How to Humanize an AI-Generated Press Release for Real Journalists

You can generate a press release in about eight seconds now. The problem is that the journalist on the other end can tell in about two. AI drafts a clean skeleton, but it fills that skeleton with the exact phrases that make a reporter’s eyes glaze over, and no amount of polish fixes a release that has nothing real inside it.

So this isn’t a guide to sneaking AI copy past anyone. It’s a guide to taking a fast first draft and turning it into something a working journalist would actually open, read, and use.

Key takeaways

• A press release that reads like a template gets deleted, not because a detector flagged it but because a busy reporter can smell filler in two seconds.
• Humanizing an AI draft means adding the things AI can’t invent: a real quote from a real person, a specific number, and a reason this matters now.
• Cut the throat-clearing. AI loves opening with “In today’s fast-paced market” — journalists hit delete on that line.
• Reporters scan for the news hook in the first two sentences. If your lead buries it under adjectives, the rest never gets read.
• Run your final draft through a detection check the way you’d reread it aloud: not to game a tool, but to catch the robotic patterns a tired editor would also catch.

Why AI press releases fail with reporters specifically

Most advice about AI-sounding text is aimed at students worried about a grade. Journalists are a different audience with a different filter. They don’t run your pitch through Turnitin. They run it through experience.

A reporter at a mid-size outlet might wade through a hundred-plus pitches before lunch. They’ve internalized what a real story looks like, and they spot PR boilerplate even faster. AI-generated copy tends to hit every boilerplate marker at once: the grand opening claim, the symmetrical three-point structure, the quote that sounds like it was approved by a committee, and zero specific numbers. It’s not that the writing is bad. It’s that it’s frictionless in a way that signals nobody really stood behind it.

Worth knowing: detectors and editors are reacting to the same underlying thing. Tools measure statistical flatness — low perplexity and burstiness — while editors just call it “generic.” If you want the math behind why flat writing reads as machine-made, we broke down perplexity and burstiness separately. The practical upshot is the same either way: vary your sentences and get specific.

The five edits that matter most

You don’t need to rewrite the whole thing. You need to fix the parts AI predictably gets wrong.

1. Kill the throat-clearing opener

AI almost always opens with a windup. “In an era of rapid technological change…” or “As businesses increasingly seek…” Delete it. The first sentence should state the news. If a stranger read only your first line, would they know what happened? If not, rewrite until they would.

2. Replace every quote with something a human would actually say

This is the big one. AI writes executive quotes that restate the press release in corporate voice: “We are excited to announce this milestone as we continue our commitment to innovation.” Nobody talks like that. Real quotes have an opinion, a number, or a small concrete detail.

Don’t paraphrase a generated quote into a slightly better generated quote. Call the person. Ask them one question. Use what they actually say.

3. Add numbers that aren’t round

AI gravitates to “thousands of customers” and “significant growth.” Reporters want a count and a percentage they can check, like “up roughly a third over the prior quarter” backed by a real figure. Specific, slightly odd numbers read as true because they sound like they came from a spreadsheet, not a brainstorm. Only use real ones — making up a precise stat to sound credible is the fast track to a correction and a burned source relationship.

4. Break the symmetry

AI loves three balanced clauses and parallel structure that never lets up. Real writing is lumpy. Some sentences are long and wind through a couple of ideas. Some are short. Read your draft out loud; wherever it sounds like a metronome, break the rhythm.

5. Answer “why now?”

Every real story has a reason it’s happening today and not last month. A funding round closed. A regulation changed. A season started. AI drafts almost never include this because the model doesn’t know your timing. Add one sentence that gives the reporter their news peg.

A quick before-and-after

Here’s a made-up example so you can see the gap. All the names and numbers below are invented for illustration — they aren’t real companies or stats.

> “In today’s rapidly evolving educational landscape, LearnLoop is thrilled to announce the launch of its innovative new platform designed to revolutionize how students engage with learning materials and unlock their full potential.”

That’s 40 words and says almost nothing. No news, no number, no reason to care today. Now the humanized version:

> “LearnLoop, a Boston study-tools startup, raised $4.2 million this week to expand a flashcard app that 60 high schools quietly piloted last fall. The round was led by Reach Capital and closes just as districts finalize next year’s software budgets.”

