I’ve been in the Learning and Development trenches for 11 years. I’ve seen the shift from clunky, Flash-based modules that took six months to build to the current era where I can spin up a storyboard draft in seconds using generative AI. But here’s the reality that most "AI evangelists" leave out: speed is rarely the goal if the tone is inconsistent.
If you’ve ever had a learner tell you, "The introduction sounds like a friendly human, but the assessment questions sound like a lawyer having a breakdown," you know exactly what I’m talking about. Maintaining tone consistency when AI is writing 40% of your course isn't just a stylistic preference; it’s a credibility issue. If your brand voice sounds schizophrenic, your learners will tune out.
In this post, I’m pulling from my "gotchas" doc—the one where I track every time a project went off the rails—to show you how to keep AI-assisted content grounded, human, and on-brand.
The Foundation: Defining Tone Before You Prompt
Most instructional designers fail at AI prompting because they ask for "a professional tone." That’s meaningless to a machine. To get consistency, you need a style guide for eLearning that is explicitly designed for an LLM (Large Language Model) to consume.
If your style guide is a 50-page PDF, the AI will ignore it. You need a "Voice & Tone Prompt Card." This is a concise snippet you paste into your system instructions every single time you start a new draft. It should include:
- The "Anti-Persona": Who are we definitely not? (e.g., "Do not use corporate jargon like 'synergy,' 'leverage,' or 'best-in-class.'") Sentence Structure Constraints: "Use active voice. Keep sentences under 15 words. Avoid passive constructions at all costs." The "Humanity" Check: "Use contractions. Address the learner as 'you.' If you find yourself using a semicolon, delete it."
Risk-Based QA: Why Everything Can’t Be Treated Equally
One of my biggest pet peeves is the "blanket QA" approach. If you treat a 2-minute microlearning on 'How to fill out a travel expense report' with the same rigor as a 'Sexual Harassment Prevention' compliance module, you are wasting your SMEs' time and your own bandwidth.
I use a Risk-Based QA Framework to determine how much human intervention is required for AI-generated text.
Content Type Risk Level QA Strategy AI Usage Compliance/Legal/Safety High Full Human Audit; Fact-check against policy source documents; Legal review. Drafting structure/outlines only. Soft Skills/Leadership Medium Review for tone/nuance; Ensure scenario validity. Drafting scripts and branching scenarios. Process/Software UI Low Spot-check for technical accuracy. Drafting step-by-step instructions.Fact-Checking and Source Tracking: The "Gotcha" Prevention
AI is a confident liar. It will hallucinate a policy update with the same tone it uses to describe a company mission statement. If you are using AI to draft content, you must implement a source-tracking workflow.
My rule is simple: If the AI generates a fact, it must be supported by a link to an internal source document. If I’m reviewing a draft, the first thing I look for is the reference. If there’s no source cited, it’s an automatic flag. I tell my team: "Trust, but verify." In L&D, we don't have the luxury of "oops" when it policy risk in ai generated content comes to technical accuracy.
Pro-tip: Use Retrieval-Augmented Generation (RAG) tools if your organization has them. Don’t ask ChatGPT to "explain our new HR policy." Instead, paste the policy into the chat and say, "Using only the provided text, write a summary for new hires." This drastically reduces the likelihood of the AI making up its own version of your internal processes.
SME Review: Moving Beyond "Looks Good"
If your SME replies with "Looks good," they haven't actually read it. They are just trying to get the task off their plate. As an ID lead, I’ve learned that if you ask for vague feedback, you get vague results.
When you send AI-assisted content to a Subject Matter Expert, don't just send a Word doc or a link to a prototype. Send a structured feedback request. Here is how I frame the review to keep the tone and content on track:
By forcing the SME to think about the learner's perspective rather than just reading for typos, you turn a passive review into a high-value collaboration. It also forces them to realize that the AI draft is a starting point, not the finished product.
The Editing Ritual: Why I Rewrite Every Sentence Five Times
My quirk—the one that drives my peers crazy—is that I rewrite one sentence five times until it’s perfect. When working with AI, you have to adopt this level of obsessive editing. AI tends to be wordy, passive, and overly optimistic. It loves to say, "It is important to remember that..." or "Understanding this process is a critical component of..."
Delete those filler phrases. They are the hallmark of AI-generated content and they scream "corporate buzzword vomit."
When I edit AI drafts, I focus on the "Readability Score." If a sentence has three commas, it’s too long. If it uses a word that doesn't appear in our company’s daily Slack conversations, it’s too formal. Your job as an Instructional Designer is to act as the curator of the AI’s output. The AI provides the clay; you are the sculptor.
Conclusion: The Human-in-the-Loop Mandate
AI can give you a draft in seconds, but it cannot give you the empathy required to teach someone a complex skill. It doesn't know the frustration a learner feels when a process is broken. It doesn't know the culture of your organization.
If you want to keep your tone consistent, stop treating AI as a "content generator" and start treating it as a "junior copywriter." You wouldn’t send a junior copywriter’s first draft to the VP without reading it yourself, would you? The same applies here. Keep your style guide tight, use your risk-based QA strategy, and never, ever hit "publish" without putting your own human eyes on the final product.
Now, go check your most recent draft. How many times did you see the word "comprehensive"? If it's more than once, you’ve got some editing to do.