---
title: "Can Journalists Tell If a Pitch Is AI Written? Usually, Yes"
description: "Journalists usually can tell an AI-written pitch, not with detectors but from eight tells. Score your pitch, fix the tells, and use AI for research only."
author: "Shilika Jain"
date: "2026-10-28T02:03:50.255+00:00"
tags: ["media relations", "pitching", "ai and pr", "crypto pr", "earned media"]
canonical: "https://www.shilikajain.com/blog/ai-written-pitches-journalists-spot-them"
---

# Can Journalists Tell If a Pitch Is AI Written? Usually, Yes

By [Shilika Jain](https://www.shilikajain.com/authors/shilika-jain) - 10/28/2026

Journalists usually can tell an AI-written pitch, not with detectors but from eight tells. Score your pitch, fix the tells, and use AI for research only.

---

# Can Journalists Tell If a Pitch Is AI Written? Usually, Yes

Most of the time, yes. Journalists don't need detector software to spot an AI-written pitch, because the giveaways are in the content: a generic compliment about "your recent article", an announcement with no clear news, the same polished three-paragraph shape as the forty other pitches that morning, and no sign the sender knows the reporter's beat. Whether a reporter can prove AI wrote it doesn't matter. What matters is that the pitch reads as mass-produced, and mass-produced pitches get deleted.

I've written pitches for Web3 and AI founders for years, and I also write for publications myself as a freelance author at The Chainsaw and a contributor at CryptoDailyUK. So I think about pitches from both sides of the inbox. This post covers the specific tells, a quick way to score your own pitch, and how to use AI for the research part without sending AI-sounding email.

## Why AI pitches are easy to spot right now

Reporters were already overloaded before AI drafting became normal.

> journalists get tons of pitches everyday!
>
> sometimes it's more that what an average investors get!!
>
> [Shilika Jain on X, 10 Nov 2023](https://x.com/Shilika_jain/status/1723024943313494112)

Now picture that inbox when a large share of senders use the same few tools with similar prompts. The pitches converge. Same opening flattery, same "I'd love to explore a potential feature" close, same rhythm. A reporter doesn't need to identify the tool. They just notice that this email looks like the last ten they deleted.

I noticed the same flattening in social content this year.

> ppl have started using AI to such an extent that it's difficult to differentiate between actual human content & AI content, it's become so monotonous lol
>
> [Shilika Jain on X, 13 Aug 2026](https://x.com/Shilika_jain/status/2087873093507440923)

Monotony is the real tell. A pitch can be grammatically perfect and still read like nobody in particular wrote it, and that's what gets it skipped.

## The eight tells that get an AI pitch deleted

These are the patterns that show up most often when a pitch is drafted by a model and sent with light edits. Each one is fixable.

| Tell | What it looks like | The human version |
| --- | --- | --- |
| Vague flattery | "I really enjoyed your insightful piece on the crypto market" | Name the article and the one point in it your news extends or disputes |
| No news in line one | Two sentences of context before anything new | The news, with a number or a name, in the first sentence |
| Wrong beat | DeFi funding pitch to a reporter who covers policy | Pitch only reporters whose last few bylines match |
| Inflated adjectives | "revolutionary", "groundbreaking", "first-of-its-kind" | One verifiable claim instead of three adjectives |
| Perfect symmetry | Three paragraphs, three bullets, three benefits | Uneven length, as long as it needs to be |
| Generic close | "I'd love to explore a potential collaboration" | A specific offer: an exclusive, a data set, a founder at a set time |
| Invented specifics | A statistic or quote that doesn't trace to anything | Only facts from your own fact sheet |
| No point of view | Describes the product, says nothing about the industry | One sentence on why this matters now |

The invented-specifics tell is the dangerous one. The others cost you a reply. A made-up number costs you the reporter's trust, and they'll remember your name for the wrong reason.

## How to score your own pitch before you send it

I use a quick count before anything goes out. Read the pitch and give it one point for each tell from the table above that applies.

- **0 to 1 points:** send it. One stray adjective won't sink you.
- **2 to 3 points:** rewrite the opening and the close by hand. Those two spots carry most of the AI flavour.
- **4 or more points:** delete it and start from a blank page. Editing a pitch that's mostly tells usually leaves you with a smoother version of the same problem.

Then run one last check: read the first sentence out loud and ask whether you'd say it to the reporter on a call. If you wouldn't, neither would they want to read it.

A quick worked example, using a made-up company. The AI-flavoured opening: "I hope this finds you well. I really enjoyed your insightful coverage of the evolving DeFi space, and I wanted to share an exciting development." That scores three points before the news even appears: vague flattery, no news in line one, inflated adjectives. The rewrite: "Your piece last week argued lending protocols still can't price real-world collateral. We just closed a pilot that does, with numbers we can share before Thursday." Same company, same news. The second version tells the reporter why it's for them, what's new and what they get, in two sentences.

