AI-written pitches in 2026: the operator cut

Contributors

1. Josiah Roche, Fractional CMO at JRR Marketing

   About 70,80% of the cold outreach landing in inboxes now reads at least partly AI-written, and most of it gets ignored for the same reason: it sounds like it was written for a category, not a person. The only AI-assisted pitches I've seen earn replies are the ones where AI did the prep work, not the final voice. ChatGPT is useful for turning interview notes, product docs, or a source brief into a rough first pass, but the sendable version still needs a human rewrite of the opener, angle, proof, and why-now hook. In practice, a fully manual pitch might get a reply rate around 8,12%, an AI-drafted but lightly edited one often drops to 2,4%, and an AI-assisted draft that's heavily rewritten can sit much closer to manual. One workflow that's held up is ChatGPT for structure, Perplexity for quick source checking, then a manual edit that cuts the email down by 33%. The parts most worth rewriting are the first two sentences and any line that sounds flattering in a vague way. A B2B software pitch sent this way for a survey story got replies from roughly 9% of targets, versus about 3% from an earlier version that kept the generic AI phrasing and broad trend claims. An AI-sounding draft also cost a placement in one case because it used the same stale pattern editors see all day: I thought this would be of interest, given your audience, and a list of talking points with no specific news hook. The tells are pretty consistent on inbound pitches. Over-explaining the obvious, praise that could fit any publication, fake specificity, odd em dashes everywhere, and subject lines that sound polished but say nothing. If a pitch can swap the outlet name, reporter name, and company name without changing the body, it's usually dead on arrival. The ones that get through tend to reference one recent article, bring one clear data point or contrarian view, and sound like a person who knows why this journalist, on this week, should care.

2. Runbo Li, CEO at Magic Hour AI

   Every pitch I send starts as an AI draft. Every single one. But the ones that land have something no model can generate on its own: a detail so specific it proves I actually care about the person reading it. Here's what I mean. Earlier this year I was pitching a creator partnership. Claude wrote my first draft in 30 seconds. It was clean, professional, and completely forgettable. So I rewrote the opening with a reference to a specific video they'd posted two weeks prior, named the exact transition they used, and explained why our tool would let them do that effect in one click instead of four hours in After Effects. Got a reply in 11 minutes. My workflow: I use Claude to generate structure and first-pass language, then I gut the first two sentences and the last sentence every time. Those are the only parts people actually read in a cold email. The middle can be AI-assisted. The open and close have to be unmistakably human. Response rates with this method run about 35-40% on cold outreach. When I've sent lightly edited AI drafts without that rewrite pass, I'm at maybe 8-12%. That's the difference between a tool and a crutch. On the receiving end, I get dozens of inbound pitches weekly. What gives AI away instantly: the pitch could be sent to literally anyone. It references your company's mission without naming the mission. It uses phrases like I'd love to explore synergies or I believe there's a natural alignment. No human who has actually looked at Magic Hour would write that. They'd say I saw your NBA edit or your face swap template is blowing up on TikTok. Specificity is the new authenticity. The other tell: length. AI drafts are almost always too long. A real person who respects your time keeps it under six sentences. If I see a pitch that's three paragraphs of context before the ask, it's getting archived unread. One more thing. I had an AI-drafted response to a journalist that used the phrase at the intersection of and I almost sent it. Caught it last second. That phrase is the scarlet letter of AI-generated content right now. The journalist later told me they auto-filter any pitch containing it. The rule is simple: AI writes the bones, you write the blood.

3. Raj Baruah, Co Founder at VoiceAIWrapper

   I use AI to draft cold emails and source-request answers, and I receive a lot of them on the other side as well. The pattern that consistently gets through (and the pattern that gets filtered) has stabilized enough in 2026 to write down. What earns a reply: AI-drafted, heavily edited, with a specific push-back line at the end. My workflow is Claude generates the first pass with the recipient's recent content as context, I edit at least 40 percent of every sentence, and I append a closing line that names something I would push back on or challenge in their published thinking. The push-back line is the part that proves a human read their work. Reply rate on this workflow runs roughly 3x what I got with fully manual templated outreach. What gets filtered or ignored: the unedited AI signature. Tells include opening with I'm reaching out to discuss, I hope this email finds you well, bullet lists in the body of a cold email (humans rarely format that way under length pressure), em dashes paired with vague verbs, and the dead giveaway, a closing line that summarizes what was just said. Real humans do not summarize themselves at the end of a four-paragraph email. AI does. A specific cost I paid early. I sent a journalist a Featured-style response that was AI-drafted and lightly edited. They published a story two weeks later citing five other sources, not me. Three months later they replied to a different pitch and mentioned the earlier one read too clean. The placement I had been chasing went to people whose drafts had more friction in them, not less. What I do now on inbound. When I screen pitches for our company, the giveaways are structural. Identical paragraph length across three paragraphs. A LinkedIn profile that says Founder & CEO but the email signature lists three accomplishments unrelated to the pitch. The compliment in paragraph one that is too specific to be improvised but too generic to mean anything. The line that holds up in 2026: AI gets you to a draft. Edits get you to a reply. Specificity gets you to a citation.

