admin
The rapid integration of generative artificial intelligence into the communications sector has moved past the initial phase of novelty and into a critical stage of operational refinement. As public relations professionals and corporate communicators increasingly rely on custom Large Language Models (LLMs), such as OpenAI’s Custom GPTs, a recurring technical hurdle has emerged: the tendency for these systems to "forget" complex instructions or revert to generalized training data over specific organizational guidelines. Sarah Evans, partner and head of PR at Zen Media, argues that the solution to this inconsistency does not lie in the technology itself, but in the rigorous standardization of internal workflows and the conversion of institutional knowledge into a repeatable digital infrastructure.
The challenges cited by Evans reflect a broader industry struggle to maintain brand voice and accuracy in an automated environment. Even when provided with explicit rules, AI models can suffer from "drift" or "hallucination," leading to errors as minor as incorrect punctuation—such as the persistent use of em dashes when prohibited—or as significant as the fabrication of quotes and sources. These failures often stem from a misunderstanding of how LLMs process information. Rather than possessing a human-like memory of a conversation’s beginning, AI models operate within a "context window," and as a conversation progresses or tasks become more complex, the weight given to initial instructions can diminish.
The Evolution of AI in Communications A Brief Chronology
To understand the current shift toward standardized AI instructions, one must look at the timeline of AI adoption within the media and PR landscape. The release of ChatGPT in late 2022 triggered a wave of experimentation, with communicators primarily using the tool for brainstorming and rough drafting. By mid-2023, the industry moved toward "prompt engineering," where users learned to craft more detailed queries to achieve better results.
The introduction of Custom GPTs in late 2023 marked a significant turning point, allowing organizations to upload proprietary documents and set "System Instructions" to govern the AI’s behavior. However, by early 2024, it became evident that simply creating a custom bot was insufficient for high-stakes professional work. The current phase, as highlighted by Evans and other industry leaders, involves the creation of "Knowledge Infrastructure"—a system where editorial standards, banned phrases, and formatting rules are not just uploaded once but are reinforced through every single interaction.
Turning Institutional Knowledge into Infrastructure
The core of Evans’ philosophy is that AI cannot independently replicate the nuance of an organization’s unique perspective. "You can’t magically just say, ‘Hey, ChatGPT, build me a GPT for optimizing press releases,’ because it has to come from your knowledge," Evans noted. This necessitates a thorough audit of how an organization thinks, writes, and reviews content.
For a communications team, this infrastructure typically includes:
- Editorial Style Guides: Beyond basic AP Style, these include brand-specific preferences regarding tone, voice, and formatting.
- Banned Phrase Lists: Identifying overused industry jargon or "AI-isms" (words like "delve," "unlock," or "transformative" that LLMs frequently overuse) to ensure the output sounds human and authentic.
- Trusted Source Repositories: A list of approved white papers, internal data sets, and industry experts that the AI should prioritize when generating content.
- Formatting Requirements: Specific rules for subheads, bullet points, and metadata designed for both human readers and search engine algorithms.
By documenting these standards, teams create a "source of truth" that can be fed back into the AI system. This prevents the model from relying on its general training data, which may be outdated or misaligned with the brand’s specific goals.
The Power of Redundancy and the Addendum Method
One of the most tactical shifts Evans advocates for is the use of "addendums" for every task. Rather than relying on the AI to remember instructions stored in its global configuration, Evans’ team attaches a detailed editorial PDF to every prompt. This document serves as a constant reminder of the rules, ensuring that the model’s "attention" is always focused on the most relevant constraints.
"Repetition matters for each task. Redundancy is everything," Evans emphasized. This approach treats the AI less like a seasoned colleague and more like a "toddler" that requires step-by-step guidance. In technical terms, this is often referred to as "Chain-of-Thought" (CoT) prompting, where the AI is instructed to perform a task in a specific sequence: first, research the topic; second, outline based on the style guide; third, draft the content; and fourth, check the draft against the banned phrase list.
Supporting Data The Impact of AI on PR Productivity
The move toward standardized AI workflows is supported by emerging data regarding productivity in professional services. A 2023 study by researchers at MIT and Stanford found that generative AI can increase the productivity of workers in writing-intensive tasks by as much as 40%, with the quality of work improving by 18%. However, these gains are only realized when the AI is properly managed.
Evans provided a concrete example of this efficiency in practice. A junior writer on her team previously spent approximately two hours researching and drafting an article. By mapping the writer’s workflow and turning it into a reusable "AI skill" attached as a prompt addendum, the time required for the same task was reduced to just 12 minutes. This represents a 90% reduction in production time, allowing the staff member to focus on higher-level strategy and creative oversight rather than manual drafting.
Adapting to AI-Driven Search and LLM Optimization
The necessity for standardized instructions is also driven by the changing landscape of search engine optimization (SEO). As search engines like Google and Bing integrate AI-generated answers (such as Search Generative Experience, or SGE), and as users increasingly turn to LLMs like Perplexity for information, the structure of press releases and owned media must evolve.
Evans noted that LLMs often prioritize the beginning of a document, sometimes reading only the first 240 characters to determine relevance. Consequently, her team has standardized new structures for press releases, including:
- Multiple Subheads: Breaking up text to make it easier for AI agents to "scrape" and categorize information.
- FAQ Sections: Designing content specifically around the prompts audiences are likely to use when asking an AI about a brand.
- Metadata Integration: Ensuring that key facts and figures are presented in a way that AI models can easily identify as authoritative.
This shift represents a fundamental change in the "audience" for PR content. Communicators are no longer just writing for journalists and consumers; they are writing for the algorithms that mediate how those groups find information.
Industry Implications and the Path Forward
The strategies outlined by Evans suggest a broader professionalization of AI use within the communications industry. As the "speed of technology" continues to accelerate, the "rules" of engagement are in a state of constant flux. The ability to adjust to these changes—not through ad-hoc experimentation, but through systematic infrastructure—will likely separate the industry leaders from those who struggle with the technology’s inconsistencies.
For many organizations, this requires a cultural shift. It moves the role of the communicator from "content creator" to "systems architect." The value of a PR professional in 2024 and beyond is increasingly found in their ability to curate institutional knowledge and build the frameworks that allow AI to execute that knowledge at scale.
"Everyone thinks they’re behind. I think I’m behind," Evans admitted, reflecting the pervasive sense of urgency in the field. However, the investment in building these systems is not a recurring cost but a "one- or two-time investment" that yields exponential returns in time savings and output quality.
Conclusion
As the communications industry prepares for upcoming milestones, such as Ragan’s PR Daily Conference in Brooklyn, the focus is clearly shifting from "what" AI can do to "how" it can be governed. The insights provided by Sarah Evans underscore a critical truth: the most powerful tool in the age of AI is not the algorithm itself, but the human-led systems that define its boundaries. By standardizing instructions, embracing redundancy, and treating institutional knowledge as a digital asset, communicators can overcome the limitations of current AI models and unlock a new era of organizational efficiency. The transition from 120-minute tasks to 12-minute tasks is not just a gain in speed; it is a fundamental reimagining of what a communications team can achieve.
