I used ChatGPT's new settings to kill the AI voice — and it actually worked
by Ben Khalesi · Android PoliceWe all know how it goes. First is a big bubbly hello, and next, we’re delving into a rich tapestry of a basic topic. The default ChatGPT voice is easy to spot.
I kept trying to prompt my way around it and telling it to act normal or stop using emojis, but it never really worked. OpenAI recently introduced a Personalization dashboard that changes things.
After a few changes to the new toggles, the AI voice became less obvious. Here’s exactly what I changed.
Related
Google Gemini is awesome, but it needs to copy these features from ChatGPT
Why ChatGPT still feels more complete
Posts 8
By Rajesh Pandey
Choosing the right Personalization options is the first step to cutting fluff
Go to Settings, then Personalization. You’ll see sections for Base style and tone and Characteristics. If you haven’t touched it yet, everything will be set to Default.
In OpenAI’s world, Default is like a golden retriever that swallowed a corporate thesaurus. It wants to please you, it wants to use big words, and uses them with far too much enthusiasm. Let’s change that.
Base style and tone
By default, the base style and tone are neutral with a heavy dose of corporate-speak. I usually go with Efficient and adjust it depending on the task.
| Base Style Option | Algorithmic Behavior | Optimal Use Case |
| Efficient | Direct answers, brief reasoning and less fluff | Baseline for stripping verbosity and establishing a foundation |
| Candid | Highlights risks, blunt honesty, no sugarcoating | Effective for auditing, editing, and receiving feedback |
| Cynical | Sarcastic and critical | Useful for a non-robotic edge into text generation |
| Professional | Polished and heavily uses formal business conventions | Highly prone to generating the AI voice. Avoid authentic prose |
Characteristics
The second set of toggles controls the Characteristics. Set Warm and Enthusiastic to Less. High enthusiasm adds a fake tone that kills professional writing.
Excess enthusiasm feels forced and unnatural. Emojis and too many bullet points are another dead giveaway of machine-generated text.
The default ChatGPT loves to structure every single response as a five-point list with rocket ship and lightbulb emojis. Lowering the Emoji and Headers & Lists settings gets the bot to write continuous paragraphs.
Custom instructions force the AI to write more human-like text
The characteristic toggles are a great start, but the real deal happens in the Custom instructions box. They force the model to bypass its inherent token-saving laziness and the Latent Semantic Indexing terms that trigger AI detection software.
Avoid vague prompts like “be professional.” That retriggers the corporate jargon we’re trying to skip. The key is specificity.
Set clear expectations and boundaries, and the AI is far less likely to default to vague, bloated writing.
Use these instructions to start:
Never use delve, tapestry, robust, testament, realm.
Avoid overused corporate buzzwords such as synergy, leverage, cutting-edge, world-class unless explicitly requested.
Human writing has burstiness, with punchy sentences mixed with longer, complex ones. Machine writing sticks to uniform sentence lengths, which feels monotonous.
Paste this command to disrupt the algorithmic pacing:
Write this content with human-like burstiness. Mix short, punchy sentences with longer, complex ones. Avoid uniform sentence lengths that feel monotonous or algorithmic. Make the rhythm feel natural, as if a person were speaking or writing spontaneously.
Get rid of the robotic empathy and endless meta-commentary. Generative models avoid firm positions since their training discourages strong opinions.
Paste this command to cut the noise:
Write in a direct, confident voice. Avoid robotic empathy, overused comforting phrases, and endless meta-commentary about the writing process. Take firm, clear positions where appropriate. Do not hedge unnecessarily or qualify statements with phrases like ‘it seems’ or ‘one could argue.’ Keep the text focused on information, analysis, or argument, not commentary on itself.
If you feel like your prompts are still vague, ask the AI to help optimize them. Feed it your draft prompt and request it to make it more specific, add examples, or clarify boundaries.
Free ChatGPT users should manage memory to avoid confusion
If you’re using free ChatGPT, you might see a red warning saying “Memory 100% full.” This is a bottleneck.
Memory full of junk from previous chats confuses the AI, so you must curate it. Click Manage and remove the junk from memory. With a clean database, the AI prioritizes the relevant facts.
Even better, merge your rules. Extract the important facts, remove the scattered entries, and feed them into one Master Memory Entry. Bringing 100 scattered rules together in one block reduces memory usage.
AI will remain artificial, but you can guide it to sound better.
Even after fine-tuning every setting, banning the word “delve,” and giving the AI a custom backstory, the output is still not fully human, and it never will be.
Large language models lack the lived experience or real intuition of a human drafting an email on a Tuesday.
Despite these adjustments, some transitions may feel stiff or sentences may be too balanced. We’re only trying to tip the scale a bit.
The goal is to cut the fluff. Do what I did, and what you get is a much better drafting tool.