Using LLMs for blogs: Cheating?
We ran the same prompt through four model aliases. One scored 33% AI on GPTZero. The others scored 100%. Here is what gives it away.
It's easy to prompt an agent or chatbot to write a blog. Why wouldn't you? Don't have the time? Not confident in writing? Better than what you could do? Create an internal and external version? I used to dread creating documentation, however that burden is now behind us!
We wanted to see how writing with AI differs to actual people, and if a hybrid approach is the way forward. We ran the same prompt through model aliases hosted on PrivateMind: Fast, Default, Reasoning, and Coding. The prompt asked each one to write a short paragraph about the best restaurants in NYC and why they stand out.
The results were revealing not because any model got it right, but because each one was wrong in a different way, and some sounded more human than others. Add in specific skills for blog writing, and it comes out pretty well. To the trained eye however, there are red flags everywhere. Common three-word phrases, directness, and the trusty em dashes.
The setup
PrivateMind routes requests through four aliases. Each maps to a different open-weight model running on our own infrastructure:
- Fast: Gemma 4 31B (NVFP4)
- Default: Qwen3 27B (NVFP4)
- Reasoning: Kimi K3
- Coding: GLM-5.2
The prompt was deliberately simple: Write a short paragraph about the best restaurants in NYC and why they stand out.
What gives it away
The tell tale signs are purely stylistic, and they are everywhere.
Fast read like a press release. "Incredible diversity," "uncompromising commitment," "irreplaceable cultural experience." Every phrase was a cliché. It named two restaurants and then spent the rest of the paragraph describing what restaurants are, rather than recommending any. The closing line, "capture the city's relentless energy and multicultural spirit on a plate," is the kind of sentence that appears in every AI-generated food post ever written.
Default did not name a single restaurant, not one. It wrote 150 words about "gastronomic innovation meeting deep-rooted tradition" and "unparalleled density of talent" without ever saying where to eat. It used an em dash to wedge in a clause about "hyper-local farm-to-table concepts." If you have read one AI paragraph, you have read this one.
Reasoning was the only one that sounded like a person. It named six restaurants, gave specific details (Eric Ripert, Thomas Keller, "since 1888," "three Michelin stars"), mentioned neighborhoods (Lower East Side, Greenwich Village, Midtown, Brooklyn), and ended with a concrete image: "a perfect slice eaten on a folding countertop." No em dashes. No generic abstractions. It had opinions, and it committed to them. Given this uses Kimi K3, there's no surprise it scored the best out of the LLMs we tested.
Coding was better than Fast and Default but still had the tells. It named five restaurants, which is what we wanted. But "seamlessly blends rich immigrant history with relentless culinary innovation" is a sentence no food writer would write. "Unparalleled gastronomic excellence" is a phrase no human would use in conversation. The structure made sense, the content was specific, but the voice was still a machine doing an impression of a food critic.
The patterns
Three things show up across every model except Reasoning.
First, generic three-word phrases. "Uncompromising commitment to," "incredible diversity and accessibility," "unparalleled density of talent." These are the LLM equivalent of filler words, and they always stand out. They sound impressive and carry no meaning. A real food writer might say "the pastrami is hand-sliced and worth the queue," not "an irreplaceable cultural experience."
Second, describing instead of recommending. Fast and Default both explained what makes NYC restaurants good in general. None of them said "go here, eat this, it is better than the place next door." A local who knows NYC will have opinions, whereas an LLM has descriptions.
Third, em dashes. Default used one. Em dashes are not wrong in general writing, but they have a clear signature.
Is it cheating?
Using an LLM to write a blog post is not cheating, but it should be used as a tool to help rather than just throwing out slop. Every paragraph above was curated by a model, and none of them are publishable without editing. Only the Reasoning alias produced something close to usable, and even that needs a once over.
This is a great example for people to understand when to use the right model too! You've got to be using frontier intelligence if you want it to sound somewhat normal in our experience. If it's for an internal document, you will get away with the cheaper, smaller models.