4. Callum Gracie, Founder at Otto Media

   I use AI heavily for pitches, cold emails and source-request answers, but I do not use it as a send button. The workflow that works is context first: Claude Projects hold the journalist request, source fit, proof points, hard limits, past examples and the claim we can safely make. Manus then helps with research prep, task breakdowns and first drafts. Every answer is still edited by a human before it goes out. What gets through in 2026 is not AI writing. It is specific relevance, a real point of view and a source who clearly fits the request. What gets filtered is the polished generic pitch that could have been sent to 200 journalists. AI gives itself away through vague praise, over-explaining, fake certainty, inflated adjectives and missing the one detail the journalist asked for. I would not put a fake response-rate percentage on it because we do not send pure AI versus pure manual in a clean test. The better signal is placement quality. My Featured account moved from 3,000-plus to the top 85 in less than a year by using AI for prep and humans for judgement. The lesson is simple: AI can make a good pitch faster, but it makes a bad pitch louder.

5. Christopher Coussons, Director at Visionary Marketing

   I sit on both sides of this. I draft source-request answers and cold outreach with AI help, and I get pitched constantly because we do digital PR for clients, so I screen inbound pitches every week. What earns a reply, in my experience, is a draft where AI did the typing and a human did the thinking. My workflow is to dump the raw answer into a model as bullet points I have already worked out, ask it to tidy the prose, then rewrite about half of it by hand to put the specifics back in. The bit that gets a pitch picked up is a real number from a real account, a named mistake I made, an opinion an editor can quote. A model will not invent that I tanked a client's traffic by 34 percent one quarter, because it has no stake in being honest. I do, so I say it, and that line is usually the one that gets used. What gets filtered is the opposite: a pitch that is fluent and says nothing. When I screen inbound, the tell is never a typo. It is the shape. Three tidy paragraphs of equal length, an opener that restates my own question back to me, hedging words like seamless and best-in-class, and not one concrete figure or first-hand detail the writer could only know by having done the work. If I can swap the company name out and the pitch still reads fine for any other source, it goes in the bin. Fully manual answers used to win on effort alone. Now AI-assisted drafts can match them on polish, so the only thing that separates them is whether there is a real human and real evidence underneath. The draft that cost me a placement was an early one I let a model write end to end and barely touched. It read smoothly and the journalist never replied. The fix was not less AI, it was more of me on top of it.

6. Kevin Lourd, Founder at Distribute.you

   At Distribute, we build AI infrastructure to automate high-volume outbound campaigns across sales and PR, so we see both sides of the AI pitching equation: the drafts our users generate, and the thousands of inbound emails our own servers process. For our own outreach, we use our platform's AI to aggregate research and build the initial scaffolding of a pitch. But we don't send the raw output. If we don't heavily edit the draft, usually by stripping out about half the adjectives and making the syntax deliberately choppy, our response rates drop fast compared to fully manual outreach. A few months ago, we tested fully automated, zero-edit AI sequences. Inbox providers flagged them almost instantly. We actually cost ourselves a major PR placement because an AI drafted a follow-up that sounded so relentlessly polite and perfectly structured that the journalist assumed we were a bot and blocked the domain. On the receiving end, we process thousands of inbound pitches and replies. Cloud compute is expensive, so we recently deployed basic, hard-coded scripts at the bottom layer of our servers to automatically strip out obvious AI-generated noise before it ever hits our main processing environment. What triggers those server filters, and what gives an AI draft away to my human eyes, is exactly the same. It's perfectly symmetrical paragraph lengths. It's words like delve or seamlessly. Most of all, it's the weird habit of summarizing the whole pitch in a concluding sentence. Real people rarely do that. When I get a genuine, manual pitch from another founder, it's usually a little messy, includes a quick fragment, and ends abruptly. Lately, the slightly unpolished drafts are the only ones that actually make it through our bottom-layer filters and get a reply.

7. Sahil Agrawal, Founder, Head of Marketing at Qubit Capital

   I almost replied to a pitch last week because the subject line mentioned a tool we use. Then the body was 4 paragraphs of perfectly balanced sentences that all carried the same weight. I deleted it. We sit between founders and investors, so I read inbound to investors and outbound from us. What gives an AI draft away is not a word. It is that every sentence is equally polished and nothing is left hanging. A real person trails off, overstates one thing, forgets to explain another. The drafts that get replies here are the ones where someone wrote 2 sentences themselves and let the tool fill the boring middle, then cut half of it. Fully generated and unedited gets filtered fast, maybe 1 reply in 40 against closer to 1 in 8 for the edited ones. You can sand a draft until it sounds like nobody.

8. Natalia Lavrenenko, Marketing Manager at Smarfle CRM

   I run marketing at a CRM startup and I send roughly 30 pitches a month, plus answer source requests like this one. I've tested AI-drafted pitches against fully manual ones for about six months now. What works: using AI for the structural outline and the data formatting, then writing the actual hook and the closing ask by hand. Reply rate on this hybrid sits around 12 percent for me. Fully AI-drafted, lightly edited, hovered closer to 3 percent and most of the responses I did get were polite passes. Fully manual was similar to the hybrid on reply rate but cost me roughly four times the time per pitch. What got me ignored every time: AI-drafted source-request answers that tried to sound like a quote ready for the article. Reporters can smell a pre-packaged quote a mile away. The ones that worked were when I shared a specific observation from my customer base in plain English and let the reporter pull the quote shape they wanted out of it. What costs the placement: opening sentences that start with a setup-and-payoff structure. There are three things every founder needs to know. The biggest mistake I see is this. Those sentences are the unmistakable AI fingerprint right now. I had a tier-1 placement nearly fall through because a draft used that structure and the editor flagged it. I rewrote opening with a concrete moment and it ran. When I screen inbound pitches that come into Smarfle's press inbox, the tells are: zero specific names or numbers, perfectly balanced sentence rhythm, and a closing that reframes the opening in slightly different words. Real PR pros write less tidy paragraphs. The mess is the signal.