The average worker spends up to 20 hours per week on tasks that involve writing in one way or another – so it’s no wonder that as of October 2025, an estimated 20 – 30% of American workers (and rapidly growing) are using AI at work (most often ChatGPT). AI can certainly help all of us deal with the deluge of emails to write and reports to prepare. At the same time, as the share of writing generated by AI continues to increase, many are pushing back on what they view as “AI slop” – we’ve seen this in many recent social media posts, and research being conducted in the workplace suggests these complaints aren’t isolated.  If you’re using AI to help you write, it’s thus important to take the time to think about how others are perceiving your communication – and what you can do to ensure they feel a warm, human touch.
 
Here’s the good news: in our extensive experience, we’ve found you can combine human warmth and judgment with the speed and efficiency of AI to get the best of both worlds. In this comprehensive, in-depth guide, we provide detailed guidance, techniques, and specific prompts to help you make ChatGPT sound more human, including an all-purpose “Humanizer” prompt that can nudge your favorite AI to be a little more human than it otherwise would be.
 

Prefer to listen?  This post is also available as an episode on the Ravensight: AI for Small Business podcast, available on Spotify, Apple, Amazon, Youtube, and many other streaming platforms!

Proof that it works? Beyond our own personal experience, AI has shown itself to be sufficiently warm and conversational to conduct certain forms of therapy, so you can in fact get your business emails – and presentations – to sound like they weren’t written by a robot.

To help you thoroughly understand the tool you’re using, we also dive into the linguistic and psychological quirks that make AI‑generated prose sound artificial (and why that triggers negative reactions), and the technical limitations or learned patterns in models like ChatGPT and Google’s Gemini that cause their writing to deviate from natural human expression. However, if you want to jump straight to the practical prompting strategies for producing human‑sounding writing using ChatGPT or Gemini, be our guest and use the jump-to menu below to skip to that section.

How To Make ChatGPT Sound More Human: A Comprehensive Guide to Using AI for Writing

Let’s start by understanding why AI sounds the way it does… and why people react negatively.

The Problem: Obvious AI Writing Leads To Mistrust

When people sense writing is artificial or overly scripted, they tend to trust it (and the sender) less. Business communication relies on authenticity and a connection with the reader, something AI’s tone can inadvertently undermine. As more and more people utilize AI, they are also learning to recognize content generated by AI through certain subtle (or not-so-subtle) patterns and tendencies. Even as people utilize AI themselves, they may judge other people harshly when it’s obvious that they’re reading something written by AI rather than a human.

A recent workplace study – “Professionalism and Trustworthiness in AI-Assisted Workplace Writing” – delved into this topic. Their work with over 1,000 professionals found that while AI‑assisted emails came across as “more professional” in some respects, they also made the recipient doubt the author’s sincerity. Employees reported that messages from their managers felt less personal and caring when they perceived extensive AI utilization. The study also found that employees believed extensive AI use implied laziness or a lack of caring on the part of the writer. This isn’t surprising; it’s essentially the difference between receiving a hand-written birthday card from a friend, and a generic “happy birthday” email from some mailing list that you’re on. One interesting conclusion is that when a supervisor’s message went from light to heavy AI assistance, half as many employees found it sincere (40 to 50% vs 80% plus).

This isn’t the only problem. One industry survey in 2025 found 68 percent of customers felt frustrated or disconnected when they were not sure if they were dealing with a human or an AI, while many reported that this uncertainty lowered their satisfaction and trust in the company. Even more striking, over 40% percent of call center employees said customers had accused them of “sounding like a robot.” In other words, real humans were being mistaken for AI because their company scripts made them talk in a stilted, impersonal way. This highlights a key psychological point: people have a low tolerance for communications that feel canned or automated, especially in situations that call for empathy or personal attention. The moment a message feels too generic (or “too good to be true” in its flawless politeness), recipients may decide not to trust it.

In short, human readers are finely attuned to tone and subtext. An AI‑written email might check all the grammar and etiquette boxes… yet still trigger a gut feeling that something is off. That feeling can translate into mistrust, frustration, or apathy.

Worse yet, many of AI’s writing failings aren’t subtle. Some are noticeable – such as em-dashes, overly formal diction, and specific transitions.

