Reporters reply to about three in every hundred pitches they receive, and AI has made the pile they are ignoring considerably larger. The tool is useful here, just not where most people are pointing it.
Propel measured 405,964 real pitches and found reporters respond to 3.43% of them. Cision asked 1,899 journalists in 19 markets what they think of machine-written outreach and 53% object to it. Those two numbers set the shape of this guide: the volume problem is real, and adding AI prose to it makes you part of the problem rather than the exception. AI earns its place in media relations by reading a reporter’s last 90 days of coverage, building and then pruning a list, arguing against your own story until it either survives or dies, and generating subject line options. It does not earn its place writing the pitch. Journalists have published the tells, and detection tools now sit inside the platforms they use.
Reporters reply to 3.43% of the pitches they receive. That figure comes from Propel, which tracked what happened to 405,964 pitches sent through its platform in a single quarter, so it is recorded behaviour rather than people describing themselves in a survey.[7] Its more recent quarterly barometer covers close to 700,000 pitches, which tells you which direction volume is moving.[8]
Cision surveyed 1,899 journalists across 19 markets in January and February 2026. Most of them are sorting through more than 50 pitches a week, and 72% say a quarter or fewer are relevant to what they actually cover.[1,2] Muck Rack polled 897 reporters over roughly the same weeks and found 54% rarely or never respond to a pitch at all.[3,5]
I run campaigns for a living, which means I send pitches for a living, and the honest admission is that most of mine have gone unanswered too. Those odds were already ugly before generative AI. What has changed since is the floor.
In July 2026 Press Gazette reported that Qwoted, the platform that matches reporters with sources, is now banning roughly 50 accounts a day and dealing with upwards of 15 violations daily, ranging from AI-written responses to “experts” who appear not to exist at all. Its chief executive Dan Simon put it bluntly: “AI has just poured petrol on the disinformation fire.”[10]
So here is my position, and it is the spine of everything below. AI is the wrong tool for writing the pitch and close to the right tool for almost everything surrounding it. Point it at the research, the list, the story test, the subject line and the follow-up log. Then write the 140 words yourself.
My honest take, after a decade of writing outreach that mostly got ignored: the pitch body is the single worst place in the whole workflow to point a language model.
Start with what journalists have said out loud. In Cision’s 2026 survey, 53% somewhat or strongly object to receiving AI-generated pitches and press releases, softening slightly from 56% the year before.[2,15] Just over half of your recipients have told a researcher, unprompted, that they do not want the thing you were about to send them.
Notice that this is not squeamishness about the technology. Muck Rack found 82% of journalists now use at least one AI tool themselves, up from 77% a year earlier, while concern about unchecked AI rose to 26%.[3] They are perfectly comfortable with the tools and deeply uncomfortable with being handed slop by someone who could not be bothered.
Then there is the awkward fact that the writing has fingerprints. Press Gazette asked working journalists what gives machine-written copy away, and the answers are specific enough that you can check your own draft against them. Tech journalist Chris Stokel-Walker flagged constructions like “flipped the script” and the “they’re not just X, they’re Y” pattern. Freelancer Harry Wallop says no human writer would ever reach for “flip the script”, so he treats it as instant grounds for suspicion. Author Roger Miles looks for three or more consecutive paragraphs of identical length and a reluctance to use active verbs. One freelance tech writer quoted in the piece has stopped using long dashes in his own copy because they now attract the wrong kind of attention.[9]
Detection has also stopped being a party trick and turned into infrastructure. Pangram, an AI detector, was integrated into Qwoted in August 2025 specifically so journalists can see how much of a pitch was machine-written before deciding whether to read it.[11] A month later it was built into HARO and Featured on the source side.[12]
None of this makes AI writing tools useless. It makes them wrong for a document whose entire job is to demonstrate that a specific human read a specific reporter’s work and had a thought about it. We have written before about where AI copywriting holds up against human writers and where it collapses, and media pitches sit firmly on the human side of that line.
| Task | Give to AI | Keep for yourself |
|---|---|---|
| Reporter research | Cluster their last 90 days of coverage and name the recurring question | Decide what it means and whether you actually fit it |
| List building | Draft a longlist, flag obvious mismatches | Cut it, name by name, out loud |
| Story test | Argue against you as the reporter would | Kill the pitch or fix the story |
| Subject line | Fifteen options, six words or fewer | Pick one, then rewrite it |
| Pitch body | Nothing | All of it |
| Follow-up | Track dates, flag the three-day mark | Write and send the note |
The split this article argues for. Everything on the left operates on research and process; the prose stays human, which is what journalists have said they want.[2,9]
A few years ago I was trying to reach one trade reporter about a payments client, and I had already burned two pitches on her. Both were competent. Both were ignored.
