Prompt to insert content without sounding like AI

[Gossamer]

I’m trying to create a prompt with chatgpt-5 that can be used to review an existing blog post and add a new section with an H2 that seamlessly integrates with the current content. I’ve studied best practices to avoid it generating content that sounds like AI, but I’m not having any success.

Here is my basic prompt. What am I missing?

using content available at Top Email Threats: Combatting BEC, EAC, and Fileless Malware. please seamlessly incorporate a new section with an H2 on the benefits of using secure file transfers as a means of protecting sensitive data in transit and at rest. Limit new section to no more than two brief paragraphs. Be sure to write in a natural, human way using the same editorial/journalistic approach from the original article. Avoid the “if this, then that” / “by doing X, you get Y” cadence, and instead let the ideas flow in a more observational, narrative-driven style. This leans into texture and pacing, making it feel more natural and human, and aligning it with a journalistic/editorial voice. Maintain the same keywords and SEO best practices, but vary the sentence structure. Follow best practices for avoiding content that sounds like it was written by AI. Review these instructions carefully after writing your response to be sure you are following the instructions.

Here are the results:

The Role of Secure File Transfers in Safeguarding Sensitive Data

In an era where sensitive information is constantly exchanged through email and other communication channels, securing file transfers has become a critical layer of protection. Secure file transfer protocols ensure that data is encrypted both in transit and at rest, preventing interception or unauthorized access. As more businesses move toward remote work and cloud storage, the risk of exposure during file sharing grows, making secure transfer tools essential for protecting against data breaches.

These solutions not only encrypt files but also offer audit trails, adding an extra level of accountability. Without secure transfers, even the most robust email security measures can be undermined by vulnerabilities in file-sharing systems. In a world where the integrity of sensitive data is non-negotiable, secure file transfers are no longer a luxury—they are a necessity.

It is reported as 100% written by AI according to zerogpt.

[mcfinley]

zerogpt has a proprietary algorithm that identifies patterns that are typical in AI generated content.

Your prompt is asking the AI model to generate content… so it is going to trigger those patterns! Your prompt asks it to take a natural approach and follow the style of the original source, which it is doing really well. Bottom line, the model is doing the only thing it knows how to do, and doing it well.

If you want it to generate some novel content that breaks outside the patterns typically used by a generative AI, you need to do something that breaks out of the pattern of using generative AI. Your goal is to generate meaningful text that follows the style of the original paragraph but not in the way that AI normally does? Maybe ask it for 5 separate paragraphs, then choose two of them randomly? Maybe ask the model for logprobs and intentionally choose low probability tokens? Set temperature high? Use several different models and combine parts of the results of each of them?

The bottom line is that if you use a popular model to directly generate text, it is going to follow the patterns normally associated with that model and be detectable. The technique you are looking for is a form of Compound AI Systems.

[Tesla3D]

I apologize for mentioning a brand name.

In general, I use one of many writing tools to check the content in my favorite word processor or PDF editor. From there, that resulting document is re-submitted to ChatGPT for review. Then, once again, use one of the many writing tools to look over that output. My idea is to avoid the patterns normally associated.
As a ChatGPT subscriber, I feel that creating a Project, using bookmarks increases the ChatGPT using more of the terms and models. I typically write about more complex physics, technical or regulatory subjects to be read by undergraduate level readers.