For marketers and PR professionals, AI tools like ChatGPT and Google Gemini have quickly moved from novelty to necessity. When used strategically, they can dramatically speed up research, ideation, and first-draft creation. However, when used poorly, they can flood your content engine with generic, repetitive, and error-prone copy that dilutes brand credibility and customer trust.
This shift has been one of the biggest learning opportunities for me to navigate as a Content Manager at Merritt Group.
As an agency, Merritt Group was quick to adopt generative AI as a critical workplace enabler. We use it for everything from note-taking and setting up new project management boards to drafting messaging frameworks, design concepting, and fine-tuning content. We also frequently evaluate new tools and use cases through our cross-functional AI testing team, MG Labs. In doing so, we’ve discovered that how you use AI makes a big difference.
The most effective teams treat AI as a draft accelerator and thinking partner, not an infallible content machine. That means being intentional about how you prompt it and disciplined about how you edit what it produces.
Below are practical best practices for both sides of the equation: first, how to prompt LLMs to generate stronger V1 drafts, and second, how to edit AI-generated content into something accurate, differentiated, and genuinely valuable.
Part 1: How to Prompt AI for Better V1 Drafts
1. Garbage in, garbage out: One of the biggest mistakes marketers make with AI is asking it to “just write” about a topic and trusting it to pull the right information from the open internet. Instead, you should be grounding the model—secure, encrypted, and compliant versions, of course—with pre-vetted reference materials, such as brand guidelines; approved messaging or positioning frameworks; and curated news articles, research reports, or past content that tie into your chosen narrative.
I primarily use our agency’s ChatGPT business version for content generation and have had a lot of success with creating dedicated projects for each client or major project initiative. I recommend turning on memory in projects to ensure ChatGPT stays anchored to that specific project’s tone, context, and history. This helps cut down on the risk of hallucinations or off-brand language.
2. Break it down: AI also performs better when you guide it step by step—especially for longer-form assets like white papers, eBooks, or thought leadership pieces. Rather than asking the AI model to write an entire eBook in one shot, I recommend breaking the process down into bite-sized chunks. Start by providing vetted reference materials and asking AI to summarize the key takeaways. Then, define your target audience and outline their key motivators and pain points.
Once I feel like the model understands the basic framework, I move into actual content generation. Ask AI to draft one section at a time, explaining the purpose of each section and highlighting the specific stats, examples, or narrative points to emphasize. This approach gives you greater control over content structure and makes editing far more manageable, since you can refine individual sections instead of having to rework the entire asset.
3. Provide guardrails and negative constraints: Of course, telling AI what to do is only half the equation. It’s equally important to tell it what not to do. Effective guardrails might include instructions to avoid buzzwords or hype language, exclude certain talking points or outdated narratives, not speculate beyond the sources you provided, or steer clear of overly sales-driven messaging. These negative constraints help prevent generic, over-polished copy and keep the draft factual and aligned with your strategic intent.
4. Be explicit about voice, tone, and point of view: Even with strong source material, AI models will often default to a neutral, vaguely academic tone unless told otherwise. To avoid voice drift, specify key brand elements like point of view (first person vs. third person), desired tone (e.g., authoritative but conversational), and the audience’s familiarity with the subject matter (e.g., “Assume you’re speaking to CISOs with deep subject matter expertise in Zero Trust but limited knowledge of AI attack vectors.”)
Voice consistency is often where AI-assisted content falls apart most visibly, so this is an important step to get right. I also find it helpful to provide the model with a few examples of the writing style I want it to emulate, often by pulling publicly available blogs or social media posts.
5. Know when to re-prompt vs. when to edit manually: If a draft is fundamentally off-base—wrong audience, wrong argument, wrong structure—it’s usually faster to revise the prompt and ask AI to regenerate the section. However, if the draft is directionally sound and just needs tighter language or stylistic polish, manual editing is often more efficient.
The key mindset shift here is understanding that AI can help you get to a usable V1 faster, but humans are responsible for making it good. There should always be a human in the loop reviewing, editing, and validating AI-generated content before it goes anywhere near publication.
