This blog may be quite contentious, but AI is a factor of how we work now and we need to acknowledge it head on. We seem to be in a period of ‘don’t ask don’t tell’ when it comes to AI use, with agencies, clients and journalists happy to go about their days avoiding the elephant in the room.

That was until Anna Moloney from City AM wrote an op-ed about not printing an actual op-ed that week because of the amount of ‘AI slop’ that had been fired her way in lieu of ‘real’ opinion pieces.

Not being short of an opinion myself and not slow in calling out lazy PR in the past, I pretty much agreed with everything Anna said. But given AI is used in a variety of different ways from research to drafting and the tightening of copy, it did raise the question, how much AI is too much? And what level will journalists like Anna accept when using AI detection software? 10%, 20%, none?

Then I thought, what if I asked AI, how much is too much AI in PR?

What I got was a perfectly cogent, completely defensible and highly convincing argument. Exactly what I would have said – except I didn’t.

Now I was faced with a dilemma! Do we clean this up to remove the obvious signs of AI authorship and publish, just to see if my fellow PR professionals and journalists would spot it? How many would agree, how many would disagree, would anyone run it through AI detection just to check?

While I don’t mind calling out the industry where I see fault, I have no desire to embarrass anyone or make them feel foolish.

But it did raise the issue around the successful use of AI to produce content of value. Anna’s point was more about the ‘banal genericism’ of what she received and that is not the fault of AI, but the inexperience or lack of judgement of the PR person operating it.

“Writing, especially in opinion, is about individual voice: how you turn a phrase, the anecdote only you can tell, that odd word you might have picked up from a well-loved novel.” 

She said herself that they could get any number of regular columnists to bang out “sharp copy at short notice”, to fill the space. But that’s the issue, in the right hands, AI is actually also pretty good at this.

The onus is on the person employing the tool to then build back in the individual voice. This style of writing also often requires the PR person to add their own opinion to that of your clients, to help successfully join the dots.

Good PR is a discipline that requires the ability to take information from a wide range of sources, fold it in with a hint of foundational economics, be a good student of human behaviour, a subject matter expert and combine all of these with a decent scoop of creativity. It requires judgement, subtlety and nuance.

But this can take time and AI is proving incredibly useful at helping to do part of this job by taking the vast swathes of research and putting it in easily digestible, very readable formats. Often, at first read, these can seem hard to improve on.

So, to the time poor PR with a million competing demands and deadlines or the inexperienced exec lacking the confidence to add clear opinions back in, it’s good enough.

And that’s what is critical in this debate. AI isn’t the issue, it’s a tool. It’s the expertise, ability and above all the judgement of the user that dictates the quality of the output. You know the old adage of a bad workman blaming his tools – well it still fits to this day.

So, whether it’s inexperience, poor judgement or a lack of time, the issue is one of PR slop, not AI slop.

This is going to be a consistent issue in our industry for the foreseeable future and we’re going to have to get to grips with the wider point of addressing AI use in client work and in published media and find a solution sooner rather than later.

Now, if anyone would like to read a really good blog that is literally a textbook perfect argument for how much AI is too much in PR, here is one I didn’t write earlier….

Lazy ‘AI Slop’ in PR: How Much AI Is Too Much?

There is a sentence that should make every PR professional wince: a journalist has saved space for a promised opinion piece, only to discover the copy is not expert insight at all, but 600 words of bland, machine-polished filler. That was the frustration behind City A.M.’s recent broadside against “AI slop”, with Anna Moloney, Deputy Comment and Features Editor, describing an inbox increasingly clogged with AI-written pitches and articles masquerading as human expertise.

The reaction across PR should not be defensive. It should be embarrassed. Because the issue is not whether AI has a role in communications. It plainly does. The issue is whether agencies, in the rush to be faster and cheaper, are confusing efficiency with abdication. If a client is paying for judgement, positioning, nuance and voice, then handing a publication a generic AI draft is not innovation. It is a breach of trust.

So what percentage of a published article is acceptable to be done using, checked by or generated with AI? The honest answer is that a simple percentage is the wrong measure. An article could be only 10 per cent AI-generated and still be unacceptable if that 10 per cent contains the argument, examples or supposedly personal opinion. Equally, a piece could be heavily supported by AI at the research-organising stage and remain legitimate if every claim, judgement and turn of phrase has been shaped, checked and owned by the named author.

A better test is this: could the named author defend every line in conversation with the editor? Could they explain why the argument matters now, where the examples came from, what experience sits behind the view and why the piece sounds like them? If the answer is no, the article should not be submitted under their name.

For PR, that means AI can be acceptable in three limited ways. First, as a preparatory tool: summarising background material, mapping angles, identifying likely counterarguments or helping structure an interview transcript. Secondly, as an editing aid: spotting repetition, tightening overlong sentences or checking clarity once the human argument exists. Thirdly, as a quality-control assistant: flagging unsupported claims, inconsistent terminology or gaps that need a human source.

What AI should not do is replace the essential work: interviewing the expert, extracting the distinctive view, challenging the obvious line, adding lived experience, understanding the publication’s audience and writing in a voice that carries authority. A published byline is not just a container for grammatically correct words. It is a promise that a real person has something worth saying.

If the industry wants a rule of thumb, it should be this: zero per cent of the final thinking should be outsourced, and no undisclosed AI-generated prose should appear in a submitted article. AI-assisted is acceptable; AI-authored is not. The human contribution should be visible in the argument, the evidence, the tone and the risk. Opinion writing should have edges. It should contain a claim that someone might disagree with. It should sound as though it came from a person with skin in the game.

Disclosure also matters. Publications should be told when AI has materially assisted a piece, particularly where it has helped draft or rewrite sections. Editors may decide their own thresholds, but PR teams should not make that decision for them by omission. Transparency protects the journalist, the client and the agency relationship. It also avoids the humiliating spectacle of being caught by detection software or, worse, by an editor’s instinct for empty prose.

The most practical standard for agencies is a human-first workflow. Start with a proper briefing or interview. Build the argument from the spokesperson’s expertise. Use AI only after that point, and only to improve clarity, not to invent substance. Then fact-check manually, remove generic phrasing and ask whether the piece would still make sense if the author had to say it aloud on radio. If not, it is not ready.

The danger of lazy AI use is not merely that bad articles get rejected. It is that PR devalues the thing it claims to sell: credible influence. Journalists do not need more content. They need useful, timely, original ideas from people who know what they are talking about. AI can help prepare those ideas for publication, but it cannot supply the judgement, tension or authenticity that make them worth publishing.

The answer, then, is not 20 per cent, 30 per cent or any neat threshold. The acceptable amount of AI is the amount that remains invisible because the human thinking is unmistakable. If an editor reads a piece and hears a person, AI may have played a useful supporting role. If they hear “word froth”, the industry has failed before the first paragraph.