Why Does Your Content Sound Like a Bot?
- Helena Goos
- Apr 4
- 8 min read
Updated: Apr 5
AI is everywhere. It’s in our feeds. It’s on our computers. It’s in our hands. Throw a virtual rock, hit an AI. Of course, for the copywriter and content creator, this problem is twofold---creating content that avoids sounding like AI or, if using AI, avoiding AI detection.
Recent studies have demonstrated that humans have a very poor ability to distinguish between human and AI-generated content. Given the amount of content flooding our socials, it’s no wonder that the reaction is to assume that everything you come across is somehow AI-generated.
If you’ve ever noticed or asked yourself, “Why does my writing sound like AI?” you’re not alone.
An Aside Into Semantics: AI vs LLM
As the industry and technology has emerged, the general public has been using the term AI to cover a wide array of models from LLM’s to video diffusion models. However, I want to be clear that in this article, when referring to AI, I am specifically referring to Large Language Models. A Large Language Model is a highly advanced computational model that uses incredible amounts of data and power in order to predictively generate responses to inputs. These models appear in the form of chatbots, such as ChatGPT or Claude. In this article, the chatbot I will be basing most of my advice and analysis on is Claude, as it is the main chatbot that I use.
How AI Talks
It’s easy to forget that AI is not a person, especially when it responds so convincingly. AI doesn’t “speak” in the sense that a human does (there’s a philosophical debate here somewhere). Grammar comes as a byproduct of its generation process, not as the goal.
Claude doesn’t choose to use a subordinate clause in the sentence: “Claude doesn’t construct grammar, it predicts the likeliest structure based on its training data.” Claude is using the data it has to predict the likeliest response based on the probability of a grammatical structure occurring in a given linguistic context.
This is, in some ways, a similar process to what humans do when we speak---the key difference being that we can override our predictive ability and put something out of order on purpose.
Unedited AI responses will often have a few dead giveaways of their origins---and it’s more than just the em dash.
In the spirit of using AI, I went ahead and asked Claude to explain to me what its most significant response features are. This breakdown helped me put into words what I had been observing in the months since I had been using it.
Right-Branching Sentences
A right-branching sentence is a sentence where the main subject appears at the beginning before additional, contextual information about the subject. A left-branching sentence occurs when the subject comes in the latter half of the sentence OR is not the first information provided.
Right-Branching: “The horse named Friday rode into the town on a Saturday.”
Left-Branching: “On a Saturday, the horse named Friday rode into the town.”
Right-Branching Sentence (AI Typical)
Subject | Subordinate Clause |
[The horse named Friday]... | …rode into the town on a Saturday. |
Left-Branching Sentence
Subject | Subordinate Clause |
[the horse named Friday] | On a Saturday [ ] rode into the town. |
According to Claude, right-branching sentences are a marker of AI, although it is not uncommon to see this used in human copy. It is worth noting that human writing often utilizes both right and left branching sentences interchangeably, but with more creativity---especially to change an emphasis of the subject
Nominal Style
You may notice that AI, especially Claude, tends to use adverbs such as "genuinely" and “honestly”. However, Claude seemed to think that the nominal style was a more common feature of AI writing. Nominalization is the use of a word, typically a verb, as a noun.
Example sentence: “They were explaining to us how they chose to come to this conclusion.”
Nominalization: “They gave us an explanation on the choice to come to this conclusion.”
Verb | Noun Form (Nominalization) |
choosing, to choose | the choice |
explaining, to explain | the explanation |
Hedging Structures
Hedging is a style in which the speaker (for a variety of reasons) attempts to introduce ambiguity into the argument. This style can be commonly found in pieces like medical disclaimers or damage disclaimers.
The hedging structures baked into AI writing are there for two primary reasons: safety (to avoid the spreading of misinformation) and RFHL (Reinforced Human Learning). You may notice that this entire section has been in a plausible hedging structure.
Common Hedging Devices
Verbs | Adverbs |
suggest, imply, indicate, may, could, might, estimate | possibly, perhaps, plausibly, apparently, probably |
Adjectives | Modal Nouns |
probable, likely, certain | possibility, probability, estimate, estimation |
The modal form may indicate permission, possibility, ability, necessity---in this context, we are focusing on the possibility aspect of this form.
You may also note that once again, nominalization makes an appearance with the modal nouns in the chart above.
Neat Little Bow
AI writing tends to resolve all of its arguments, unlike human writing, which can tend towards hanging threads or unresolved arguments. With Claude, this feature is a result of its RFHL training, and the human trainers (myself included) score completed, resolved responses higher than incomplete responses.
Human writing, like all art (even if it’s corporate swill) is an act of creation. This is what separates human writing from AI writing. This is why it can be hard to make AI writing sound human. AI content is generated, not created. It follows a predicted rhythm and beat---humans are freestyle jazz.
Instead of resolving every statement in your LinkedIn post or Twitter article, try asking more questions. Maybe that’s the whole tweet.
So…What Do I Do Now?