Same facts a release would carry, but now there’s a dollar figure, a named investor, a count of pilot schools, and a timing hook. A reporter can see the story. In the first version, a reporter just sees a press release.

And the quote, generated:

> “We are committed to empowering every learner on their educational journey,” said CEO Dana Okafor.

Versus what you’d get from a thirty-second phone call:

> “Teachers kept telling us the same thing — kids make flashcards once and never open them again,” said Okafor. “So we built the app to nag them. That’s it. It just nags.”

The second one sounds like a person because a person said it.

Where an AI humanizer fits — and where it doesn’t

A humanizing tool is genuinely useful for one job: smoothing out the mechanical patterns AI leaves behind. The repetitive cadence, the over-tidy transitions, the hedging phrases. Those are the things that make copy feel manufactured even after you’ve added your facts.

What a tool can’t do is the part that actually wins the pitch. It can’t call your CEO for a real quote. It can’t know that your funding round closed on a Tuesday or that a regulation shifts next month. Treat the humanizer as the last 20% — the proofreading-and-polish layer — not the source of the story. The order matters: facts and quotes first, then smooth the prose.

If you want to dig deeper, there’s more on the writing-craft side of what makes prose read as human, from sentence variety to specificity. And if you’re weighing how often you’ll run drafts through a checker, what our plans cover lays out the limits.

A simple workflow you can reuse

Here’s a repeatable loop for any release:

1. Draft fast with AI. Get the structure down. Don’t fuss over wording yet.
2. Strip the openers and superlatives. First pass is pure deletion. Cut “thrilled,” “revolutionary,” “game-changing,” and any sentence that doesn’t carry a fact.
3. Inject the real material. Drop in actual numbers, the news peg, and a genuine quote from a real person.
4. Break the rhythm. Read aloud. Vary sentence length on purpose.
5. Run a detection check as a mirror. Not to beat a tool, but to surface the sentences a human editor would also find flat, then fix those specifically.

That last step is worth reframing. A detector isn’t a gatekeeper here the way it is in a classroom; PR wires don’t usually screen for AI. It’s a proofreading instrument. The sentences it flags as machine-like are, more often than not, the same ones a skimming reporter would skip. (And remember detectors output a probability, not proof — a 2023 study by Liang and colleagues even found they can be biased against non-native English writers, so treat the score as a hint, never a verdict.)

Frequently asked questions

Can journalists actually tell when a press release was written by AI? Often, yes, though usually not by running it through a detector. They notice the tells: vague superlatives with no number behind them, a quote that sounds like a mission statement instead of a person talking, and openings like “In an increasingly competitive landscape.” A reporter reads dozens of pitches a day, so they pattern-match fast. The fix isn’t tricking them, it’s giving them the concrete, quotable material they actually need.

Is it unethical to use AI to write a press release? No. Using AI to draft a first version is no different from using a template or a junior staffer’s rough copy. The ethics question is about accuracy, not authorship. As long as every fact, name, title, and quote in the final release is true and approved by the person quoted, how you generated the first draft doesn’t matter. Problems start only if you publish fabricated quotes or stats — a journalism problem regardless of who or what wrote them.

Will an AI detector flag my press release and hurt my pitch? Probably not in the way people fear. Reporter inboxes don’t run AI detectors as a gate the way some schools do. The bigger risk is human: an editor who finds your copy generic and moves on. That said, detectors are a useful mirror. They flag the same flat, predictable prose that bores a human reader.

What’s the single biggest tell that a press release was AI-generated? Quotes that no human would say out loud. AI tends to write executive quotes that just restate the headline in corporate language. Real people say specific, slightly messier things. Replace every generated quote with an actual line from the actual person, even if you have to call them to get it.

The bottom line

AI gets you a draft in seconds, but a draft isn’t a story. The work that earns coverage — the real quote, the odd-but-true number, the reason this matters today — is still yours to do. Use AI for speed, do the human edits that give a reporter something to grab onto, then polish the rhythm last.

When your facts are locked and you want to catch the leftover machine cadence before you hit send, paste the draft into PaperBleach and fix the flat sentences one at a time.