I’ve done it. You’ve probably done it. Pressed for time to complete a project, we copy-paste several pages of AI-generated content, lightly skim them, and ship it. This rule is #1 because even though we know better, professionals are breaking it constantly.
The Illusion of AI Perfection
AI writing tools produce polished, grammatically correct text that is often more well-written than what most of us can craft on our best day. The collective consciousness of large language models gives us easy access to a sophisticated writing expert. It’s far too easy to be tempted by the assumption that the content is suitable for our purpose.
AI’s tendency to create well-written nonsense comes down to one simple fact: AI does not understand your business, your relationships, the goals or motivations of your project, or any of the specific context it is writing about. As a result, the silliness that comes out of the current chatbots cannot be overstated.
Recent examples include:
- McDonald’s ending their AI drive-thru trial after customers had to stop the AI from adding hundreds of chicken nuggets to their orders
- NYC’s MyCity App encouraging businesses to break the law
- An AI-generated Halloween Parade event that duped thousands in Dublin into showing up for a non-existent celebration
Large Language Models are Fancy Auto-Complete
We should all think of LLMs as an incredibly sophisticated version of predictive text on our phones. The earliest version would typically complete a word based on the letters you had typed, with a context window of just the first part of the word. As this advanced, the context window broadened, and you could type “Going to the” and it would suggest “store” or “gym.”
LLMs are built on neural networks and are far more complex than predictive text but still rely on a context window. They are simply guessing the next word, or more accurately, the next “token.”
Understanding Tokens
Tokens are the bite-sized pieces that AI models use to process language – but they’re not exactly words. A token might be a full word like “cat,” part of a word like “ing” or “pre,” a piece of punctuation, or even a space. When you type “hamburger,” an AI model actually sees it as something like [“ham,” “bur,” “ger”]. This chunking system exists because it’s a more efficient way to handle language, especially across different languages and writing systems.
When you make a request of AI, it takes the context window (your request, previous requests in your chat, other prompt information provided by the chatbot developer) and uses an algorithm to guess the most likely next token. If its model tells it that the next word after “You should add chocolate to your” is “concrete,” it will gladly become the worst barista in history.
The Competitive Advantage of Careful Review
In a business world where AI usage is becoming universal, the ability to effectively review and refine AI-generated content is becoming a crucial competitive advantage. The executives who master this skill will stand out by consistently producing communications that are not just technically correct, but strategically sound and relationship-enhancing.
The “Every Word, Every Time” Protocol
Implementing a rigorous review process for AI-generated content isn’t just about catching errors – it’s about maintaining professional integrity and relationships. Here’s how to do it effectively:
First Pass: Content Accuracy
Read through the entire text focusing solely on factual accuracy. Check every number, date, policy reference, and business claim. Highlight items you need to verify. AI can be remarkably confident about completely incorrect information.
Second Pass: Context and Tone
Evaluate whether the content truly fits your specific situation. Does it match the history of your relationship with the recipient? Does it acknowledge any recent conversations or developments? Does it maintain your usual communication style? Essentially, does it sound like you?
Final Pass: Strategic Alignment
Consider how the message aligns with your broader business objectives. Ask yourself:
- Does this support our current strategic initiatives?
- Is this consistent with our brand voice and values?
- Couuld this content impact other business relationships or projects?
- Are there any potential legal or compliance implications?
Making It Manageable
This process might seem burdensome, but it will help you avoid costly and embarrassing mistakes. You can dramatically decrease the effort required for review by:
- Working in small chunks (we’ll cover this in another rule)
- Creating a personalized review checklist for the content you generate most often
- Setting aside dedicated review time for AI-generated content
- Using a “cooling period” between generation and review
- Having a second pair of eyes on critical communications
Conclusion
The sophistication of AI writing tools makes it tempting to skip thorough review, but that’s exactly why careful review is more critical than ever. By treating “Read Everything, Every Time” as an inviolable rule, you transform AI from a potential liability into a powerful tool that enhances rather than endangers your professional communications.
