AI vs Human Pitches: What Journalists Prefer

Every PR agency is using AI now. Most won’t admit how much.

I’ll be transparent: at Presslei, we use AI tools every day. For research, for data analysis, for first drafts of campaign summaries. AI is embedded in our workflow and it makes us faster.

But there’s one place where the AI question gets genuinely complicated: the pitch itself. The email or LinkedIn message that lands in a journalist’s inbox and has about 8 seconds to earn their attention.

Should you let AI write your pitches? Can journalists tell the difference? Does it matter?

We decided to stop guessing and actually test it. Here’s what we found.

The Test: How We Set It Up

Over a four-week period earlier this year, we ran a controlled comparison across 600 pitches for three different PR campaigns. Each campaign had a genuine data story behind it — we weren’t testing fictional pitches, these were real campaigns going to real journalists.

For each campaign, we created three pitch variants:

• Variant A — Fully AI-generated. We gave Claude the campaign brief, the journalist’s name and beat, and their last 3 articles. The AI wrote the complete pitch, including subject line. No human editing.
• Variant B — AI-assisted. A human wrote the core pitch. We used AI to refine the subject line, tighten the opening sentence, and suggest personalization angles. The human made all final decisions.
• Variant C — Fully human-written. Written from scratch by a human who had read the journalist’s recent work. No AI involvement at any stage.

Each variant was sent to a roughly equal, randomized subset of journalists. Same campaign, same data, same story — just different pitch execution.

We tracked four metrics: open rate, response rate, positive response rate (journalist expressed interest), and sentiment of the response (coded as positive, neutral, or negative).

The Results

The numbers tell a clear story, but it’s not the one most people expect.

AI-assisted pitches slightly outperformed fully human pitches. The combination of human judgment and AI refinement produced the highest response and positive response rates. The AI caught things the human missed — a tighter subject line, a cleaner opening sentence, a more precise way to frame the data hook.

Fully AI-generated pitches significantly underperformed. Nearly half the response rate of the other two variants. And here’s the telling detail: the negative response rate was 5x higher. We got emails back that ranged from curt (“Please remove me from your list”) to explicitly calling out the pitch as AI-generated.

The open rate anomaly. AI-generated pitches had the lowest open rate, which is interesting because the subject lines were objectively well-crafted — they hit all the “best practices.” My theory: AI subject lines, despite being technically sound, have a sameness to them that experienced journalists recognise subconsciously. They look like every other pitch in the inbox.

Why Fully AI Pitches Fail

We did a qualitative review of the AI-generated pitches that got negative responses. Three patterns emerged:

1. The Personalization Was Correct But Hollow

AI is good at referencing a journalist’s recent work. It can name the article, summarise the topic, and connect it to your pitch. But the connection often feels mechanical.

Here’s a real example from our test (details changed):

AI-generated: “I noticed your recent piece on rising energy costs in the UK and thought you’d be interested in our latest data on household spending trends, which shows a 23% increase in utility-related financial stress.”

Human-written: “Your piece last week on energy bills made me think of something we’ve been seeing in our data — the families getting hit hardest aren’t who you’d expect. Middle-income households are cutting spending faster than low-income ones. Counterintuitive but the numbers are clear.”

The AI version is technically correct. It references the article, connects to the pitch, includes a data point. But it reads like a template with variables filled in. The human version reads like someone who actually read the article and had a reaction to it.

Journalists process hundreds of pitches. They’ve developed a finely tuned radar for authenticity. The AI pitch passes a checklist test. The human pitch passes a gut-feel test. The gut-feel test is the one that matters.

2. The Tone Was Professional But Lifeless

AI writes in a register that’s polished, grammatically perfect, and utterly forgettable. It avoids risk. It smooths out rough edges. It never says anything weird or surprising.

That’s exactly wrong for a pitch. The best pitches have personality. They have a voice. Sometimes they have a slightly unusual turn of phrase that makes the journalist pause and re-read.

One of our most successful human-written pitches started with: “This is going to sound strange, but we think we’ve found the most boring statistic in Britain — and it’s accidentally fascinating.” No AI would write that opening. It’s too risky, too informal, too human. It also got a 44% response rate.

