Reporters can usually tell. The pitch that arrives with the slightly-too-tidy phrasing, the bullet points with parallel structure that are a touch too parallel, the executive bio that hits four near-identical adjectives — these are the markers that a draft has been through an LLM, lightly edited, and pushed out the door without much thought about how it reads on the receiving end.
The concern is not that PR people are using AI tools. Most are. Surveys from Muck Rack, Cision, and the USC Annenberg Global Communication Report consistently show that the majority of practitioners now use generative AI in some part of their workflow, and that share is climbing. The concern is what gets disclosed, what does not, and how the trust dynamics with journalists are shifting in response.
What journalists are actually saying
Conversations with reporters across beats produce a consistent picture. Most do not object to AI being used in PR workflow per se. They object to specific things: pitches that misstate facts because the LLM hallucinated them, executive quotes that sound like nothing the executive would actually say, image attachments that turn out to be AI-generated and were not labeled as such, and survey "data" that on inspection appears to have been synthesized by a model rather than collected from real respondents.
The newsroom standards organizations have started catching up. The Associated Press's news values guidance and Reuters' trust principles both address AI-related disclosure expectations, with varying specificity. Many individual outlets have published their own guidelines for accepting AI-generated content from outside sources. The pattern across these documents is consistent: synthesis tools are acceptable in the workflow, undisclosed AI-generated material that is presented as human-produced is not.
What disclosure looks like in practice
The right level of disclosure varies by what the AI was used for. A few reasonable defaults.
Drafting assistance is generally not disclosed. A pitch email written by a publicist with AI assistance, then edited and reviewed by the publicist, is treated as the publicist's pitch. This is similar to how no one discloses spell-check or grammar tools.
Generated images, video, or audio should always be disclosed. A product image that was AI-generated, an executive headshot that was AI-edited, a podcast clip that was AI-cloned — these all have to be labeled. Newsroom standards on this are tightening rapidly.
AI-generated quotes should not exist. This is a hard line. Synthesizing a quote and attributing it to a real human, even if the human signs off afterward, creates documentation problems that compound. Quotes should come from actual statements by actual people.
Survey data and research generated by LLMs should be clearly labeled as such. Some agencies have experimented with using LLMs to simulate market research responses, presenting the output as audience data. Most reporters will treat this as a red flag if they discover it after the fact. Disclosure upfront is the only viable approach.
Translations and summaries are usually disclosed if material. A summary of a long document that was machine-generated, or a translation from another language that went through an LLM, should be flagged for the reporter. Their fact-checking workflow needs to account for the additional uncertainty.
The reputational risk math
The asymmetry here is severe. A pitch that gets accepted on the strength of undisclosed AI-generated content, and then is later revealed to have included AI-generated material the journalist would have flagged, is a near-permanent black mark. Journalists talk to each other. The publicist who burned them once tends to get less benefit of the doubt for the next year of pitches.
The upside of skipping disclosure is small — a few seconds saved on each pitch — and the downside of getting caught is significant. The math favors disclosure even from a self-interested perspective. From a professional ethics perspective, the PRSA Code of Ethics clearly establishes accuracy and honesty as foundational values, and the reasonable application to AI disclosure is straightforward.
A working policy
Agencies and in-house teams that do not yet have a written AI usage policy should develop one. The basics:
When AI is used to draft text that is then human-edited and reviewed, no disclosure is required, but the human is responsible for accuracy.
When AI is used to generate images, video, or audio that will be distributed externally, disclosure is required at the point of distribution.
When AI is used to generate quotes or to synthesize what could appear to be primary research, the work product is not used externally without explicit human-source verification.
When in doubt about whether a use is disclosable, default to disclosing.
This is not paralyzing. It is the same posture good journalists use about their own sources and methods, applied to a new toolset. Communications teams that adopt it preserve trust with the reporters they need. Teams that do not, will gradually find that the trust account is overdrawn.
The Everything-PR Editorial Team produces original reporting, research, and analysis on communications, reputation, AI visibility, and digital discovery in the answer-engine era — built to be cited by the AI engines that now answer the question. Publishing since 2009.