So I pulled the last 30 pieces she had published, pasted the headlines and opening paragraphs into a chat, and asked one question: what does this person keep coming back to that her editor probably is not asking her for? The answer was chargebacks. Not payments generally, not fraud, not fintech funding. Specifically, small merchants losing disputes they should have won. She had written about it four times in ten weeks from four different angles.
The pitch I sent opened with a bakery in Leeds that had lost eleven disputes in a quarter and still had all the paperwork. My client appeared in the fourth sentence. She replied in about forty minutes.
That is the whole trick, and notice that none of it was a writing trick.
Paste real text. Do not ask a model who covers what from memory, because it will invent bylines and outlets and do it with total confidence. I have watched a junior colleague build a list from a chat window and put two reporters on it who had left the publication eighteen months earlier. Every name and link gets checked against the outlet itself or a media database before it goes anywhere near a send.
Three prompts that have earned their place in my process:
Ninety days is the right window. Beats move, people get reassigned, and a reporter’s obsession from two years ago is often somebody else’s beat now. Muck Rack has productised a version of this, with AI for research and list building, a badge showing which journalists appear most often in AI-generated answers, and a media brief assistant that pulls recent coverage and likely questions.[16] Useful. The judgement about whether you belong in that reporter’s world is still entirely yours.
The most expensive four minutes of my career were spent exporting a list.
Early on, on a product launch with a nervous client and a deadline I had agreed to without thinking, I pulled a keyword-matched list of 380 contacts and sent one release to all of them. Among those 380 was an editor I had spent four years building a relationship with, someone who had run three features off the back of my pitches. She replied to the blast with a single line: “Please take me off whatever list this is.” She did not open anything I sent for the next eighteen months.
The data explains why that reaction is normal rather than harsh. In Muck Rack’s 2026 survey, 88% of reporters immediately delete a pitch that is irrelevant to their coverage, 50% delete anything that reads like a mass email, and 35% delete pitches addressed to the wrong name or outlet.[5] Cision found 82% naming lack of relevance to their audience as a reason to reject, and its chief marketing officer Amy Jones told PR Daily that high-volume spray-and-pray outreach erodes credibility and damages relationships.[1,15]
The saddest number in the whole Muck Rack report: only 3% of journalists say PR outreach always reflects the community their outlet serves, and another 13% say it usually does.[5]
Worth knowing where the cheap-volume routes now stand, because they have changed twice in two years. Cision permanently discontinued Connectively, formerly HARO, on 9 December 2024.[13] Featured then acquired and relaunched HARO, and both HARO and Featured now run Pangram over source responses so journalists can see how much of an answer was machine-written.[12] The path that used to be the fastest way to get quoted is now the most heavily policed one in the industry.
Here is the part that most pitching advice skips, including advice I have given.
I once spent a fortnight arguing with a founder who wanted a national business desk to cover a partnership. It was a genuinely good partnership. It was also two mid-size companies agreeing to work together, which happens somewhere every Tuesday. He kept asking me to make the pitch stronger, and there was nothing wrong with the pitch.
What ended the argument was making him read it from the other side of the desk. I put the draft into a chat with the instruction: “You are a senior business reporter at a national daily. You will get 100 pitches today. Give me the three reasons you delete this one, and tell me what would have to be true for you to reply.” The model was blunter in nine seconds than I had managed in two weeks. He read it, went quiet, and dropped it. We put the announcement on the company blog and pitched the trade press instead, who ran it properly.
That is the honest failure mode of AI in this job: it will cheerfully produce 200 elegant words about a non-story, because it has no view on whether anything happened. Ask it for the objection rather than the copy.
Muck Rack asked journalists what makes a pitch genuinely relevant beyond simply matching their beat. The winner, by a 33-point margin, was that it directly affects the community their audience belongs to.[5,14] So before anything gets written, answer three questions: what community does this journalist’s audience belong to, does this story affect that community, and can the reporter see the effect immediately.[15]
It also helps to know what journalists are short of. In Cision’s survey, 47% want more data or research from PR teams, 45% want embargoed or early access, and 42% want access to experts. Only 22% asked for more press releases.[1]
If the honest answer is that a customer got a good result with your product, you have a case study rather than a news story, and our guide to writing a case study with AI is a better home for it than a reporter’s inbox.
Short. Shorter than you think, and shorter than your client will want.
Muck Rack found 69% of journalists prefer pitches under 200 words.[5] Cision’s data puts the general sweet spot at 100 to 200 words, with some flexibility by pitch type.[1] Propel’s behavioural data backs both up: pitches with bodies of 51 to 150 words drew a 7.51% response rate against a 3.43% average across everything sent.[7] Roughly double, for writing less.
