LinkedIn Feed Collapse. Why Readers Stop Trusting You.

This year, more people will use ChatGPT and LLMs to damage their reputations than to build them. Let me ask you this: When was the last time you spent time on LinkedIn without wondering, “Is this written by a person, or generated by AI?” Before I show you how to make AI sound better than the rest, let me make the case.

I’ve just searched Reddit for posts about LinkedIn and AI. People are raging: “Half the posts are AI-generated now. So there is just no reason to scroll anymore. I check my messages once a week, and that’s it,” says one user

But even LinkedIn-related posts about fake AI-generated posts are fake. In one such thread, another user comments: “This is an entertaining post even if faked, but if people start doing this, it'll make LinkedIn profiles like car bumper stickers, full of random nonsense, much of it offensive.”

It’s everywhere right now.

Emails sound smooth but empty, polished but generic, confident without stakes. Common signs include sentences that explain rather than decide. Abstract nouns stacked on top of each other. Vague claims with no numbers or consequential reasoning. Everything sounds reasonable. Nothing feels like a real person actually lived the experience that drove the post. The tone does not reveal any of the author's stories or experiences that demonstrate their journey that led to it.

You’d read something like: “AI is reshaping how professionals approach growth and efficiency.” The tell: No actor, decision, cost, or effect. Stacked abstractions with no mechanism, like “This creates alignment, clarity, and momentum across teams.” Nothing moves, or changes, and no one acts.

Tell words and phrases are: ‘unlock’, ‘signal’, ‘the shift is’, ‘the lift shows in’, ‘he names a real shift’, ‘scalable’, ‘in today’s fast-paced world’, ‘at scale’, ‘going forward’, ‘it’s important to note’, ‘teams are seeing’, ‘this really resonates’, etc. Too many to list them all here.

But the number one tell? Once you see it, you stop asking who wrote it. You start asking who is responsible.

GPT avoids making decisions because accountability follows them.
It cannot own cost and lose trust as we do. It doesn’t face a client, a court, a board, or a market. It offers options without choosing. We humans decide because we must live with the consequences of our actions. Wherever we are in life is a result of our actions and decisions. AI doesn’t have that. When you see a sentence with no trade-off, no downside, no risk, no owner, it was written by AI.

People are annoyed with it. So, are we doomed?

In one of my recent LinkedIn conversations, one of my contacts wrote: “Sounds like AI, but I like it.” It wasn’t AI, which made me realize that GPT had already brainwashed me into a language that didn't serve me. I found out that I am not the only one.

Hiromu Yakura is a postdoctoral researcher at the Max Planck Institute for Human Development in Berlin. One day, he noticed he was using the word “delve” more. He wondered if others were experiencing the same trend.” So he asked ChatGPT to edit millions of pages of emails, essays, and academic and news articles. The prompts weren’t anything special, such as “polish the text” or “improve clarity.”

They isolated the words that the model repeatedly added during editing. Examples include ‘delve,’ ‘realm,’ and ‘meticulous.’ The researchers labeled these terms ‘GPT words.’ I call them ‘GPTisms.’

Next, they analyzed more than 360,000 YouTube videos and 771,000 podcast episodes. The data spanned periods before and after ChatGPT’s release. The goal was to track changes in GPT word usage over time.

We Even Start To Speak Like ChatGPT

They compared those words to synthetic controls. These controls came from weighted synonyms that ChatGPT used less often. For "delve," the controls included words like "examine" and "explore." The surprising part? GPTisms went beyond scripted formats. They also appeared in spontaneous conversation.

“AI technology seems to send patterns back to the human mind,” says Levin Brinkmann, a co-author from the Max Planck Institute for Human Development. He describes a feedback loop. Humans train AI on text. AI recombines that text. Humans then absorb and repeat its patterns. Yakura adds context. “AI is not unique in influencing behavior,” he says. “The difference is speed and scale.”

In other words, as AI gains cultural authority, meaning people are more likely to mimic its language than that of other sources. Three years into adoption, AI has already altered everyday speech. The question now is, “When it reads and sounds like AI, can it be trusted?” If the answer is “no,” you’ll disappear in the crowd of LinkedIn users who take the easy route rather than taking the time to be genuine and authentic.

These are the people who buy courses like: “Create your entire consulting program in just three hours” and fall for promises like: “AI done for you. Custom AI Team delivered in 1 week,” or “Create your website with AI in just 45 seconds” and “3 months of AI strategy in 3 hours. Free course, 5 steps, 23 tools, 1 proven roadmap.”

Can you do all of that with AI? Yes, but it will cost your reputation.

High Performance Consultants don’t fall for that.

They are looking for results beyond standard norms. They take a shortcut, simplify work, and automate operations. But they do this without shortchanging their clients or risking delivery quality.

You need a key tool: the ‘Voice Card.’ It stands apart from typical style guides, brand voice documents, tone rules, prompts, or system instructions. Those describe preferences but do not enforce behavior.

A Voice Card is an operating constraint with hard, specific requirements. It blocks GPTisms and instead forces cost, mechanism, and decision in your writing. Think of it as a behavioral contract between you and AI. It makes your content feel human. Most AI-generated output seems rehearsed and often fails to inspire readers to act decisively.

