5 Signs your Content is AI Generated & How to Fix It

Recently, I’ve been working with a client who wants to ‘speed up’ the publishing process to ‘get more content out there’. His argument was that we could use AI to crank out material faster and quicker than his competitors, get more traffic and so forth. Of course, it doesn’t work like that. As you’ve probably seen yourself, auto-generated AI content is very forgettable. It’s highly unlikely that anyone would share, recommend or sign up for more to read.

However, he does have a point. You can use an LLM to speed things up. It’s just that he’s trying to speed up the wrong things. Instead of using it to generate cookie cutter content, I’d suggest that you use it to refine different parts of the writing process.

PS – in the following articles I’ll go into more depth about this, so sign up for the newsletter and I’ll drop you a line when they’re ready.

Let’s take an example.

When you start a new project, there’s a temptation to write a simple prompt (i.e. an instruction to the LLM) and ‘hope’ that the LLM understands what you’re trying to achieve. This can lead to a lot of frustration if you have to edit, repeat, refine etc. the prompt.

Create a ‘Foundation’ Prompt

In prompt engineering, a “foundation” prompt (sometimes called a “seed” or “base” prompt) refers to the initial instructions given to an LLM that establish the fundamental parameters for how it should respond. This serves as the underlying framework that guides all subsequent interactions.

A foundation prompt typically contains:

• Role definition: Establishing who or what the AI should act as (e.g., “You are an expert financial advisor”)
• Core behavioral instructions: Setting fundamental rules for how the AI should respond (e.g., “Always provide balanced perspectives” or “Use simple language without jargon”)
• Formatting preferences: Defining how information should be structured (e.g., “Use bullet points for lists” or “Include headers for different sections”)
• Topical boundaries: Clarifying what subjects are in-scope and out-of-scope
• Tone and style guidance: Specifying the voice to use (formal, conversational, technical, etc.)

The foundation prompt effectively “primes” the model by setting the initial conditions that influence all its outputs. It’s the stable base that remains consistent while additional prompts or questions may vary. This is different from iterative prompts or follow-up instructions that build upon this foundation.

This means that when you start a new project – in this case, how to ensure your content feels less AI generated – create a ‘foundation’ prompt to ask an LLM for advice on how to write high-quality AI material.

It will be something like this:

‘As an LLM with deep knowledge of content generation, help me write a prompt that will I can give to an LLM to help me co-author content that avoids typical AI cliches in its responses’

See what I’ve done here?

Here’s the response from Claude:

“I want to co-author content with you that feels original, nuanced, and distinctly human. Please help me avoid these common AI writing patterns:

1. Avoid formulaic structures like “In this article, we will explore X, Y, and Z” or concluding with “In conclusion, we have explored X, Y, and Z.”
2. Don’t use predictable transition phrases like “It’s important to note,” “It’s worth mentioning,” or “Interestingly.”
3. Skip unnecessary qualifiers like “relatively,” “fairly,” and “quite” that weaken writing.
4. Avoid excessive hedging with phrases like “it could be argued that” or “some might suggest.”
5. Don’t overuse passive voice or corporate jargon.
6. Never start responses with “As an AI” or reference your limitations unprompted.
7. Present unique perspectives rather than balanced “on one hand, on the other hand” constructions.
8. Use varied sentence structures and lengths rather than consistently medium-length sentences.
9. Incorporate figurative language, cultural references, or analogies when appropriate.
10. Feel free to take clear positions on non-controversial topics rather than remaining neutral on everything.

So, what it’s doing now is providing me with guidelines on how I can ask any LLM (ChatGPT, DeepSeek, Google Gemini and others) to craft content that will avoid the giveaway signs you typically. For instance, words like ‘delving’ cropping up everywhere.

Refine prompt with another LLM

At this point, we could start using the instructions Claude gave us to start writing.

However, my preference is to use the guidelines from one LLM as input to another. In other words, I’ll put Claude’s guidelines into Google Gemini to see if it can further refine the prompt.

While this extra step takes a bit of effort, it means you’re getting multiple inputs from different ‘expert copywriters’ in the form of LLMs. Does that make sense?

Tip: This process of iterative refinement is highly recommended as it also ensures you avoid any bias or limitation in one of the LLMs. It’s like asking 5 different professors to critique your thesis. They’ll all provide different angles on the material you’ve submitted.

In Google AI Studio, I first inserted the guidelines from Claude, then asked the following:

‘As an expert copywriter who specializes in high quality in-depth articles, provide me with 5 guidelines on how I can use an LLM to write content that will have an original tone of voice, precise knowledge, and will avoid the typical cliches responses that suggest that the content has been AI generated.’

Google AI Studio, which tends to get overlooked as an LLM, provides a much richer set of instructions:

Guidelines for Using LLMs for High-Quality Content

Here are 5 guidelines to help you leverage LLMs to create high-quality, in-depth articles with a unique voice, precise knowledge, and that avoid the telltale signs of AI generation. These focus on how you *use* the LLM, rather than just relying on it to do all the work.

