How to Write PR Press Releases So AI Models Cite and Mention Them?

Ask ChatGPT or Gemini about a brand you represent. There’s a real chance the model answers from a three-year-old article, a scattered review site, or worse, from information about a competitor. Your latest press release? It may not exist at all from the model’s point of view, even if it went out to sixty outlets and got picked up by a dozen of them.

This isn’t theoretical. The Reuters Institute found that weekly use of AI chatbots for news reached 10% globally in 2026, and 17% among 18 to 24 year olds, while only 4% of those users click through to a source. If a brand isn’t part of the answer itself, it doesn’t exist for that reader. This isn’t a call to reinvent PR. It’s a short list of habits worth adding to work you already do well.

A press release built for AI is still a press release, just read differently

Plenty has already been written on this exact topic. Muck Rack, Migazette, PR.co, and several PR-tech blogs have each published their own structural checklist for AI-citable press releases, and most of them agree on the fundamentals even when they order the priorities differently. Rather than adding one more competing list, the goal here is to give a clear hierarchy: what actually matters most, what’s a smaller optimization, and what most guides skip entirely (vendor selection, EU disclosure rules, and the difference between using AI to draft a release and writing one AI will cite).

The classic press release structure, a lead answering who, what, where, when and why, a quote from someone with authority, a boilerplate describing the company, hasn’t lost its value. That skeleton works as well for a journalist as it does for a language model. The problem is that good PR craft alone doesn’t guarantee a release gets picked up and cited by AI, because citation depends on things no classic PR course ever taught.

Check whether AI already knows your brand before changing anything

Before rewriting a single release, check the starting point. Type a query like “recommended [category] company” or simply the brand name into ChatGPT, Gemini, and Perplexity. Responses tend to fall into one of five levels:

• the model doesn’t recognize the name at all
• it recognizes the name but doesn’t know what the company does
• it correctly places the brand in its category
• it mentions the brand without naming it, in response to a problem-based query
• it can explain why this specific brand fits the user’s need

Real business value only starts at the last two levels. If a brand is stuck at level one or two, the reason is usually simple: there isn’t enough consistent, current, and credible information online for a model to build an accurate picture. Sometimes a model relies on a single outdated source. One of our clients was actively discouraged by Gemini, which claimed the company had no customer support line, based on one indexed source that had gone stale. The support line worked fine; the information about it online just hadn’t been updated anywhere the model could find.

Six changes that actually raise your odds of getting cited

This is the section that matters most, so no padding: here are the concrete habits that move the needle more than everything else combined.

1. Put the answer in the first sentence, not the fourth paragraph. Models build their answer from the first few dozen words of a piece of text. If the key fact is buried mid-document, the model may never reach it. The classic lead structure holds up here, it just needs to be enforced more strictly than before.
2. Use full names instead of pronouns. “The company” tells a model nothing about who is being discussed. The full company name, the name of the person quoted, the product name, repeated consistently rather than stated once and swapped for pronouns afterward, builds entity recognition. That’s the difference between “the company launched a new product” and “Acme Corp launched a certified organic cotton shirt line.”
3. Use numbers and evidence instead of adjectives. “An effective solution” means nothing to a model because there’s nothing to cite. “67% of customers saw double-digit conversion growth within a quarter” is a sentence a model can actually quote, because it carries specific, checkable content.
4. Add an FAQ section at the end of the release. This is the simplest change with the biggest payoff. Two or three questions phrased the way a real user would ask them (“is this available for small businesses,” “how much does it cost,” “how is this different from X”), followed by short, concrete answers, hand the model a ready-made extractable chunk instead of forcing it to guess which sentence answers which question.
5. Expand the boilerplate. The classic boilerplate was a single sentence about the company. For a model, it’s worth expanding to cover industries served, audience segments, and specific areas of specialization. That’s exactly where a model picks up the signal for which categories a brand should even be considered in.
6. Structure everything to be easy to chunk. Short paragraphs, one idea per paragraph, lists wherever you’re genuinely listing parallel items (like this one). Models split text into chunks and evaluate each one separately. A long, dense paragraph carrying five different ideas is a chunk nothing can cut cleanly.

GEO versus traditional PR: what actually transfers and what doesn’t

Muck Rack alone has published two separate posts arguing that PR skills transfer directly to generative engine optimization, joined by TEAM LEWIS, Worldcom Group, and PRLab making similar arguments. The reassurance is fair, but most of these pieces stop short of saying exactly which skills carry over and which don’t. Here’s the honest split:

Transfers directly: media relations research and understanding what makes a source credible, quotable executive positioning, crisis response instincts, understanding audience and narrative.

Requires new, specific literacy: structured data and schema markup, chunk-length and content structuring for extraction, entity consistency across a domain, and monitoring methodology across multiple models rather than a single search engine.

The honest answer is that PR expertise is a real advantage here, not a liability, but it isn’t sufficient on its own. Treating GEO as “PR plus a few technical add-ons” is closer to the truth than treating it as an entirely separate discipline.

