Journalists can tell your pitch was written by AI

Australian journalists use AI and punish you for using it. In the 2026 survey of 803 of them, 78% said an AI-written pitch reduces their trust in the PR sender, and 48% said they can almost always tell. In the same survey, 54% said they use generative AI in their own work.

That is not hypocrisy. It is a rule about who carries the risk.

Why the double standard is rational

A journalist using AI to transcribe an interview, summarise a report or draft a first pass is using it inside a process where they check the output and put their name on it. The verification stays with them.

A pitch written by AI and sent to them moves the risk the other way. If the claim is wrong, they wear it. So an AI-written pitch reads as a signal that the sender did not check either. The trust penalty is not aesthetic. It is a judgement about whether anything in the email can be relied on.

The tells are structural, not factual

Half of them can spot it almost always, which means the detection is not happening at the fact level. It is happening in the shape of the writing.

Uniform sentence length. Every sentence lands between 15 and 22 words, so the email has no rhythm. Human writing is lumpy.

Three-item lists everywhere. Three benefits, three reasons, three quotes. Real arguments rarely arrive in threes.

The flattery line that quotes nothing. Saying you really enjoyed their recent piece on the housing supply shortage is the loudest tell in the entire email, because a model can generate it from a headline. If you actually read the piece, quote a sentence from it, or disagree with a point in it. A journalist can tell the difference in half a second.

Hedged claims. Could potentially help address. May offer valuable insights. Models hedge because they were trained to. A person who knows their subject states the thing.

No local detail. No suburb, no council, no industry body, no dollar figure, no date. A model that does not know your business fills the gap with category language.

The split that works

Use AI for the parts where the risk stays with you and the output is checked before it leaves.

Research it is fine for: finding which journalists cover a round, summarising a long report before you decide whether it matters, checking whether a claim has been covered already, building and cleaning a media list, drafting the internal brief.

The pitch body it should not touch. Not because a human draft is automatically better, but because the pitch is the one artefact where the signal you are sending is that a person who understands this thought it was worth your time.

What a working pitch looks like

Five lines. Roughly 120 words.

Line one is the news, in plain words, with the number in it.

Line two is why it lands on this journalist's round. One clause. Reference something they wrote, and quote three or four words from it so it is obvious you read it.

Line three is what you have that nobody else does. Data, access, a person who will go on record, photographs from a site, a document.

Line four is what you are offering, concretely. An interview this week, the full data set, twenty minutes on site.

Line five is your mobile number.

No attachment on the first email. No boilerplate. No paragraph about the company's mission.

The uncomfortable part

Writing it yourself does not save a pitch with nothing in it. Specificity is what the journalists are actually detecting, and a human can produce generic writing without any help.

The test before you send: strike out every sentence that could appear in a pitch from a different company in a different industry. If more than two lines survive, send it. If not, you do not have a story yet, and no amount of drafting will produce one.

Write the pitch yourself, and give it something only you know.