It would be easy to walk away from this conversation with Claude feeling confident in your ability to detect AI writing. But here’s the problem. You don’t sound like AI. AI sounds like you. AI is trained on human-created content and is known to scrape the internet for its data (warning: NSFW topics), including social media. Human linguistic ability is so fluid and creative that even if someone has never touched AI a day in their life, they may still sound like AI. The reality is, there isn’t a true way to avoid sounding like AI, but there are some steps you can take to fix AI writing or your own, and/or avoid AI detection or setting off the alarm bells.
Let your writing go unedited.
Yeah, I know, I know. That’s some pretty stinky advice. Editing is so fundamental to the job of writing that it can make or break a story. That goes for everything, from television to poetry collections. But in the day of instantaneous tweets and long-form, unedited or “raw” footage might be one of the best ways to “prove” your content authenticity: odd punctuation choices---the em dash keeps people guessing---unfinished thoughts and arguments. When AI makes everything look perfect, embracing the imperfection of human creation is how you prove your humanity.
Write like you talk.
You don’t sound like AI. AI sounds like you. Write like you, the person, talk. Keep your casual slang and your contractions. In some long-form formats, this may not be stylistically possible (due to style guidelines). But in personal pieces like those that go on your blog or Twitter articles, your idiolect will let you shine.
Make a tone of voice document.
Tools are for everyone, and marketing tools are no different. A tone of voice document/style guide/etc. is a tool that marketing teams use for brand products. But you can make one for yourself.
To start, go through your own socials or writing and take account of them: topics, tones of voice---make note of how you want to sound.
Take these examples, as well as your target voice, and create a tone of voice document by using characters and archetypes to guide your style. You can feed this to your LLM of choice and/or use it as a reference to tone check what it generates.
Take a look at Jane Doe - Tone of Voice for a better idea!
Use AI’s system prompting.
Most models like Claude and ChatGPT have the ability to personalize their responses. This is called “system prompting” and it’s incredibly easy to use, if you haven’t already.
To personalize Claude's responses:
Go to the lower left-hand corner of your Claude dashboard
Toggle your profile → select Settings
Scroll past your name and personal info
Find the prompt: "What personal preferences should Claude consider in responses?"
Click the dialogue box and type anything — direct instructions (avoid politics) or personality notes (always talk like Yoda from Star Wars)
Here's what that looks like in practice:
Go to the lower left-hand corner of your Claude dash, and toggle your profile: then select “Settings.”
Underneath your name and other personal information, you will see the prompt: “What personal preferences should Claude consider in responses?” with a dialogue box attached.
Select the dialogue box and input anything you want.
Remember that AI is a very useful tool, but the tool is only as good as the master. You can use AI to push your work, reasoning, and art, all with decent prompting.
Examples of our personal system prompts:
Helena: “Be objective, though not unkind. Do not sweet-talk me. Challenge me. Ask me questions that force me to think in three dimensions. Ask me questions that force me to consider alternate perspectives or even contradicting perspectives. When appropriate, play Devil's Advocate. Do not allow me to be complacent in my own thinking or reasoning.”
Laurisha: “ Have a conversational tone and keep my words as much as possible. I value humanity above all else. Content should evoke a feeling. If you need more context, ask me questions.”
Hire a human for long-form.
Even after you do all that---backend parameters, tone of voice documents---you may still find that the result produced by AI is not quite right. You will probably have to go through and tweak everything from words to whole paragraphs. That’s when it might be time to hire a human. For long-form and most especially for pieces that need to tug the emotional heartstrings of the readers, you need a human touch. AI copywriting isn’t enough to cut it when you are trying to explain your life experience.
Run your work through an AI detector.
AI slop reads as laziness, and your audience feels it. Readers want to feel like you wrote for them specifically, not for everyone and no one at the same time.
There are many tools on the market. CryptoFilma uses Originality.AI, it’s not perfect and works 85% of the time. Even well-written human copy can get flagged. [You can see this blog's draft score here.]
Beyond your audience, some editors won't accept op-ed submissions that fail the test. If earned media is part of your strategy, it's worth knowing where you stand before you pitch.
Know when to use AI.
Is it not Sun Tzu who said “know your enemy?” While AI has its faults, it’s still an incredibly useful tool. Learning when to use it is just as important as a skill as being able to write on your own.
Examples of where to use it may be…
Product update blog: give it an outline with key bullet points on the product, the goal (including SEO), and the target audience. Come back later to add images and screenshots.
Quick social posts for investors or partners to share: keep the prompt conversational, or it'll sound like a press release.
Punching up social posts or formatting for longer threads.
Skip AI when you want to evoke a human emotion. AI isn’t built for connection, but yes, it can help optimize or repurpose.
Want to Hire a Human?
If you ever want help with your content— whether that's ghostwriting long-form, building a tone of voice document, or adding authenticity to AI, give us a holler!
Special thanks to Laurisha Cotton, whose original concept, editorial guidance, and contributions to the actionable steps made this piece possible.
Comments