AI-generated expert pitches rarely win a reporter's pick on HARO, Connectively and Featured, according to a case study AI-detection company Pangram Labs published on September 23, 2026. Journalists selected about 1 in 10 fully AI-written pitches against 1 in 3 human ones, and banned accounts were 3.6 times as likely to have sent one.
Key points
- Pangram scored 1,394,740 pitches sent to 42,392 journalist requests on Featured's three platforms between June 2025 and July 2026.
- About 21% of those pitches were flagged as fully AI-generated.
- Journalists selected roughly 1 in 10 fully AI-generated pitches for publication, compared with about 1 in 3 written by people.
- 35% of journalists turned on an option to stop receiving flagged pitches, though it is off by default.
- Banned accounts were about 3.6 times as likely as active ones to have submitted a fully AI-generated pitch.
Why do AI-generated expert pitches lose on HARO?
Because reporters want a real expert, and a fully machine-written reply often comes from a service sending answers in bulk rather than from someone with first-hand knowledge, the case study suggests. On Featured's platforms, reporters chose about one fully AI-generated pitch in ten for publication, while about one human-written pitch in three made it into an article.
The study counts only clear cases. Featured supplied aggregate data, and pitches classed as mixed or AI-assisted were left out of the comparison, so the gap is between pitches written entirely by a model and pitches written by a person.
Many reporters do not wait to compare. In the case study, 35% of a pool of more than 2,500 journalists chose to screen out pitches Pangram flagged as fully AI-generated, an option they have to switch on themselves.
What do flagged pitches say about the account behind them?
Often that something is off. In Pangram's figures, 24.7% of banned accounts had sent at least one pitch flagged as fully AI-generated, against 6.8% of active accounts, which is where the 3.6 times figure comes from (Pangram).
HARO does not ban a source only for sending AI-written pitches, the case study notes; a flagged pitch is one signal that can start an internal review. Pangram ties the pattern to services that answer journalist requests automatically to collect as many placements as they can.
Why are fake experts a problem for journalists?
Because some get through. The release cites a case Press Gazette uncovered in September 2026: an "art therapist" who probably does not exist was quoted more than 30 times, by Vice, Forbes, Glamour and others, with words that scored as fully AI-written.
That case followed a wider one. An earlier Press Gazette investigation found 1,000 articles in the British press attributed to 50 fake, non-existent or AI-enhanced experts, and the UK's two main PR trade bodies, the CIPR and the PRCA, launched a campaign against fake experts in response.
What does this mean for PR professionals?
Our analysis: PR Thrive is published by Connectively, which Featured owns, so these are our own company's platforms and figures. For anyone answering requests on HARO, they make the trade-off plain: a fully AI-written reply is less likely to be picked, may never be seen, and can draw attention to the account that sent it.
- Keep the expert's own words. The comparison left out AI-assisted drafts; the losing group is pitches written entirely by a model. We covered how the detection filter works and how many reporters use it.
- Volume is not a strategy. Sending an automated reply to every request is the pattern the case study links to banned accounts. Featured CEO Brett Farmiloe has argued that fewer, better pitches now win.
- Make the expert checkable. Fake-expert cases turn on credentials nobody can confirm, so link a real profile and match the request's stated qualifications. Our contributors share ways to research journalists before reaching out.
Frequently asked questions
Who is Pangram Labs?
A company that detects AI-written text. Max Spero and Bradley Emi started it in 2023, it is based in Brooklyn, and its customers include universities, schools, publishers and businesses.
Since when has Featured used Pangram?
Since April 2025, to score every pitch across Featured, HARO and Connectively. The case study covers pitches scored from June 2025 to July 2026.
Does HARO ban sources for sending AI-written pitches?
Not for that alone, the case study says. A pitch flagged as fully AI-generated is one signal that can start an internal review of the account.
Who wrote the case study?
Annamieka Aerts and Max Spero of Pangram. It is a four-minute read on Pangram's blog, with its methodology at the end.
Sources
- Journalists Hate AI-Generated PR | Pangram x Featured: Featured's release on its press site, also distributed via HARO Newswire, September 23, 2026.
- Featured case study: Pangram Labs' case study and methodology, by Annamieka Aerts and Max Spero, September 23, 2026.
- Prominent art therapist quoted by Vice, Forbes and others is AI-generated: Press Gazette, September 18, 2026.
- PR bodies launch campaign against fake experts after Press Gazette investigation: Press Gazette, January 29, 2026.