Is That a Robot Writing? When and how to use AI without losing your voice

[Earl Duque]

AI is here. AI is there. AI is everywhere. 

When it comes to writing, we have a new tool that promises to help: Artificial Intelligence. When generative AI started to appear a couple years ago, it felt like magic to me, but nowadays it can also feel like a trap.  

Honestly, nowadays, I almost always scroll past content that looks like it was completely AI-generated*. (* with a couple of exceptions, more on this later) 

How do we use it to help us create without losing our unique, human perspective? How do we balance assistance with authentic creation? 

In this post, we will cover when and how to utilize generative AI for text content and talk about (current) telltale signs that something is AI generated so that you can make sure your real content isn’t mistaken for a robot. 

This is a lot, Earl. I am allergic to walls of text.

I get it, this blog is more than 3000 words long. Here are some alternative ways to consume this information:

Podcast style (generated via NotebookLM, audio gaffs and all): 

Instructional handout: https://gist.github.com/earlduque/04bd46b2741912c15abf40ec2be913da 

Table of Contents 

1. Three Ways to Start with AI 
   1. For brainstorming 
   2. Building a Narrative Arc 
   3. Getting the Knowledge Out 
2. Using AI as a Collaborator, Not a Ghostwriter 
   1. One Powerful Question to Ask an AI 
   2. A Non-Negotiable Rule for Code Snippets 
3. A quick aside about also having grace when seeing AI content online 
4. A Deep Dive into Keeping Your Content Human 
   1. Why AI Sounds Robotic 
   2. Level 1: Macro-Level Flaws (Structure and Flow) 
   3. Level 2: Meso-Level Flaws (Rhetoric and Sentence Patterns) 
   4. Level 3: Micro-Level Flaws (The "AI Vernacular") 
   5. Level 4: The Human Touch (Tone, Voice, and Punctuation) 
5. Defining Your Relationship with AI 
   1. The Spectrum of Collaboration 
   2. Core Principles of an Effective Partnership 
6. What to Delegate and What to Own 
   1. Tasks You Can Safely Delegate 
   2. Guardrails: Responsibility and Ethics 
7. Conclusion 

Three Ways to Start with AI 

That initial hurdle of getting started is where I find AI to be most powerful. Again, it’s not about letting AI write for you, but letting AI help you build momentum. Here are three ways I do this: 

Brainstorming Your Topic 

AI shines the most when you have a general idea but don't know the best angle. I'll give it context: who I am, who my audience is, and what my goal is. Then I'll ask it to help me ideate. It might suggest different titles, angles, or additional topics I hadn't considered, all of which gives me a solid starting point. 

Building a Narrative Arc 

A list of facts isn't a blog post. The best content tells a story. I learned early on about things like the Hero's Journey (Wikipedia link) and how compelling a good story arc can be. Why are stories like Star Wars so popular? Because they follow a pattern that resonates with us. 

Your technical blog can do the same. I often ask an AI to help me structure my post with a story arc: 

• The Problem: Start with a relatable issue. 
• The Stakes: What does this problem mean for you and others? 
• The Solution: This is the core of your content. 
• The Resolution: Connect it all back and provide next steps or a call to action. 

Asking an AI to create a template with this kind of narrative structure helps ensure your post is engaging from beginning to end, not just a dry recitation of information. 

Getting the Knowledge Out 

This is my favorite technique, and it's exactly how this blog post came to be. I gave the AI a ton of context: my goal, my writing style, my audience, and what I wanted to cover. Then, I asked it to interview me. 

I told it: "Keep asking me questions and follow-up questions until you think you have enough information to take MY thoughts (not generate new thoughts, but take MY thoughts), and organize them into a proper blog post. Use MY language as much as you can." 

This process turns a daunting writing task into a simple conversation. It helps get the information out of my head and onto the page in my own words. 