AI-generated text occupies a narrow band of tone: professional-friendly. It can’t do dry wit, self-deprecation, genuine excitement, or controlled provocation. And those are precisely the tonal tools that make a pitch stand out.

3. The Structure Was Too Perfect

This was the subtlest issue. AI pitches had perfect structure: hook, context, data point, relevance to journalist, call to action. Every element in the right order, every paragraph the right length.

Real emails from real humans aren’t that tidy. They might have a P.S. that adds a tangential thought. They might start with the data instead of the context. They might be three sentences long because the story speaks for itself.

Perfection in structure is paradoxically a signal of inauthenticity. Journalists recognise it the same way you recognise a stock photo — everything is technically right, but it feels wrong.

Where AI Actually Helps

The test wasn’t all bad news for AI. The AI-assisted variant (Variant B) outperformed pure human writing, and the specific areas where AI added value are worth understanding.

Subject Line Optimisation

AI is genuinely good at tightening subject lines. Humans tend to write subject lines that are either too long or too vague. AI consistently produced shorter, punchier versions that included concrete data points.

Human draft: “New research on how UK household spending is changing in 2026”

AI-refined: “UK middle-income families cutting spending faster than low-income — new data”

The AI version is more specific, more surprising, and more likely to get opened. In our research on subject line performance, data-led subject lines outperform every other format. AI is excellent at identifying and foregrounding the data hook.

Research and Personalisation Prep

AI is a powerful research assistant. Before writing a pitch, we use AI to:

• Summarise a journalist’s last 5-10 articles
• Identify recurring themes in their coverage
• Find connections between their beat and our campaign
• Draft bullet points of potential angles

This prep work takes 15-20 minutes per journalist when done manually. AI does it in 30 seconds. The human then uses that research to write a genuinely personalised pitch — but AI did the homework.

Catching Errors and Tightening Copy

AI is an excellent editor. After a human writes a pitch, running it through AI to check for:

• Typos or grammatical mistakes
• Sentences that could be shorter
• Claims that need qualification
• Tone inconsistencies

This produces a cleaner final product without sacrificing the human voice.

Scaling Variations

When you’re pitching 50 journalists on the same campaign, you need 50 slightly different pitches. AI can help generate variations — different opening hooks, different ways to frame the same data, different angles for different beats — that a human then reviews and personalises.

This is the legitimate use case for AI in pitching: not writing the pitch, but giving the human more material to work with.

The Journalist Perspective: What They Told Us

After the test, we surveyed 40 journalists who had received pitches across all three variants (without telling them which was which). We asked two questions: “Can you usually tell when a pitch is AI-generated?” and “Does it affect your likelihood of responding?”

87% said they can usually tell. The most commonly cited signals:

• “It reads like a template” (mentioned by 68%)
• “Too polished, no personality” (52%)
• “The personalization feels forced” (44%)
• “Every sentence is the same length” (31%)
• “It uses phrases like ‘I came across your insightful piece’ — nobody talks like that” (28%)

72% said it negatively affects their response. The reasoning was consistent: if you can’t be bothered to write a personal email, why should I take time to read it? AI pitches signal that the sender doesn’t value the journalist’s time or their relationship.

One journalist put it bluntly: “I get 200 emails a day. If I can tell a robot wrote yours, I delete it instantly. Not because I hate AI — because it tells me you sent this to 500 people and I’m not special. And if I’m not special, the story probably isn’t either.”

That last line is the key insight. An AI-generated pitch doesn’t just fail on execution — it undermines the credibility of the story itself. If the pitch feels mass-produced, the journalist assumes the campaign is mass-produced too.

The Personalization Gap

The single biggest difference between AI and human pitches is what I’ve started calling the personalization gap.

AI can personalise at Level 1 and Level 2 in our response rate framework — using the journalist’s name and referencing their beat or recent article. It does this efficiently and at scale.

But Level 3 personalisation — the kind that connects your pitch to the journalist’s specific worldview, interests, and patterns — requires understanding that AI doesn’t have. It requires having actually read their work, not just processed it. It requires knowing that this journalist is skeptical of government statistics, or that she always leads with human stories, or that he’s been building toward a series on cost-of-living and your data is the missing piece.