Propel found subject lines of one to five words earned the highest response rate at 4.73%, while subject lines of 10 to 15 words made up 36.54% of everything PR people actually sent.[7] The whole industry is doing the opposite of what works. Six words that say what the story is, with no adjectives, will do. This is the one place I do use AI on the copy: fifteen options, six words maximum, then I pick one and rewrite it so it sounds like a person typed it in a hurry.
One number that changed how I staff launches: 52% of journalists expect a PR contact to respond within 24 hours when they reach out first.[1] Your reply speed is worth more than your send time. If you are pitching on a day when nobody can get a spokesperson on the phone by tomorrow lunchtime, do not pitch.
Finally, a pitch is a short personal argument aimed at one reporter, while the announcement itself belongs somewhere else entirely. If what you are actually writing is the announcement, our guide to writing a press release with AI covers that job, and the two documents should never be the same file.
Once. Three to seven days later. That is the whole policy.
Muck Rack’s 2026 data has 50% of reporters saying one follow-up is ideal and 51% saying it should land between three and seven days after the original.[5,6] Cision’s report summarises its own findings as think twice, follow up once.[1] There is no version of this data where a fourth email helps you.
Muck Rack, State of Journalism 2026, based on 897 reporters surveyed between 30 January and 2 March 2026. Respondents could choose more than one answer.[4,5]
The blocking risk is real and it is documented. 46% of reporters delete pitches that arrive repeatedly with no response, and a 2024 Cision survey of 3,000 journalists found 48% will block a PR person outright for repeated aggressive follow-ups.[5,6] One relationship, gone, over an email you sent because a client asked whether you had chased.
AI’s job in this stage is the log, not the letter. Keep a simple table of who received what and when, have it flag the three-day mark, and let it draft nothing. The follow-up itself should be two sentences: what you sent, one new fact, and an offer to stop. I write the same one every time and it takes ninety seconds.
The last thing, and the one I wish somebody had told me at 26. Silence is not rejection. Frank Strong tells the story of chasing a reporter for months, eventually asking directly whether he had done something wrong, and getting back: “I simply don’t have time to reply to every email or pitch, much less cover everything. You’ve put the company on my radar. Keep the news flowing.”[5] That is what most non-replies mean. AI can make you faster at every part of this job except the part that only patience fixes.
No. In Cision’s 2026 survey of 1,899 journalists, 53% somewhat or strongly objected to receiving AI-generated pitches and press releases. Journalists have also published the specific tells they look for, including repeated sentence patterns, paragraphs of identical length and a reliance on abstractions. Use AI for research, list building, story testing and subject line options, then write the 140 words of the pitch yourself.
Often, and increasingly they do not have to guess. Working journalists interviewed by Press Gazette named phrases like “flip the script” and structures such as “they’re not just X, they’re Y” as instant giveaways. Detection has also been built into the platforms: Pangram was integrated into Qwoted in August 2025 and into HARO and Featured a month later, showing journalists how much of a pitch or source response is machine-written.
Under 200 words, sent before noon. Muck Rack found 69% of journalists prefer pitches under 200 words and 78% want them in the morning. Propel’s data on 405,964 tracked pitches shows bodies of 51 to 150 words drew a 7.51% response rate against a 3.43% average. Subject lines of one to five words performed best. Email is the channel: 97% of journalists prefer it.
Once, three to seven days later. Muck Rack found 50% of reporters say one follow-up is ideal and 51% say it should come within three to seven days. The downside of overdoing it is permanent: 46% of journalists delete pitches sent repeatedly without response, and a 2024 Cision survey of 3,000 journalists found 48% will block a PR contact for repeated aggressive follow-ups.
Four things. Reading a reporter’s last 90 days of coverage and naming the question they keep returning to. Drafting a longlist of contacts that you then prune by hand. Playing the reporter and arguing against your own pitch until the story either survives or dies. And keeping the follow-up log honest. Paste real text rather than relying on the model’s memory, because it invents bylines and outlets confidently.
Every number here was pulled from the original report or press release in August 2026, not from a roundup post. The Cision figures come from the 2026 State of the Media PDF itself; the Muck Rack pitching numbers were cross-checked against Frank Strong's line-by-line write-up of the same report. I deliberately left out several widely circulated pitching statistics, including a much-quoted claim about pitch open rates by day of week, because I could not trace them to a published methodology. I also left out software pricing, since none of the vendors named here publish rate cards I could verify. The campaign anecdotes are mine, with client details changed.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.