Here is a template of my exact prompt that i use: https://gist.github.com/earlduque/cee7693d3151d37a509a9424d4f018b8 

In fact, I also use voice-to-text dictation services as if it’s a real spoken interview. It makes the blog sound even more like me because the responses that it’s using for the blog are my actual speaking style. Check out this video from me and Travis about this process: 

Using AI as a Collaborator, Not a Ghostwriter 

Once you have a draft, the temptation is to let AI "clean it up." Be careful here though because this is where your voice can get lost. I try to utilize my own writing as much as possible, but I do use AI for a very specific kind of feedback. 

One Powerful Question to Ask an AI 

After I've written my draft, I give the entire text to an AI and ask one simple question: "What did I miss?" 

This is so much better than "make this better" or "edit this" or “expand this.” It doesn't rewrite my content or change my voice. Instead, it acts as a gut check. It might recommend things I can expand on or point out a relevant topic I completely forgot to include. It helps me see my own blind spots while keeping me in the driver's seat. I also use it to check for tone or idioms that might not land with our global audience. 

A Non-Negotiable Rule for Code Snippets 

AI is incredibly good at writing code. But you should never just copy and paste an AI-generated code snippet into your content without testing it yourself. 

If you're just posting code that anyone could have generated with an AI chatbot, what's the point of you posting it? Your value as an author comes from your credibility. You need to vet that code, run it, and make sure it works exactly as you expect. You are the one vouching for it. Use AI to check your code for errors or style, yes, but always, always test it yourself. 

A quick aside about also having grace when seeing AI content online 

Earlier, I said that I scroll past clearly AI generated content now, but this is a perfect time for me to clarify what I meant. 

To me, it's obvious when AI was used to 100% generate content from a simple prompt that is a few words. Nowadays, I can tell if someone still infused their own voice and/or stories/anecdotes that make it unique and why it warrants my attention.  

I want to acknowledge that AI is very helpful to our sector because of that “global audience” that I just mentioned. An entire post may feel like it’s AI generated but part of the usefulness of AI is that it breaks down two barriers: language and writing ability/experience. 

So, as we go into “signs something was written by AI”, it’s important to acknowledge that I do not mean to say that something that is mostly AI sounding is inherently bad, but the lack of personal voice, experience, and the human posting it, is what makes this content skippable. 

A Deep Dive into Keeping Your Content Human 

As time goes on, it's becoming more and more obvious when content is 100% AI-generated. It has a certain "feel" that sets off a lot of our BS detectors now. To get good at fixing this in your own workflow, you first have to understand why it happens and then learn how to spot the tells. 

Why AI Sounds Robotic  

That's how LLMs work. 

At its core, an LLM just predicts the most statistically probable next word. This is why its output is inherently predictable, repetitive, and generic. It’s leaning on common patterns from its training data, like the formal language of corporate press releases. 

And if thousands of journals and articles and educational studies are available online and open licensed to be used in training data. Bingo bango, suddenly the LLM is thinking that’s how everyone wants to speak (at least when it defaults to responding “professionally”). 

And the absence of self. 

AI can mimic human language, but it can't replicate lived experience, emotion, or vulnerability. So, we encounter the "uncanny valley" effect: the grammar is perfect, but the text feels lifeless and lacks personality. The "imperfections" of human writing (short-hand, idioms, silly phrases, slang, etc.) are now valuable signals of authenticity. 

Many of us have seen how generative AI loves em dashes, almost to the point that it’s a meme and clear giveaway that it is AI generated. But I also noticed that AI loves using the “It’s not just X, it’s Y” sentence structure. I did some research (along with asking Gemini and ChatGPT to do deep research too), and asked AI to compile all my notes to this multi-level framework of things to watch out for and how to fix them (as of September 2025): 

Side note: I also think it’s funny because you can tell the following section is obviously AI-generated. 

Level 1: Macro-Level Flaws (Structure and Flow) 

This is about the big-picture architecture of your post. 

How to Spot It: 

• Formulaic Introductions: The text opens with a rigid template, like identifying a "pain point" before hinting at a solution. 
• Predictable Outlines: The document follows a rigid, step-by-step format with an overabundance of headings and numbered lists, sacrificing flow for efficiency. 
• Redundant Conclusions: The conclusion is just a rehash of the introduction or a simple summary, offering no new insight. 
• Clunky Transitions: The connections between sections feel "stitched together" rather than flowing as a cohesive narrative. 