That gap between “I processed your article” and “I understood your article” is where pitches succeed or fail. AI can close the gap partially. It can’t eliminate it.

A Framework: When to Use AI, When Not To

Based on our testing, here’s the framework we now use at Presslei:

Use AI for:

• Research and journalist profiling (always)
• Subject line generation and refinement (always)
• First-draft variations for A/B testing (usually)
• Copy editing and tightening (always)
• Data analysis and finding the story hook (always)
• Translating pitches into other languages (with human review)

Don’t use AI for:

• Writing the final pitch that goes to a journalist (never)
• The personalisation paragraph (never)
• Follow-up messages (never — these need to feel human)
• Responses to journalist questions (never)

The rule is simple: AI does the prep work. Humans do the relationship work. Every interaction a journalist has with you should feel like it came from a person who cares about the story and respects their time. AI can help you be more prepared for that interaction. It can’t have the interaction for you.

What This Means for PR in 2026

The AI-in-PR conversation is often framed as a binary: either AI is going to replace PR professionals or it’s useless. Both positions are wrong.

AI is going to replace PR professionals who were already doing mediocre work. If your outreach strategy was “blast 1,000 journalists with the same press release,” AI does that faster and cheaper. But that strategy didn’t work when humans did it either.

AI is not going to replace PR professionals who build genuine journalist relationships, create original stories, and pitch with specificity and care. It’s going to make them faster, better-researched, and more productive.

The agencies that figure out the right human-AI split will win. Based on our data, that split is roughly: AI handles 60% of the work (research, analysis, drafting, editing) and humans handle 40% (strategy, relationship, final pitch, follow-up). But that 40% is the 40% that actually determines outcomes.

If you’re a journalist reading this: yes, we know you can tell. That’s why we don’t send AI-written pitches. The ones you get from us are written by a human who read your last five articles. The AI just helped us find those articles faster.

If you’re a PR professional reading this: use AI aggressively for everything except the moment of human contact. That moment — the pitch, the follow-up, the relationship — is the only thing that matters. Automate everything around it. Protect the moment itself.

And if you’re curious about the data behind effective pitching or how to build an AI-enhanced PR workflow, we’ve written about both.

Frequently Asked Questions

Will journalists blacklist me if they catch an AI-generated pitch?

Probably not a formal blacklist, but the damage is real. Most journalists told us they simply delete and move on. The risk isn’t a dramatic confrontation — it’s quiet invisibility. Your future pitches get pattern-matched to “that person who sends robot emails” and deleted before they’re opened. Rebuilding after that is harder than getting it right the first time. If you’re tempted to fully automate your pitching, ask yourself: is saving 20 minutes per pitch worth risking a journalist relationship that took months to build?

How can I use AI in PR without it backfiring?

Keep AI in the back office, not the front office. Use it for research (journalist profiling, article summaries, beat analysis), for data analysis (finding story hooks in your numbers), for drafting (subject line options, structural outlines), and for editing (tightening copy, catching errors). Then write the actual pitch yourself, in your own voice, with specific references that show you did the work. The journalist should never interact with AI output. They should interact with a better-prepared human.

Is this going to change as AI gets better?

Probably, but not in the way most people expect. AI will get better at mimicking human writing. But journalists will also get better at detecting it — it’s an arms race. More importantly, the underlying issue isn’t detection — it’s trust. A pitch works because a journalist believes a real person read their work, understood their beat, and thought of them specifically. That belief requires genuine human judgment, regardless of how good AI text generation becomes. The fundamental dynamics of human relationship and trust don’t change because the tools improve.

Salva Jovells is the founder of Presslei, a reactive PR agency based in Zurich. He’s spent 12 years in ecommerce SEO and has analyzed 5,272 media placements to build a data-driven approach to earning press coverage.

About the Author

Salvador Jovells

Founder of Presslei. 12+ years in ecommerce SEO across international markets. After a decade of link buying for Hockerty and Sumissura, I reverse-engineered 5,272 earned media placements and founded a reactive PR agency that builds authority through data-driven stories journalists actually want to publish. Based in Zurich.