Why Your AI Press Releases Sound Like AI (And What a Governed Workflow Looks Like)
By Gabriel Tan | March 2026
A 2024 Buzzsumo content analysis found that while 87% of content creators now use AI assistance, only 13% produce content that generates meaningful engagement. The other 87% sounds like everything else on the internet.
Press releases are no different. Most agencies using AI for press releases follow the same pattern: someone opens a chat window, types "write a press release about our Q4 earnings," and edits whatever comes back. The output is generic. It reads like it could belong to any company on any exchange. The adjectives are inflated. The executive quote sounds like a template. The journalist who receives it can tell.
According to Cision's 2025 report, 72% of journalists cite factual errors as their top concern with AI-assisted PR content. But the problem that arrives before factual errors is tonal flatness. A press release that sounds like AI will not get read carefully enough for anyone to check the facts.
The issue is not the AI. It is the absence of a system around it.
Here is what a governed AI press release workflow looks like when it runs well.
The client has a brand voice document. Before any drafting starts, someone has pulled the client's last three to five press releases and their annual report chairman's statement. They have noted the tone, the vocabulary, the recurring phrases, how the CEO speaks in quotes. They have written it up in a page or two and loaded it into a shared AI project. The AI reads this document before every draft. It matches the client's voice because it has been given the client's voice.
The project has an anti-AI writing guide. This is a separate document that lists the specific language patterns that make AI output sound machine-generated. Trailing "-ing" phrases. The "not X, but Y" construction. Em dash overuse. Words like "pivotal" and "multifaceted" that spiked in frequency after 2022. Vague attribution. Uniform sentence length. The AI reads this guide, checks its own output against it, and corrects before presenting the draft.
The team fills in a structured intake form. Who, what, where, when, why, how, quote source, objective. Every field completed. No blank fields. "N/A" is acceptable. Blank is not. This replaces the loose paragraph that produces loose output.
The AI asks clarifying questions before writing. If the "why it matters" field says something weak like "this is good news for shareholders," the AI pushes back and asks what specifically makes it good news. This mimics what a good editor would do. The draft is only as good as the brief.
There is a QA checklist. The team member reviews the draft against defined criteria: data verified, quote attribution checked, regulatory language confirmed, formatting correct. The senior sees the draft only after QA is complete. Their role is sign-off, not rewriting.
The result is a first draft that sounds like the client, not like a machine. It is not perfect. No first draft is. But it is a draft worth editing, not a draft worth deleting.
The agencies producing press releases that sound like AI are not using bad tools. They are using good tools with no voice control, no anti-AI layer, no structured input, and no QA gate. The tool is in use. The system is not.
The system takes about 30 minutes to set up for a new client. The brand voice document and the anti-AI guide carry across every deliverable for that client. Once the system is in place, every press release follows the same path. The quality is controlled. The voice is consistent. The governance is built in.
The 87% of AI-assisted content that sounds generic is not a technology problem. It is a setup problem. The 13% that works has something the rest does not: a system that tells the AI who the client is, what the client sounds like, and what the output must not sound like.
That system is not complicated to build. It is just rarely built at all.
Gabriel Tan is the founder of Mekong Bridge Advisory. He builds structured execution systems for PR and communications firms.
gabriel.tan@mekongbridge.com | www.mekongbridge.com