What this means for PrivateMind
This is the point of running multiple models behind one gateway. You route to the right model for the job: Reasoning for analysis, Coding for technical writing, Fast for quick drafts you plan to edit heavily, and Default for general-purpose work.
Every prompt on PrivateMind goes through the same gateway, the same audit trail, the same infrastructure. No data leaves your environment. The models are open-weight, running on bare metal you control. If a model gets a number wrong, you catch it in review. If a model sounds too AI, you rewrite it. The infrastructure does not solve the quality problem, but gives you the tools to manage it.
The real question was never whether using LLMs for blogs is cheating. The question is whether you know which model you are talking to, what it is good at, and where it will lie to you.
The raw outputs
For reference, here are the four paragraphs exactly as each model produced them. We used GPTZero to analyse the content for AI generation. Grammarly offer a generous free AI detector also.
Fast (Gemma 4 31B) - 100% AI Generation
New York City's dining scene is defined by its incredible diversity, ranging from historic landmarks to cutting-edge fusion spots. Establishments like Le Bernardin stand out for their uncompromising commitment to seafood precision and world-class service, while spots like Katz's Delicatessen offer an irreplaceable cultural experience through their legendary pastrami. The city's strength lies in its ability to blend global authenticity with innovative techniques, where a humble street cart's halal platter can be just as iconic as a Michelin-starred tasting menu. Ultimately, the best restaurants in NYC stand out because they capture the city's relentless energy and multicultural spirit on a plate.
Default (Qwen3 27B) - 100% AI Generation
New York City's restaurant scene is a vibrant mosaic of culinary excellence, standing out as a global capital where gastronomic innovation meets deep-rooted tradition. The city's best establishments distinguish themselves through an unparalleled density of talent and diversity, ranging from Michelin-starred institutions that redefine fine dining to unpretentious neighborhood gems that serve as cultural anchors in every borough. What truly sets NYC's top restaurants apart is their ability to capture the essence of the immigrant experience and the city's relentless energy, offering an ever-evolving array of flavors, from hyper-local farm-to-table concepts to authentic global cuisines, that ensure every neighborhood serves as a destination for discovery, making the culinary landscape as dynamic and essential as the city itself.
Reasoning (Kimi K3) - 33% AI Generation
New York City's restaurant scene is among the most celebrated in the world, spanning every cuisine and price point. Le Bernardin, Eric Ripert's Midtown seafood temple, has long held three Michelin stars for its exquisite French preparations of pristine fish. At the other end of the spectrum, Katz's Delicatessen has been hand-slicing its legendary pastrami on the Lower East Side since 1888, offering a delicious link to the city's immigrant history. Per Se impresses with Thomas Keller's refined tasting menus and views of Central Park, while Peter Luger in Brooklyn has set the standard for American steakhouses since 1887 with its iconic porterhouse. Even casual institutions like Joe's Pizza in Greenwich Village prove that in New York, culinary excellence exists at every level, from white-tablecloth tasting menus to a perfect slice eaten on a folding countertop. What unites these standout spots is their commitment to craft, consistency, and a distinctly New York sense of identity.
Coding (GLM-5.2) - 100% AI Generation
New York City's culinary scene is unmatched because it seamlessly blends rich immigrant history with relentless culinary innovation. Iconic institutions like Katz's Delicatessen and Peter Luger stand out for their unwavering commitment to time-honored traditions, serving legendary pastrami and dry-aged steaks that taste like the city's history. Meanwhile, world-class destinations such as Le Bernardin and Atomix push global boundaries with their meticulous, Michelin-starred tasting menus that elevate seafood and Korean cuisine to high art. What truly makes NYC's best restaurants exceptional, however, is their incredible diversity and accessibility; whether you are grabbing a quick, fiery bowl of cumin-lamb noodles at Xi'an Famous Foods or indulging in a multi-course extravaganza in Midtown, the city offers unparalleled gastronomic excellence at every level.
In an AI and automated era, there's something nice about doing it yourself. However, I don't know what I'd do without AI now.