## What makes a pitch read as human

The fixes in the table share one root. A human pitch carries information only the sender could have, and AI can't supply that because it doesn't know it.

> Cut the crap,nail the hook
>
> [Shilika Jain on X, 22 Dec 2022](https://x.com/Shilika_jain/status/1605933584501440512)

In practice that means:

1. **A hook built from your own facts.** A funding amount, a user number, a named partner, a first-party data point.
2. **A reason this reporter should care.** Tied to something they've actually written, not a compliment.
3. **A reason it matters this week.** A trend they're already covering, a regulatory moment, a competitor's move.
4. **An offer.** Exclusive access, a founder on a set day, data they can use before anyone else.

A good example from my own work is the MANTRA Chain funding announcement. The raise ran wider because it carried a regional angle, the Middle East, and a category word, RWA, that reporters were already covering. CoinDesk's headline literally picked up the Middle East framing. No model would have known which angle those specific reporters were primed for that month. You can see the full outcome on the [MANTRA Chain case study](/work/mantra-chain).

## How to use AI for research without sending AI-sounding email

The split that works for me: AI on the input side, a person on the output side.

**Good uses of AI before a pitch:**

- Summarising a reporter's recent articles so you know where to look (then read the originals)
- Listing the counter-arguments a skeptical reporter might raise
- Checking whether your announcement overlaps with something already covered
- Generating ten subject lines so you can see the obvious ones and avoid them

**Bad uses of AI for a pitch:**

- Writing the first sentence
- Writing the compliment about the reporter's work
- Adding any statistic, quote or claim that isn't in your fact sheet
- Writing the follow-up

Then write the pitch from a blank page with your research open in another window. It's slower than pasting a draft, maybe ten or fifteen minutes per reporter for a short list. That time is the point. I lay out where AI fits across the rest of a campaign in [how to use AI in a PR workflow](/blog/ai-tools-in-pr-workflow-what-to-automate), and the structural basics of a pitch are in [how to pitch journalists, with email templates](/blog/how-to-pitch-journalists-email-templates).

## Do AI detectors matter for pitches?

Not much. I wouldn't spend time running pitches through detector tools, and I wouldn't assume reporters do. Their judgment is faster and harsher than any tool: they're asking "is this for me, and is it news?" in a few seconds. A human-written pitch that's generic fails that test too. An AI-assisted pitch that's specific, accurate and clearly aimed at one reporter can pass it.

So the useful question isn't whether AI touched the email. It's whether the email shows that a person who knows the reporter's beat decided to send it.

## When AI-drafted pitches are fine

There are cases where a heavier AI draft does little harm. Reactive commentary offers, where a reporter has asked publicly for expert sources and wants a fast, short answer, reward speed and the content is your founder's actual view anyway. Internal pitch briefs for a founder to rewrite are fine too. And if English isn't your first language, using AI to tighten grammar is reasonable, as long as the hook, the facts and the offer are yours.

The trade-off is volume. If you're tempted to use AI to send three times as many pitches, don't. Sending more generic pitches trains reporters to ignore your name, and that's a cost you pay on every future announcement.

## FAQ

### Do journalists penalise PR people who use AI?

They penalise bad pitches, and AI makes bad pitches faster to produce. A reporter who gets a lazy, generic pitch from you will be less likely to open the next one, whatever tool wrote it. Accurate, specific pitches don't carry that cost.

### Is it okay to use AI to personalise pitches at scale?

Usually not. Automated personalisation tends to produce the vague-flattery tell, a sentence about the reporter's work that proves nothing. Real personalisation means a short list and a sentence only you could write.

### Should my follow-up email be written by AI?

No. Follow-ups are short, so there's no time to save, and a templated nudge is easy to recognise. Send one follow-up after three or four working days that adds something new, like a fresh data point or an available slot with the founder.

### What if the pitch is for a non-English outlet?

Write the hook yourself in the language if you can, or have a native speaker do it. Machine translation of an AI draft stacks two flattening layers. For APAC outlets, local framing matters as much as language.

*If your pitches go out and nothing comes back, [book a 30-minute pitch teardown](https://calendly.com/shilikajain/30min/) and I'll mark up exactly what a reporter sees. For full campaigns, see my [Web3 PR campaign service](/services/web3-pr-campaigns).*

---

**Get your pitch torn down by Shilika** - https://calendly.com/shilikajain/30min/

Canonical: https://www.shilikajain.com/blog/ai-written-pitches-journalists-spot-them