How to Fix It: 

• Deconstruct the Hook: Rebuild the intro around a surprising statistic, a provocative question, or a compelling micro-story. 
• Restructure the Narrative: Break the rigid outline. Re-sequence sections to build suspense or follow a more engaging story arc. 
• Fortify the Conclusion: Instead of summarizing, synthesize your points into a powerful final insight or a forward-looking statement. 
• Weave Seamless Transitions: Rewrite the connections between paragraphs to ensure a logical and elegant flow. 

Level 2: Meso-Level Flaws (Rhetoric and Sentence Patterns) 

This is about how your sentences are built and how they sound. 

How to Spot It: 

• The Emphatic Contrast: Overuse of the predictable phrase, "It's not just X, it's Y." 
• Syntactic Over-Complexity: A heavy reliance on verbs ending in "-ing" and "nominalizations" (e.g., "robustness" instead of "is robust"). The AI is mimicking complex academic writing without the nuanced thought behind it. 
• High Symmetry: Too many sentences follow the same template, such as "By doing X, you can achieve Y." 
• Overused Conjunctions: A repetitive rhythm created by pairs like "whether...or" and "not only...but also." 

How to Fix It: 

• Vary Syntactic Rhythms: Hunt for and eliminate repetitive sentence templates. Convert long "-ing" clauses into crisp, direct statements. 
• Prioritize Clarity Over Complexity: Replace nominalizations with active verbs to make your writing more engaging. 
• Replace Rhetorical Templates: Build a substantive argument with original phrasing and evidence instead of relying on formulas. 

(Also, this entire section is obviously AI because AI likes to construct information in the format of “bulleted list item with a title followed by explanation.” Not every list needs a summary for every single line!) 

Level 3: Micro-Level Flaws (The "AI Vernacular")  

This is about the specific, overused words and phrases that are dead giveaways. 

Hilarious and fun stats about the over-representation of specific words and phrases that appear more frequently in AI writing than human writing: 

• The phrase "play a significant role in shaping" is used 182 times more often by AI than by humans. 
• The phrase "today's fast-paced world" is used 107 times more frequently by AI than by humans. 
• Present participial clauses (verbs ending in "-ing") are used by LLMs at a rate two to five times higher than in human text. 
• Nominalizations (nouns formed from verbs or adjectives, such as "robustness" or "development") are used at 1.5 to 2 times the human rate. 

Sources: https://www.pnas.org/doi/10.1073/pnas.2422455122 and https://gptzero.me/news/most-common-ai-vocabulary/ 

How to Spot It: 

• Flowery Buzzwords: An overabundance of words like "delve," "harness," "leverage," "crucial," and "furthermore." 
• Overused Clichés: Generic opening phrases like "In today's fast-paced world..." 
• Inflated Motivational Phrases: Exaggerated statements like "stands as a testament" or "plays a vital role." 
• Corporate Jargon: A high frequency of terms like "seamlessly integrated" and "game-changing." 

How to Fix It: 

• Replace Jargon with Precision: Eradicate generic adjectives. Show, don't tell. 
• Write with Conviction: Eliminate filler words and hedging to create a direct and confident tone. 
• Pursue Original Phrasing: Actively remove clichés. Challenge yourself to express common ideas in a fresh, memorable way. 

Level 4: The Human Touch (Tone, Voice, and Punctuation)  

This is about injecting your personality back into the text. 

How to Spot It: 

• Polished but Lifeless Tone: The prose is grammatically perfect but feels sterile, neutral, and without personality. 
• Absence of "Rough Edges": The writing lacks the slang, contractions, humor, and personal asides that make a voice feel real. 
• Lack of Human Perspective: There are no personal anecdotes, real-world examples, or genuine emotion. 
• Punctuation Quirks: Look for an obsession with em dashes (—) and formulaic emoji use (🚀, ✨). 

How to Fix It: 

• Inject a Point of View: Establish a clear stance—be skeptical, enthusiastic, or instructional, but be something. 
• Ground Abstractions in Reality: Weave in personal stories, case studies, or real-world examples. 
• Adopt a Natural Cadence: Use contractions and conversational language where appropriate. 
• Punctuate with Intent: Use punctuation as a tool for style, not just correctness. Ensure any emojis or bolding feel like a deliberate authorial choice. 