1. Define Voice & Audience Rigorously

• Beyond Basic Keywords: Don’t just ask the LLM to “write an article about [topic]”. Instead, explicitly define the desired tone, style, and target audience. Think of it like briefing a human writer.
• Specify Voice Attributes: Go beyond simple adjectives like “professional” or “casual.” Example: “Write in a style similar to [Author/Publication] – known for [Specific Trait 1, e.g., witty anecdotes], [Specific Trait 2, e.g., in-depth analysis], and [Specific Trait 3, e.g., a skeptical but ultimately optimistic outlook].” Or, define it yourself: “The tone should be authoritative but approachable, like a seasoned professor explaining a complex topic to an interested undergraduate. Avoid jargon unless absolutely necessary and always explain it clearly.”
• Audience Profile: Give the LLM a detailed profile of the intended reader. What’s their level of knowledge? What are their motivations? What are their likely pain points or questions? Example: “The reader is a marketing manager with 5-7 years of experience, familiar with basic SEO concepts but looking for advanced strategies to improve organic traffic in a highly competitive niche. They are skeptical of overly simplistic solutions and prefer data-driven insights.”
• Control the Length and Detail Level: Be very specific about the desired length and level of detail. “Write a 2,500-word article providing a comprehensive overview of [topic] with a focus on [specific aspect].”
• Iterate on Prompts: Don’t expect perfection on the first try. Refine your prompts based on the LLM’s output. If the tone isn’t right, be more precise in your next prompt.

2. Knowledge Verification: Become the Subject Matter Expert Curator

• Don’t Assume Accuracy: LLMs are impressive, but they can hallucinate (make up) information. Never publish content without thoroughly verifying the facts.
• Provide Source Material: Instead of just asking the LLM to research a topic, feed it reputable sources. This could be links to academic papers, industry reports, expert interviews, or even your own notes.
• “Contextualize” the Data: After you’ve provided the LLM with source material, prompt it to “Synthesize the information from these sources, focusing on [specific aspects] and highlighting any contradictions or areas of uncertainty.” This helps avoid simple regurgitation and encourages critical thinking.
• Instruct on Specific Datapoints: “Incorporate the statistic from [source] stating that [specific statistic] and cite it appropriately.”
• Implement a Robust Fact-Checking Process: Dedicate time to verify every claim, statistic, and quote generated by the LLM. Use multiple independent sources. Don’t just rely on the LLM’s confidence; verify everything yourself.

3. Structural Editing: Transform Output From “Draft” to “Finished Product”

• Treat the LLM as a Powerful First Draft Tool: View its output as a starting point, not the final product. Expect to do significant editing and rewriting.
• Focus on Structure and Flow: Does the article have a logical flow? Are the arguments well-organized? Rewrite sections to improve clarity and coherence.
• Add Original Insights and Analysis: This is where you truly differentiate your content. Don’t just regurgitate what the LLM has generated. Inject your own experiences, perspectives, and critical analysis. “Based on my experience working with [clients/projects], here’s how this concept applies in the real world…”
• Rewrite AI-typical Sentences: LLMs often use certain phrases and sentence structures that are dead giveaways of AI generation. Actively identify and rephrase these sentences to sound more natural and human. Look out for phrases like “in today’s world,” “delving into,” “key takeaways,” and overly formal or redundant language.
• Incorporate storytelling and human elements: The content will be more engaging if you add real-life stories, examples, anecdotes, and personal experiences. The LLM might not come up with these by itself, so you’ll have to add them.

4. Focus on Depth and Nuance: Go Beyond Surface-Level Summaries

• Challenge the LLM to Explore Counterarguments: Prompt it to “Identify potential criticisms or limitations of this approach.” This demonstrates a more balanced and nuanced perspective.
• Ask Probing Questions: Instead of accepting the LLM’s initial answers, ask follow-up questions to delve deeper into the topic. “Okay, but what are the underlying causes of that trend?” or “What are the ethical considerations?”
• Explore Unconventional Perspectives: Prompt the LLM to “Consider this topic from the perspective of [different stakeholder/field].” This can reveal novel insights and challenge conventional wisdom.
• Dive Deep into Specific Examples: Use real-world case studies and examples to illustrate the points you’re making. The more specific and detailed the examples, the better.
• Encourage Critical Thinking, Not Just Reporting: Push the LLM to analyze, evaluate, and synthesize information, rather than just summarizing it. Prompt it to “Compare and contrast these two approaches and explain their strengths and weaknesses.”

5. Implement a Multi-Pass Editing and Feedback Process

• First Pass: Content & Factual Accuracy.
• Second Pass: Tone and Originality. Eliminate AI clichés, inject personality.
• Third Pass: SEO optimization. Improve the article’s reach and search engine ranking.
• Fourth Pass: Readability and Flow. Have a third person who has never read the article edit it to make sure it is easy to follow, reads well, and keeps the audience engaged.

Recap and Next Steps

Here’s what we covered today.

• Create a foundation prompt: Before you start, ask an LLM to help you craft a ‘foundation’ prompt to get started.
• Use Multiple LLMs: Avoid bias and other limitations in the LLM of your choice by using a combination of LLMs to brainstorm, draft, review, and update your materials. Don’t rely solely on one LLMs. Remember, each LLM has its own default ‘accent.’
• Separate Tasks – Use different LLMs for different tasks. For instance, some excel at creating more ‘original’ angles/insights, others offer better proofreading skills, and others are better editors.

Instead of using an LLM as a ‘fast food’ content generator, look for ways to collaborate with it. Get a feel for which LLMs work best for different writing tasks. Perform a quick test by asking 5 different LLMs to create content from the same prompt. Once you see the difference, you can lean into the ‘best’ LLM.

If you follow these guidelines, your material will avoid the tell-tale giveaways you see in generic AI material. I hope you found this useful.

To get the next tutorial delivered to your inbox, sign up for the newsletter.

If you’re on LinkedIn, please connect over here.