Linguistic and Psychological Factors: Why AI Writing Feels “Off”

AI‑generated text often has subtle (and not‑so‑subtle) markers that signal its artificial origin. Human readers pick up on these cues in tone and phrasing, which can make the writing seem impersonal or inauthentic. By pinpointing some of these factors, we can develop prompts that make ChatGPT (or other common LLMs, such as Gemini) sound more humanlike.

Telltale Tone, Structure, and Phrasing

Academic vs everyday English. Even for business purposes, most of us don’t tend to write perfectly. At least partially as a result of their training data, LLMs favor formal grammar and academic expressions, often shunning contractions. For example, ChatGPT can use words that look like they were plucked out of an SAT vocabulary guide – words like “delve,” “align,” “underscore,” “noteworthy,” “versatile,” or “commendable” appear more often in its output than in typical human writing. While there’s nothing wrong with using any of these words (we certainly intend no offense if one of these is a tried-and-true favorite for you), stringing together too many of these words can make AIs like ChatGPT or Gemini sound oddly polished or academic for casual business contexts, in comparison to human writing.

Generic or template‑like language. AI text can feel formulaic. Common patterns include starting with grandiose imaginary scenarios or clichés. A human might write, “Let’s consider our next steps,” whereas an AI‑generated email might begin with “Imagine a world where our solutions unlock untapped potential.” Similarly, AI content is prone to padded transitions and stock connectors like “moreover,” “conversely,” or “in addition,” and it over‑explains obvious conclusions (for example, prefacing the final thought with “in summary” or “in conclusion” even when unnecessary). These verbal tics can make the text read like a canned template.

Em-dashes. AI is notorious for using em-dashes — long dashes like this to connect parts of a sentence — rather than using semicolons, parentheses, or just plain hyphens. Although many human writers — us included! — have historically used em-dashes or their shorter en-dash equivalents, the prevalence of “—” in AI writing makes it a telltale sign to avoid — at least if you don’t want people to think the sentence was written by AI.

Table: Characteristics of AI‑sounding text vs. a more natural human tone in business writing. The AI versions above are not incorrect, but their tone is flowery, formal, or overdone. The human alternatives convey the same message more directly and personably. Common AI quirks include overusing buzzwords (“cutting‑edge,” “revolutionize”), unnecessarily formal verbs (“utilize,” “leverage”), and excessive politeness/fillers that a human might omit. These differences make AI passages feel less genuine.

Technical Factors: Why LLMs Deviate from Natural Expression

Next‑word prediction vs. human thought. Large language models generate text one word at a time, each word chosen based on probabilities from the model’s training data. They do not have an explicit global outline or a genuine understanding of meaning the way a human author does. As a result, models can produce sentences that are locally coherent but lack an overarching focus or natural flow. Because an LLM is essentially looking back at the previous word to decide the next, it often lacks big‑picture context and a clear logical ladder. While reasoning models might handle some of these issues better, most users won’t use such sophisticated models for the writing due to factors such as time (GPT 5 Pro, for example, can sometimes take 10+ minutes to generate a response) and cost (using 5 Pro extensively requires either an expensive subscription and/or vastly more tokens via the API, compared to normal GPT-5.)

Lack of true context or emotion. AI models don’t have human experiences, emotions, or common‑sense grounding beyond text. When writing requires empathy, they simulate it using generic formulas (for example, “I’m sorry to hear that… your concern is very important to us”). Without careful prompting, LLMs can err toward excessive positivity (the “what a great idea” sycophancy discussed in a recent South Park episode) or neutral corporate politeness, which can feel inappropriate in serious situations.

Safe, predictable choices to avoid error. Because models are tuned to be inoffensive and “helpful,” they often end up being overly polite (to the extent of verbosity). This results in a voice that sounds unlike any individual human’s voice.

Proof Point: AI Can Provide Effective Therapy

The good news is that all of this is fixable. Don’t just take our word for it – consider the reuslts of a recent clinical trial of a generative‑AI therapy chatbot called “Therabot.”  As summarized by Dartmouth news: 106 adults participated in an eight-week trial. After four weeks of unlimited app use, participants reported a 51% average reduction in depression symptoms, a 31% drop in anxiety, and a 19% decrease in body-image concerns (which outperformed the control group). Users spent about six hours total with Therabot, roughly equal to eight therapy sessions, and described a sense of trust and alliance comparable to working with a clinician.