Defining Your Relationship with AI 

So far, we’ve been talking about written copy in terms of things like social media, blog posts, articles, and journalistic mediums. I want to aknowledge though that using AI to do most of the work has its place. You just have to be intentional about when and how much to use it. You need to be strategic: 

• AI as an Assistant (Human-Led): This is for low-stakes tasks. You have full creative control. Think of using a tool like Grammarly to catch typos. You accept or reject the suggestions to ensure your voice remains intact. 
• AI as a Co-Creator (Mixed-Initiative): More of a partnership. You provide a detailed prompt (note that I said “detailed.” Otherwise, it’s garbage-in, garbage-out), and the AI generates a starting point, like a blog outline or a list of ideas. You then take that as a starting point and shape it with your own insights and experience. 
• AI as an Automator (AI-Driven with Human Oversight): This is for repetitive, high-volume tasks. A great example is what my team often does: after a livestream, which is 100% original human content, we feed the transcript to an AI and ask it to create a blog post. This allows us to reach a whole new audience without the massive overhead of writing it from scratch. But even then, a human always reviews and refines the output. 
  (Btw, the exact prompt we use to do this is here: https://gist.github.com/earlduque/26585a744265f40ad336d934d57be84a)
  Or for the developer audience, AI is really helpful when it leads the charge with things like summarizing release notes or generating a massive amount of technical documentation. 

Across all these models, three principles are key: 

1. Generic prompts get generic results. The more context you give the AI about your goal, audience, and tone, the better the output will be. You need DETAILED prompts, otherwise it’s garbage-in garbage-out.
2. Know AI's limitations. It can hallucinate (AI-speak for making random stuff up) and generate false information with complete confidence. Trust it for brainstorming, and demonstrate rigor with your verification and proof-reading processes. 
3. Treat it like a conversation. If you don't like the first response, provide feedback, ask follow-up questions, and refine your prompts to steer it in the right direction. 

What to Delegate and What to Own 

So, what parts of your workflow can you safely "AI-away" for maximum efficiency? 

Tasks You Can Safely Delegate (with Human Review): 

• Summarizing: Feed it long documents or transcripts to get the key takeaways. 
• Outlining: Give it a topic and let it propose a structure. 
• Drafting boilerplate content: Use it for first drafts of things like release notes or standard descriptions. 
• Rephrasing: Ask it to simplify technical jargon or convert a formal draft to a conversational tone. 
• Content repurposing: Take your final, human-approved blog post and ask it to create social media snippets or an email summary. 

Guardrails: Responsibility and Ethics 

Delegation is not abdication. You are always the final owner. 

• Fact-check everything: The human must be the final reviewer of truth. Treat every stat, date, and claim from an AI as unverified until you've explicitly confirmed it. 
• Always seek originality: AI learns from the internet, including copyrighted material. The only true path to originality is to infuse the content with your own new insights, unique experiences, and distinct voice. That is something that cannot be plagiarized. 
• Watch for Bias: AI models can inherit biases from their training data. It's your job to ensure the content you publish is fair, inclusive, and represents diverse perspectives accurately. 

Conclusion: Your Voice is Your Value 

I firmly believe that if you're writing content from the beginning, it should be as original as possible. Your language, your perspective, your story. That is the whole point. 

Everybody has a unique perspective that should be present in everything they do. Your experiences, your background, the reason why you're writing will make any content you create unique. Even if it's a technical subject that's been covered a thousand times, your individual experience makes it fresh. If you let an AI write the whole thing, what was the point of the post being authored by you? 

AI is a phenomenal tool. It can help us overcome hurdles, be more efficient, and get our ideas out into the world. But you must be the author. You are the expert. Your voice, your experience, and your willingness to share are what make this community great. 

Use AI to start the conversation. Use it to find your blind spots. But always, always make sure the final product sounds like you and reflects your own understanding. Otherwise, what’s the difference between reading from you and reading from an AI chatbot?