6. Re-ground AI during long or complex projects: For projects involving multiple prompts or extended timelines, I recommend periodically re-anchoring AI in the core context, including your target audience, main argument or thesis, and the overall purpose of the asset (for example, thought leadership vs lead generation). This helps prevent drift and ensures later sections remain aligned with earlier ones.
7. Organize inputs clearly for the model: Lastly, how you structure your prompt matters. Use formatting techniques such as quotation marks, separators, or labeled sections to clearly distinguish between reference materials and instructions. Clean, organized prompts reduce ambiguity and make it easier for the model to prioritize what actually matters.
I like to use three forward slashes (///) to distinguish between my prompt and the source material I want AI to reference.
Part 2: How to Edit AI-Generated Content for Impact
Of course, a strong V1 draft is just the starting point. The real value comes from how marketers and PR professionals refine AI-generated content into something accurate, differentiated, and credible. Here are some of my biggest tips on how to edit AI-generated content for maximum impact:
1. Fact-check everything: No matter how detailed your prompt or how thorough your reference materials, AI is still prone to hallucinations – it’s like a feature of the LLM architecture. When using AI to generate content, we as writers are responsible for the accuracy of the final output. That’s why every name, date, statistic, source, and factual claim must be verified, without exception.
AI models are confident writers, but confidence does not equal accuracy. This step is non-negotiable, particularly for bylined content, executive thought leadership, or media-facing assets.
2. Watch for false certainty: Similar to LLM’s tendency to hallucinate, AI-generated content also tends to present speculation as fact and predictions as inevitabilities. When editing AI content, I recommend flagging language that relies on absolutes like “will,” “always,” or “guarantees.” Where appropriate, soften claims, add attribution, or introduce nuance. This is especially important in emerging technology or rapidly evolving industries.
3. Eliminate repetition and circular arguments: When prompts are thin or overly broad, AI tends to restate the same idea in multiple ways rather than introducing new insights. As an editor, I’m constantly scanning content for repetitive sentence structures and arguments that say the same thing with slightly different wording.
When in doubt, cut aggressively. As any seasoned editor can attest, clarity almost always improves with fewer words.
4. Add differentiated, human insight: Generative AI is designed to produce content that appeals to the broadest possible audience, which means its outputs often mirror what the market is already saying. This is where human judgment matters most. Ask yourself, is this perspective overused or outdated? What’s the more interesting or contrarian angle? What experience, opinion, or context can we add that AI can’t?
AI also tends to favor shorter, bulleted sections over dense, nuanced analysis. When editing, look for tell-tale gaps. Are there unanswered questions? Does the argument need more depth or explanation? Would an example or anecdote improve understanding? Strategically expanding AI-generated content with your unique point-of-view can turn a generic draft into a compelling story. After all, differentiation doesn’t come from better prompts alone—it comes from editorial decision-making.
5. Re-humanize the language: Even polished AI copy can feel emotionally flat. Look for opportunities to add specific examples, personal opinions, and natural transitions that reflect how people actually think and speak. Sometimes, I’ll even read a block of text out loud to gut-check whether or not it sounds like something that I would actually say. This step helps ensure your final content resonates with your target audience.
6. Use AI for final optimization: Once you’re happy with a draft, feeding it back into the AI model for final tweaks can be useful. Ask for suggestions based on your target audience, desired outcomes, or distribution channel. At this stage, AI should be polishing—not steering—the narrative.
You can also use AI to pressure-test your thinking. Ask the model to identify potential objections from a skeptical buyer or journalist, flag areas where the argument feels generic or overused, and identify any unanswered questions in your current narrative. You don’t have to accept these suggestions, but they can help surface weaknesses worth addressing before publication.
The most effective marketers and PR professionals aren’t outsourcing thinking to AI—they’re using it to accelerate execution and enhance productivity.
When paired with strong prompts, clear guardrails, and rigorous editing, generative AI can dramatically reduce time to first draft while preserving (and even enhancing) content quality. However, it’s our job as writers to make the final judgment call on whether or not something is good enough to publish. After all, AI might be able to help you write faster, but only humans can make the content meaningful.
Ready to accelerate your content engine without sacrificing quality or brand credibility? Contact Merritt Group today to partner with our AI-savvy content strategists
