Optimising PR for AI citation is the new keyword stuffing

There is something odd about watching PR people optimise language for machines.

We have spent years telling clients to sound clearer and more human. Now, in the rush to appear in AI-generated answers, there’s a risk we move in the opposite direction.

Of course agencies should pay attention to what gets cited by ChatGPT, Perplexity and Google’s AI features.

The problem starts when those signals become creative rules.

That’s when innovation begins to look a lot like keyword stuffing with better branding.

We have made this mistake before

Keyword stuffing also started with a sensible idea. Search engines used words to understand relevance, so marketers made sure those words appeared in their copy.

Then the logic took over.

Suddenly, articles were being bent around awkward phrases. Headlines stopped sounding natural. Writers were asked to repeat terms long after the point had been made.

The same thing could now happen in PR. Citation optimisation is useful when it helps information become clearer, more accurate and easier to verify. It becomes a problem when the needs of a machine begin shaping what we say.

The risk goes beyond clumsy writing

A lot of the discussion on AI citations focuses on formatting. Experts recommend using clearer headings, including statistics, naming your sources and keeping brand descriptions consistent.

None of that's inherently wrong.

My concern is what happens one step earlier, when teams begin choosing stories according to what AI systems already appear to favour.

Imagine a cybersecurity business with original evidence about a growing threat facing small manufacturers. The research is strong, but the topic rarely appears in current AI answers. Meanwhile, broad ransomware articles are cited repeatedly.

A citation-led strategy may push the team towards another ransomware story because the route looks safer. A stronger PR strategy might back the neglected issue, find the right specialist publication and introduce something new into the conversation.

One approach follows the existing pattern. The other expands it, which is an important distinction

AI visibility is not a fixed league table

AI answers change depending on the prompt, platform, location, retrieval method and timing. A source that appears today may vanish next week. A brand can be mentioned without being cited, cited without being recommended or recommended without receiving meaningful traffic.

So when someone presents a universal formula for earning AI citations, I become cautious.

Research into generative engine optimisation has found that adding credible sources, quotations and statistics can improve visibility in some circumstances. That doesn’t mean every paragraph needs a number, or every opinion requires a quote.

A useful statistic strengthens an argument. A statistic added because an optimisation checklist demands one is just clutter with a footnote.

Editorial judgement must come first

I am not suggesting agencies ignore AI visibility. We should know how clients are described, which sources are influencing those descriptions and where outdated claims are being repeated.

But monitoring should diagnose problems and not dictate the creative work.

Before optimising anything, I would ask a more basic set of questions. Is there a real story here? Will the reader learn something? Is the evidence strong enough to withstand scrutiny? Would an editor still care if AI citations didn’t exist?

Once those questions have been answered, you can proceed. Dates should be clear. Claims should be sourced. Company facts should be consistent across different channels, and headings should explain what a section contains. All of which are strong editorial habits, regardless.

Dashboards can highlight an opportunity, but they can’t decide whether an idea is interesting.

When a measure becomes the work

There is a real danger is that citation share becomes the target rather than the signal.

If agencies are rewarded for increasing visibility in AI answers, they will naturally produce more of whatever appears to lift that score. The same brand language will be repeated. The same publications will be pursued. Similar stories will be commissioned because they seem easier for machines to recognise.

The numbers may improve while the work becomes less original.

That would be a poor trade for PR. Our value has never come from feeding predictable phrases into a system. It comes from judgement, evidence, timing and an understanding of what people genuinely care about.

Write for people, then make it findable

Keyword stuffing failed because marketers confused relevance signals with the purpose of writing. Keywords mattered and repeating them until the copy became unreadable didn’t.

AI citations deserve the same distinction. We should make our work accurate, structured and easy to understand. We shouldn’t allow an unstable measurement system to narrow the stories we tell.

I want a machine to understand the work.

I still want a person to find it worth reading.

Harry Kerry-Grant is a digital PR professional at Perpetual10.