You can put it in your system instructions in your custom GPT. You might want different versions. Just like in real life, you talk and write differently based on the situation. Perhaps you want to sound analytical when writing a report, inspiring when writing a LinkedIn post, educational for teaching or explaining concepts, technical for expert discussions or specialized topics, or persuasive in sales or marketing.

You build one voice card as a foundation, from which you quickly derive a version tailored to your specific needs. You can either create one Custom GPT for each purpose, or a single GPT with system instructions on which voice card to use when.

Let me give you an example of what you’ll gain.

I used this prompt with GOT 5.2: “What is the purpose of life in a short three-sentence paragraph?”

Without my voice card attached, I get this output:

“The purpose of life is to experience existence and become conscious of it. Through relationships, challenges, and curiosity, humans learn what matters and how to act in the world. Meaning emerges not as a fixed answer, but as something created through choices, responsibility, and connection over time.”

With my voice card attached, it now reads:

“Life is about choosing what you stand for while time and options are limited.” I decide meaning by the work I commit to, the people I protect, and the consequences I accept. Each day I act, learn, adjust, and choose again with intent.”

Notice how the first version explains life from a distance. It speaks in general terms about humans, experience, and awareness. Meaning develops over time. It comes from various forces, such as relationships and responsibilities.

The second version takes ownership. It places the speaker inside the problem and forces them to make decisions within limits. Meaning comes from commitment, protection, consequence, and repeated action taken on purpose. That’s how I think, work, and communicate. You may or may not like it, even though I assume that you do, otherwise you wouldn’t have read that far. But you will create your Voice Card so it sounds the way you want.

If you want to know how to create one, I’d be happy to share.

In the meantime, start by adding these prompts to your system instructions. These should immediately remove some of the most annoying GPTisms. A full Voice Card has about 30 pages. So try these two instructions, see what difference they make. If you like it, I am here for you.

Here is the difference they make on my end. I prompted GPT without instructions as follows: “What is boutique consulting?" Answer in one paragraph and three sentences.”

Here is the output:

“Boutique consulting is a form of consulting delivered by small, specialized firms that focus on a narrow set of problems or industries. These firms compete through deep expertise, senior-level attention, and customized work rather than scale or standardized methods. Clients choose boutique consultants when they want precise judgment, speed, and solutions tailored to their specific context.”

Notice how the output differs from the instructions that you, too, can use verbatim or edit to match your preferences:

“Boutique consulting is consulting delivered by small firms that focus on a defined problem, industry, or decision type. The work is led by senior practitioners who diagnose issues, make calls, and execute within a narrow scope. Clients hire boutique consultants to get focused judgment, fast decisions, and work shaped to their specific situation.”

The first version describes boutique consulting by comparison and implication. It uses familiar contrasts and broad terms to help readers find their way. This approach guides them without demanding exactness. It explains what boutique consulting feels like in the market.

The second version states what it is and how it operates, removes comparison language, and narrows the scope. It focuses on actions like diagnosing, deciding, and executing. The definition becomes operational rather than positional.

The instructions cut out some GPTisms, like comparative framing. They also swapped vague ideas for clear roles and actions.

Now see what changes when I apply my full Voice Card:

“Boutique consulting solves a specific business problem for a defined client group. Senior practitioners diagnose the issue, choose a course of action, and execute within a tight scope. If this fits your situation, define one decision you need to make this week and decide who owns it.”

This version now compresses that definition into a use case. It sees boutique consulting as a way to decide under constraints. It assigns ownership and pushes for the next step. They shift from explanation to application. This is how I think and operate.

Try it out, make them yours by changing what you want. Then share your results with me here or talk to me in person here.

Voice Card Instructions For Contrast Structure Ban

If a sentence contains a contrast clause introduced by:

• not,
• rather than,
• instead of,
• as opposed to,
• versus,

or any equivalent negating construction, the sentence must be rewritten to remove the contrast entirely.

Rewrite rule:

1. Keep the affirmative assertion.
2. Delete the contrast clause.
3. Do not replace it with another comparison.

Approved pattern:

"The firms that keep winning treat this as an operating decision."

Disallowed pattern:

"The firms that keep winning treat this as an operating decision, not a messaging tweak."

Kill switch:

If any sentence contains a contrast clause after a comma or conjunction, the sentence fails and must be rewritten using assertion only.

Voice Card Instructions For Noun Phrase Ban

Do not use generic noun phrases that rely on intensifiers or mood words instead of mechanisms.

This includes, but is not limited to:

real + abstract noun (real pressure, real strain, real impact, real value)
quiet + abstract noun (quiet win, quiet success, quiet separator)
vague modifier + vague noun (meaningful shift, powerful change, strong signal)

These constructions are forbidden.

DETECTION RULE

Before output, scan every noun phrase.

If a noun phrase meets all three conditions, it is invalid:

The adjective intensifies tone rather than meaning (real, quiet, meaningful, powerful, strong, significant)

The noun is abstract (pressure, strain, win, success, separator, impact, value)
The phrase does not specify: what changed, by how much, under what constraint, compared to what baseline

If all three are true, the phrase must be removed.

KILL SWITCH

If a generic noun phrase survives the scan, the response fails.

The model must:

• delete the sentence
• rewrite it using one of the approved forms
• re-scan the full output
• only then deliver the response

No soft substitutions. No stylistic exceptions. No explanatory notes.