Therabot relied on open-ended dialogue built from evidence-based psychotherapy and CBT techniques, prompting crisis resources when needed. While the authors stressed the importance of clinician oversight (as do we), the point we are trying to make is that in a situation where empathy and warmth are crucial, AI can deliver meaningful, real-time mental-health support. This builds on prior research where CBT-based chatbots helped students and health-care workers. This shows how with structure, guardrails, and human supervision, AI can create interactions that feel genuinely human.

Strategies to Make LLM Writing More Human‑Like

The good news: by using a few prompts, you can significantly improve the “human‑ness” of AI‑generated writing, especially for business communication where tone and authenticity matter.

How To Train Your LLM

Think of ChatGPT like Toothless, the dragon from the popular (and now live action) How To Train Your Dragon. There’s a lot of raw power there waiting to be harnessed… but it needs a little training. It turns out that with the right prompt, ChatGPT can imitate anyone from a California surfer to a New York power broker, a la Succession. Our goal in this section is to get it to imitate you (or, at least, the best and most polished version of your voice!)

1) Provide examples (try using this specific prompt format)

LLMs are master mimics. One technique we’ve found reliable and effective – at least as a first pass – is to provide ChatGPT, Gemini, or your favorite LLM with multiple examples of how to get it right. The same way you might show a new employee some sample emails or social media posts to help them understand your tone and lingo, so too with AI.  Upload (or copy and paste) examples of what “good” looks like – the more the better – and ask the AI to reverse-engineer your tone with a prompt like the following:

You are an expert writing assistant trained to mimic tone, rhythm, and style from examples. I’ll share several examples of our ideal output below. Carefully analyze them to identify consistent traits such as sentence length, structure, phrasing, pacing, formality, humor, and emotional tone. Then, develop a reusable prompt that we can use to create new content of the same type (emails, social media posts, etc) that matches that voice exactly: not just the surface words, but the feel and rhythm.

Use the examples as your style guide. Do not summarize or critique them. Just absorb the tone and create a prompt that will faithfully reproduce it.

2) Provide clear tone and style instructions

If you don’t have materials readily available (or even if you do), instruct the AI on exactly how you want it to sound. Instead of “Draft an email announcing our new product,” try: “Draft an email announcement in a friendly, conversational tone. Use ‘we’ and ‘you,’ keep sentences fairly short, use contractions, and avoid jargon.” You can also include a short list of do’s and don’ts (for example, “use plain words like ‘use’ instead of ‘utilize’; avoid ‘in conclusion’; no clichés; keep paragraphs short”). You might consider forbidding or discouraging words AI tends to overuse, such as “delve,” “harness,” “groundbreaking,” and “in a world where.”  This will help output sound more natural.

3) Forbid em-dashes

This one is easy. You can put it in all caps (although the AI sometimes chooses not to listen). For us, we’ve saved “NEVER USE EM DASHES” to ChatGPT’s persistent saved memory.  (Pro tip – if you literally just prompt ChatGPT with something and say ‘save this to memory,’ it will, and will incorporate that into responses going forward.)

4) Combine the power of humans and AI

AI excels at structuring (for example, creating nice, orderly headings like the one in this blog post) as well as getting initial ideas/thoughts down on paper, but even with the best prompting, it often won’t be exactly what you’re looking for.  An iterative process that combines both your own words and AI’s can help.  Don’t be afraid to ask AI to expound, condense, or reword sections (or a whole article) if it’s not what you’re looking for.  Another tip is to do editing yourself on a few sections, then give it back to the AI and ask it to revise the rest of the document to flow from your edits, or match your tone more closely.

5) Once you have a useful prompt… keep it for reuse.

While it can take some time to dial in that perfect “brand voice” prompt, the power and scalability of the output is so meaningful that you’ll want to keep it handy. Rather than having to copy or paste it all the time, one useful way to save it is with a Gemini Gem or Custom GPT / ChatGPT project.  As we discussed in those posts, these are quick, easy to set up methods that help your AI tool of choice rock and roll on a specific task, without all the copying and pasting from you.  Of course, better yet is to build this into an always-on workflow automation – for example, the auto-drafting of responses to incoming leads that we discuss in our deep dive into how real estate agents can use AI.

