What PR agencies get wrong about AI, and what to do instead

Buyers used to find you through a search engine and a shortlist of names they already knew. Now many of them open ChatGPT, Claude, Perplexity, or Gemini and ask a plain question: who is the best at this, and who can I trust? The answer that comes back shapes who gets the call, often before you know the buyer exists. If you are a founder deciding whether to hire a PR agency, that shift belongs at the center of the decision. The question is no longer whether an agency can win you coverage. It is whether the coverage, and everything around it, makes you the name the AI gives when a buyer asks.

What good agencies still do well

Start with the fair part. Strong agencies earn their fee on work that software does not replace. They hold real relationships with editors and reporters, built over years, and a warm introduction still lands better than a cold send to a stranger. They bring senior judgment about what to say, what to hold back, and when the timing is wrong. They run crisis response under pressure, in the hours when one careless sentence gets expensive. They operate at scale, across many markets and campaigns at once. If that is your main need, a good agency is a sound buy, and nothing below is a reason to dismiss it.

Where the old model fights the AI era

The critique that follows is about the model and the incentives it creates. The talent is often excellent. Several habits that made sense in the press-release decade now work against a founder in the AI era.

Many agencies still sell activity. You get clip counts, advertising value equivalent, and wire blasts, which measure effort rather than result. A number that looks large can say nothing about whether a buyer, or an AI, now trusts you more than it did last quarter.

Most treat AI visibility as a separate line item, a bolt-on report, or something they leave alone. That is a mistake, because when a model answers a buyer it draws on the same earned coverage, the same consistent brand, and the same credible sources that PR has always worked for. In Asia the gap is wider still. A buyer in Singapore or Hong Kong may ask in their own language, and the model answers from sources written in that language, which most English-first programs never touch.

The deeper issue is silos. Brand sits with one team, PR with another, and measurement with a third. The AI era needs them to run as one loop: a clear brand position shapes the pitch, the pitch earns the coverage, the coverage feeds the models, and what the models say feeds the next pitch. Split those pieces across separate owners and the loop never closes.

Retainers are the next mismatch. Agency pricing is built for budgets that lean teams do not have. A founder with real traction and a modest marketing spend often cannot clear the monthly minimum, so the model does not fit the company that needs the help most.

Last, the learning stays with the agency. They discover which reporter replies, which angle lands, and which subject line gets opened. That knowledge is the asset that compounds over time. When it lives in the agency inbox and walks out the door the day you leave, you have rented results without ever owning the engine behind them.

What to do instead

The fix is a different shape of work, and it rarely costs more than what you pay now. Four moves matter most for a founder.

• Insist on outcome-based measurement. Ask to be judged on results that count: qualified coverage in the outlets your buyers actually read, replies from reporters who matter in your category, and share of answer, how often the AI names you when someone asks who to trust. If a proposal leads with clip counts and AVE, push back and ask what changed for the business.
• Keep brand, PR, and AI-citation visibility joined. Run them as one loop, under one owner, with one view of what is working. A sharp brand position makes the pitch land, the coverage feeds the models, and the AI reading tells you what to pitch next.
• Own your relationships and your inbox. Build your own reporter list and send from your own address, so the replies, the context, and the goodwill accrue to you. Change your help later and the engine stays where it belongs.
• Match the model to your stage. Run it yourself on a self-serve stack, or hire help that works this way and hands the learning back to you. Either can work. The one to avoid is the retainer that keeps brand, PR, and measurement apart and keeps the learning for itself.

Cites X가 들어맞는 자리

This is the gap Cites X is built to close. It joins brand strategy, PR, and AI-citation visibility into one loop that you own: your reporter relationships, your inbox, your data, and a clear reading of whether the AI now cites you as the answer. You keep the engine, in every language your buyers use.