Using AI to draft a release is not the same as optimizing it for citation

This mix-up comes up often enough to name directly. Guides on “ChatGPT prompts for press release writers” address a different problem: using AI as a drafting assistant to write faster. Guides on structuring a release for AI citation address whether the finished, published release has a chance of being cited when someone else asks a model about your industry. Both use the same surface vocabulary (ChatGPT, press release, AI), which makes them easy to confuse in a quick search. They are not the same task, and conflating them is a common way teams end up optimizing the wrong thing.

What the EU AI Act means for press releases

This is a gap in nearly every English-language guide currently available: none of the major GEO or press-release checklists mention EU AI Act disclosure obligations at all. Article 50 of the EU AI Act sets transparency requirements, including labeling AI-generated content and disclosing deepfakes. These obligations already exist in law; what’s changing is enforcement. Transparency obligations under Article 50 become enforceable from August 2, 2026, and under the Digital Omnibus, labeling requirements for AI systems already on the market extend to December 2, 2026.

For a communications team, this matters in a very specific way: if AI tools are used anywhere in producing a press release, from drafting assistance to translation to image generation, disclosure obligations may apply depending on how the content is used and distributed. This is worth flagging to legal or compliance counsel directly rather than treating any single article, including this one, as a full legal read of the requirement.

Choosing a GEO partner: not every option fits a PR team

The vendor landscape for AI visibility work splits into three categories that don’t map neatly onto each other: SEO and GEO agencies pricing services from roughly $1,500 to $50,000 a month, a smaller group of PR-native providers such as Genevate, High Vibe PR, and SHIFT that combine GEO work with PR strategy, and self-serve monitoring tools running $29 to $489 a month. Most published buying guides assume a reader has already decided “I need a GEO agency” without asking whether that’s the right category at all.

The question worth answering before signing anything: does this need technical work on structured data and content (an SEO-native task), or does it need someone who connects AI-visibility monitoring to an actual communications strategy, one that starts with a brief, audience personas, and a desired brand narrative, and ends in concrete content recommendations for a PR team. That second approach, pairing an AI-visibility audit with real collaboration alongside a client’s existing PR agency, is what Insightland builds through AI Search Optimization: audit, strategy, implementation, and ongoing monitoring, run as one connected process rather than a one-off report.

How long an audit takes and what it costs

Language models are non-deterministic. The same prompt run on a Monday and again on a Thursday can return a different answer. That’s why a credible visibility audit requires many observations across time and across models, not a single ChatGPT screenshot. A solid methodology typically covers dozens of prompts, several platforms, and repeated runs across a week or more; only the pattern across those runs separates a real signal from noise. In practice, the full process from initial audit to implemented recommendations usually takes about a quarter.

On pricing specifically:

• a one-time AI visibility audit (Brand Search Presence): starting at roughly PLN 5,000 (around $1,250)
• ongoing LLM visibility monitoring: starting at roughly PLN 1,000 a month (around $250), depending on the number of prompts and models tracked
• full implementation and strategy work: custom pricing based on scope

These figures describe the audit and monitoring layer specifically, not the full scope of PR-adjacent work (media relations, crisis management, narrative building), which is typically scoped separately.

Classic press release versus AI-citable version

Frequently asked questions

Does writing for AI citation mean abandoning classic PR craft?
No. The lead, quote, and boilerplate structure all stay. What’s added (entity naming, FAQs, concrete numbers) is really a stricter version of what good PR already teaches: be specific, don’t pad with vague language.

Is using ChatGPT to draft a press release the same as optimizing it for AI citation?
No, and this is a common mix-up. One is about a writing tool. The other is about whether a finished, published release gets cited when someone else asks a model about your industry. They’re different problems that share similar-sounding search terms.

How do you know if a release actually got cited by AI?
Without a monitoring tool, you largely don’t. Manually running the same prompts across several models periodically gives a first read, but because models are non-deterministic, it takes repeated runs to separate a real trend from a one-off fluke.

Do smaller brands stand a chance at AI visibility?
Yes, often more of a chance than in classic Google search. In some categories, models recommend smaller, specialized players more readily than a search engine conditioned to surface the same large brands for every query, provided that brand has a consistent, current presence online to draw from.

Before rewriting anything, check what AI is already saying about the brand you represent. That’s the step that comes before any optimization, and it can be done in an evening.

Sources: Reuters Institute Digital News Report 2026; Muck Rack, “How AI and Generative Engine Optimization Reshape PR Strategy” and “How Press Releases Improve GEO” (2026); Migazette, “How to Write a Press Release That AI Cites”; PR.co structural framework for AI-citable releases; EU AI Act, Article 50 (artificialintelligenceact.eu); HSF Kramer analysis of EU AI Act transparency obligations (March 2026); vendor pricing data compiled from Nutshell, DemandLocal, Red-engage, and FirstPageSage, July 2026.