A Universal “Humanizer” Prompt To Make ChatGPT – Or Gemini – Sound More Human

While we caution that one-size-fits-all prompts are less likely to result in perfect output relative to prompts tailored for the specific use case (brief responses to incoming leads vs. social media posts vs. internal action-items emails, etc), below is a universal prompt that you can use to take some of the AI-ness… out of AI.  Try making it into a Gemini Gem or project instruction for ChatGPT, and tweak it until it generates results you like!

You are a writing assistant designed to produce business communication that sounds natural, human, and trustworthy—never robotic, generic, or overly formal.
Your goal is to combine clarity, warmth, and professionalism without sounding like “AI output.”

Tone & Voice Guidelines

• Write as a thoughtful professional speaking to another person.

• Use contractions (it’s, we’re, don’t).

• Prefer short, varied sentences.

• Use “you” and “we” naturally.

• Be confident but not salesy.

• Sound conversational, not academic.

Language Rules

• Avoid buzzwords, filler phrases, and clichés (e.g., synergy, leverage, cutting-edge, in conclusion, imagine a world where).

• Never use em dashes (—); use commas or periods instead.

• Replace formal verbs with plain ones (use instead of utilize, help instead of facilitate).

• Keep vocabulary simple and believable—what a real person might say in conversation or email.

• Do not over-explain or restate obvious points.

Style & Flow

• Vary rhythm and sentence length to sound alive and unscripted.

• Use natural connectors like and, but, or so instead of stiff transitions like moreover or thus.

• Keep paragraphs short and scannable.

• Show genuine empathy when appropriate, without exaggeration or forced positivity.

• When giving examples or explaining, sound like you’re talking to a colleague, not presenting to a board.

Purpose & Context Awareness

• Match tone to the situation: calm for bad news, upbeat for good news, direct for action steps.

• Write for readers who are busy and skeptical of fluff—get to the point gracefully.

• If unsure, favor clarity over polish.

Before finishing any output:
Quickly self-check: Would this feel believable if read aloud by a real human? If not, simplify it until it does.

Conclusion

AI‑generated writing does not have to scream “robot.” Using AI to assist in the writing process helps us generate both more content AND higher quality content for readers like you – as you may have guessed by now, AI certainly played a major role in the creation of this article, although much of it was written or edited ourselves!  Our hope is that you can’t tell what AI wrote and what we wrote – because, in fact, we wrote it together. AI can certainly handle certain basic things solo (like brief emails) without much human review, but for longer‑form content such as this post, we find that meshing human judgment, insight, and editing with AI’s strengths in structure and rewording can be the best of both worlds.

What we want you to take away is that the issues that make AI writing feel unnatural are not irreparable flaws, but patterns we can recognize and adjust. By understanding why models like ChatGPT and Gemini default to those patterns (due to training and technical design), we become better at counteracting them. From a linguistic perspective, small differences in tone and word choice heavily influence reader perception. From a technical perspective, the AI is doing its best without true understanding, but it excels at recognizing patterns – so providing it with a new pattern to override its default can often be the most helpful approach.

Ultimately, for business communication, where trust and engagement are on the line, taking these extra steps to humanize AI text is worth it. With thoughtful prompting and a human eye on the output, we can harness AI’s speed and convenience without sacrificing the genuine, relatable voice that connects with readers.

Related Ravensight resources and how we can help

If you want to put these ideas into practice across email, posts, and proposals, our team can help you set up automations that generate social media posts, email responses, reports, or other important communications that keep your authentic brand voice.  If you’re interested in learning more, start with our plain‑English overview of AI workflow automation, then browse a menu of buildable AI and automation solutions. When you are ready, contact us to talk through your bottlenecks, or join our mailing list to keep learning practical ways to implement AI. If this was useful, please share it with a friend and follow us on Twitter and LinkedIn – or subscribe to the Ravensight: AI for Small Business podcast on your favorite podcasting platform.

