Specialization isn't a step backward from general intelligence. It's a step forward into utility. The general models will keep getting better at being general. The specialized models will keep getting better at being useful. PR teams will benefit most from a stack that includes both, and from being thoughtful about which tool to reach for when.

A Final Thought, From One Practitioner to Another

I want to be clear about something. I'm not making the case that PropeLLM is universally smarter or universally better than the big general models. It isn't. ChatGPT will continue to be incredible at writing your kid's birthday speech, summarizing a long PDF, and helping you brainstorm names for your new puppy. Use it. I do.

The case I'm making is narrower and, I think, more honest.

If your job is public relations, if you are responsible for landing coverage, building relationships with journalists, managing brand reputation, and producing measurable outcomes for stakeholders, then you deserve tools that were built for that work. Not tools that can approximate that work because they can write fluently. Tools that understand the work.

A generic AI can write a pitch. It cannot tell you what happens after it's sent. It can generate a media list. It cannot tell you who is actually likely to engage. It can rewrite a release a dozen times. It cannot tell you why the original was being ignored.

PropeLLM exists because we believe communications professionals deserve more than fluent text. They deserve intelligence that has actually seen the workflow they live in every day, the outreach, the relationships, the timing, the patterns, the outcomes. That intelligence comes from training on the right data, and the right data for PR does not live on the open internet. It lives in the practice of PR itself.

So the question facing comms teams in 2026 is no longer whether to use AI. That question is settled. The real question, and it's a more interesting one, is this:

Do you want AI that merely understands language? Or AI that understands public relations?

That's the choice. And I think, once you've spent enough years in this work, the answer becomes pretty obvious.