A features editor scrolls through 180 unread pitches before her second coffee. Three of them open with the same tidy compliment about her recent piece. Two use the word "resonate." One promises a story that will unpack a market she covers weekly.

She archives all three inside twenty seconds. None of them were badly written. That was the problem.

The tells have shifted in 2026. Editors, journalists, and recruiters are no longer squinting for stray em dashes. They've learned the rhythm of a generated draft, and they learned it fast. If you send outreach for a living, the small habits that used to read as "polished" are now reading as "processed." Here's what that looks like across the four inboxes it matters most in, and the small edits that put a human back on the page.

The Media Pitch That Reads Like Every Other Media Pitch

Journalists are the most saturated audience in this whole conversation, and they're getting louder about it. More than half of reporters now oppose AI-drafted pitches, and the objection isn't philosophical. The drafts all sound the same, and once you've seen the pattern you can't unsee it.

The tells are consistent. A flattering first line about the reporter's beat that never quotes a specific piece. A middle paragraph heavy on "landscape," "leverage," and "in an era where." A closing sentence offering an executive who is "uniquely positioned" to comment.

Any one of those, on its own, is forgivable. Stacked in the same email, they read as an assembly line.

The edit is smaller than people think. Name the actual headline you read, not the beat. Cut every adjective you didn't need to say out loud. Offer one concrete data point or one specific anecdote the reporter couldn't get from a press release.

A pitch with one real thing in it beats a pitch with five polished ones.

The Cover Letter That Says Everything and Proves Nothing

Recruiters were the first group to build a shared vocabulary for this. Ask any in-house talent lead what an AI cover letter looks like and you'll get a version of the same list: flawless grammar, generic enthusiasm, achievements that don't tie to a number, and a closing paragraph that could belong to any candidate for any role. Reed's recruiters have documented the pattern in detail, from repetitive sentence structures to CV-and-letter disconnects where the two documents describe subtly different people.

There's another tell that slips past most writers: the letter that answers a question the job ad didn't ask. Generated drafts tend to reach for universal virtues, like collaboration, curiosity, growth mindset, instead of the specific problem the hiring team wrote the ad to solve. A reader who wrote the ad notices immediately.

Fix it by writing the letter against the ad, line by line. Which requirement does paragraph one answer? Which project on your CV is the proof?

If a sentence doesn't tie to something a hiring manager asked for, cut it. The letter gets shorter and, oddly, more distinctive.

The Recruiter's Cold Email That Never Gets Opened

Outbound recruiters are on the other side of the same problem, and the numbers are unkind. The same AI drafts performed better on LinkedIn than in email, which tells you what candidates are pattern-matching on.

The edits that move that number are unglamorous:

• Open on the person, not the pitch. A single sentence that shows you read their last two roles beats any subject-line trick.
• Name the role in plain words. "Senior backend engineer, payments team" outperforms "exciting opportunity with a market leader" every time.
• Cut the second paragraph. The generated middle, all culture and mission and growth, is what candidates skim past. Ask the question instead.

The Follow-Up That Sounds Suspiciously Well-Written

Follow-ups are where AI drafts are getting caught most often now, because the register jumps. The first email sounds like a person; the follow-up sounds like a press release. Editors clock the shift instantly. So do hiring managers who exchanged two casual notes with a candidate and then received a third one written in corporate-brochure English.

Consistency is the whole game here. If the original email used contractions, the follow-up uses contractions. If the first note was three sentences, the second isn't nine. Reference something specific from the earlier exchange, like a date they mentioned or a question they asked, and keep it in the same voice you used the first time.

Use AI Earlier in the Process, Not Later

None of this is an argument against using AI. It's an argument for using it further from the send button. The teams whose outreach still reads human in 2026 tend to use models for research, angle-testing, and first-draft scaffolding, then rewrite in their own voice before anything goes out. A practical breakdown of using using AI for PR without sounding like a robot without sounding like a robot covers the specific places automation helps and the places it hurts.

One approach that tends to work: let AI help with much of what a reader will rarely see, and write the parts they will yourself. Media lists, background research, meeting prep, subject-line variants, fine. The opening line of the pitch, the specific compliment, the ask itself, those tend to stay in your hands.

The reader can often tell which is which. That's the whole tell.