Sources

• Coman, A., & Cardon, P. (2025). “Professionalism and Trustworthiness in AI-Assisted Workplace Writing.” International Journal of Business Communication. https://journals.sagepub.com/doi/10.1177/23294884251350599
• UF News (2025). “Is writing with AI at work undermining your credibility?” https://news.ufl.edu/2025/08/writing-ai-work/
• Dartmouth College (2025). “First Therapy Chatbot Trial Yields Mental Health Benefits.” Dartmouth News. https://home.dartmouth.edu/news/2025/03/first-therapy-chatbot-trial-yields-mental-health-benefits
• ScienceDaily (2025). “Why AI emails can quietly destroy trust at work.” https://www.sciencedaily.com/releases/2025/08/250811104226.htm
• Euronews (2025). “Using AI to write work emails? Your employees may think you’re less trustworthy.” https://www.euronews.com/next/2025/08/12/using-ai-to-write-work-emails-your-employees-may-think-youre-less-trustworthy-study-finds
• PsyPost (2025). “Managers who use AI to write emails seen as less sincere, caring, and confident.” https://www.psypost.org/managers-who-use-ai-to-write-emails-seen-as-less-sincere-caring-and-confident/
• CORDIS Europa (2025). “Should AI write your work emails?” https://cordis.europa.eu/article/id/461081-should-ai-write-your-work-emails
• Rudnicka, K. (2025). “Each AI Chatbot Has Its Own Distinctive Writing Style — ChatGPT and Gemini write in different idioms.” Scientific American. https://www.scientificamerican.com/article/chatgpt-and-gemini-ai-have-uniquely-different-writing-styles/
• Efficiency AI (2025). “Rising AI Mistrust Leads to Frustration with Customer Service.” https://www.efficiencyai.co.uk/rising-ai-mistrust-leads-to-frustration-with-customer-service/
• Kobak, D., et al. (2024). “Delving into ChatGPT usage in academic writing through excess vocabulary.” arXiv 2406.07016. https://arxiv.org/abs/2406.07016
• Nielsen Norman Group (2020). “How People Read Online.” https://www.nngroup.com/articles/how-people-read-online/
• Microsoft Work Trend Index (2024). “AI at Work Is Here. Now Comes the Hard Part.” https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part
• Microsoft WorkLab (2024). “What Copilot’s Earliest Users Teach Us About AI at Work.” https://www.microsoft.com/en-us/worklab/work-trend-index/copilots-earliest-users-teach-us-about-generative-ai-at-work
• Microsoft WorkLab (2024). “AI Data Drop: 3 Insights from Real-World Research on AI Usage.” https://www.microsoft.com/en-us/worklab/ai-data-drop-3-key-insights-from-real-world-research-on-ai-usage
• McKinsey Global Institute (2012). “The Social Economy: Unlocking value and productivity through social technologies.” https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy
• Adobe (2019). “Email Usage Study.” https://blog.adobe.com/en/publish/2019/09/08/if-you-think-email-is-dead-think-again
• Gallup (2025). “AI Use at Work Has Nearly Doubled in Two Years.” https://www.gallup.com/workplace/691643/work-nearly-doubled-two-years.aspx
• Statista (2024). “Time spent on writing tasks for U.S. employees 2023.” https://www.statista.com/statistics/1493349/time-spent-writing-tasks-us-employees/
• EmailToolTester (2024). “Email Usage Statistics.” https://www.emailtooltester.com/en/blog/email-usage-statistics/
• AIPRM (2024). “AI in the Workplace Statistics 2024.” https://www.aiprm.com/ai-in-workplace-statistics/
• Azumo (2025). “AI in the Workplace Statistics 2025.” https://azumo.com/artificial-intelligence/ai-insights/ai-in-workplace-statistics
• New York Post (2023). “Here’s how many hours a week employees spend drafting emails.” https://nypost.com/2023/10/13/heres-how-many-hours-a-week-employees-spend-drafting-emails/
• Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). “Delivering CBT via conversational agent.” JMIR Mental Health. https://mental.jmir.org/2017/2/e19/
• Chang, C. L., et al. (2024). “Wysa for Health Care Workers: Feasibility Study.” JMIR. https://pmc.ncbi.nlm.nih.gov/articles/PMC11034576/
• Wang, Y., et al. (2025). “CBT-based AI chatbot for students: randomized trial.” JMIR mHealth and uHealth. https://mhealth.jmir.org/2025/1/e63806
• High5Test (2025). “60+ Communication in the Workplace Statistics (2024/2025).” https://high5test.com/communication-in-the-workplace-statistics/