A SURVIVAL GUIDE to Mental-Health Posts made with AI.
AI is producing the content—not the influencers who publish it without actually understanding it.
Note : this text was traduced from French by AI.
I am going to demonstrate this using quotes, and explain what is dangerous about this chain of recoded, unverified information: it eventually settles in your mind as belief, evidence, or an adopted logic that you then repeat to coworkers over lunch because you think you are helping.
Do you want the conclusion before the full argument?
Here it is.
To decide who you are allowing to put beliefs into your mind about your mental health, your difficulties, and the mechanisms supposedly driving them, you will have to determine who genuinely understands the subject and who actually owns the claims being made.
To assess the real level of expertise behind a visible personality or practitioner, there may soon be only one reliable question:
What remains of their expertise when they are asked to defend, one by one, the links in their reasoning?
Why are they saying this? Why are they making that connection? Under what conditions does it hold? What is the conceptual architecture behind it?
And here, the only obstacle may be you—and the image you want to project.
You do not want to be the annoying person asking difficult questions. You only want to leave a small, visible comment. But by doing so, you reproduce and reinforce a fundamental problem that has already caused years of misdirection for people in distress: someone lands on the post, sees that many people agree with it, and adopts the same explanation without examining it.
Take the concept of burnout. Beginning in 1973 with the first scientific publications, the humans who were supposed to verify what was being thought, claimed, and published completely botched the job—see my FAKING BURN-OUT dossier.
Now imagine handing sensitive subjects to a machine designed to churn out text at scale, a machine that reproduces the mistakes it has encountered everywhere—including mistakes about burnout—without questioning them either.
The result is a chain in which none of the links performs the necessary scrutiny: from the AI to you, the reader, who assumes that the content possesses all the attributes of truth and self-evidence.
When you care about your physical health, you check the quality of what you eat.
When you care about your mental health, you should also check the quality, meaning, and logic of what your mind is absorbing without examination.
I use AI too—to brainstorm certain concepts, identify what may still be missing for the reader, and make an argument clearer and more intelligible. That part is useful.
But when it comes to conceptual articulation, 90 percent of the time it talks nonsense and fails to account for the relationships required to locate the issue properly.
Now, for patients, curious readers, and anyone who genuinely wants to go deeper, let’s begin.
AI Does Not Only Fabricate False Information. It Fabricates Reasoning That Already Looks True.
The most troubling problem created by AI may not be the obvious errors it produces.
Those usually become visible eventually.
The more difficult problem appears when AI assembles phrases and chains of reasoning that we are already accustomed to treating as correct, true, logical, or self-evident—even when their seams are obvious and the claims themselves are debatable or already disputed.
The more expected a formulation is within a field, the less likely it is to be questioned. The more it resembles something we have already read a hundred times, the more easily it can be asserted without a precise definition, a demonstrated logical connection, or any examination of its consequences.
All that remains is to connect several such statements to create the appearance of an explanation.
And AI does this remarkably well.
It retrieves the concepts already circulating, brings them together, generates a few causal transitions, and gives the whole thing the fluidity of a coherent argument.
Each individual piece seems plausible.
The chain itself has never been demonstrated.
When Vocabulary Produces the Explanation by Itself
Consider a formulation that has become extremely common:
“Just because you are on vacation does not mean your nervous system understands that you need to rest.”
This is usually followed by references to an “internal state of alert” that will not deactivate, a mind remaining in “surveillance mode,” and the conclusion that “this is not a matter of willpower.”
The sentence works immediately because it connects a familiar experience—being unable to switch off—to physiological vocabulary that gives the explanation apparent depth.
But what exactly does it mean for a nervous system to “understand” something?
How is this supposed state of alert identified in the person concerned? What allows us to distinguish a specific physiological mechanism from ordinary anticipation, a habit, a situated concern, or a learned relationship to activity? How do we move from a few work-related thoughts during a vacation to the claim that an internal system is refusing to rest?
The sentence provides no answer.
And it barely needs to, because its components already belong to the contemporary repertoire of stress, trauma, emotional regulation, letting go, and related topics.
The vocabulary is familiar, so the causal relationship feels familiar too.
Yet the reader has not received a demonstration. The reader has received a physiological metaphor attached to a presumed state of alert—one that has become familiar enough to function as an explanation.
Burnout and “Systemic” Causality
The same mechanism becomes particularly visible around burnout.
Today, all it takes is to line up a few expected propositions:
The problem does not come from individual fragility; it is systemic.
The organization of work produces suffering.
Work overload, lack of autonomy, value conflicts, and loss of meaning are its causes.
We must therefore measure risk factors, listen to workers, and strengthen primary prevention.
You have read this collection of claims—and countless variations of it—three thousand times.
The whole thing appears coherent. It even possesses a powerful moral progression: we stop blaming the individual, identify the organization, and propose collective action.
Yet almost none of the connections has been examined.
What is burnout if its own science invalidates the construct? How is it distinguished from depression, stress, general exhaustion, or other forms of psychological distress? Are the elements being cited causes, associated factors, classification categories, or broad descriptions of difficult working conditions? Are they specific to the phenomenon being discussed? What exactly does the instrument measure? And what evidence establishes that reducing these indicators actually reduces the suffering supposedly being prevented?
I have documented elsewhere, piece by piece, both the scientific instability of the construct and its invalidation by the very field that created it, from its origins onward—see my open-access dossier.
The word “systemic” answers none of these questions.
It indicates a presumed level of explanation. It does not demonstrate the causal chain that would allow us to move from working conditions to a specific psychological entity, and then from a measurement to an effective intervention.
Saying that a problem is systemic creates the feeling that one has moved beyond a naively individual explanation. It creates the feeling of having understood, of possessing knowledge. It also enhances the importance of the person making the claim—with the assistance of their AI.
But relocating the presumed cause does not establish that cause.
A collective explanation can be just as poorly demonstrated as an individual one. This is also true of the science surrounding burnout.
When Familiar Concepts Become an Imaginary Mechanism
The same process appears in content about the repetition of workplace difficulties.
The standard scenario is simple: someone changes companies, managers, or careers, then finds themselves in a similar situation a few months later.
The explanation arrives immediately:
“What does not change is your relationship to work.”
Then come the supposedly deep needs for recognition, belonging, security, and identity. Saying “yes” too often is rewarded. Availability is valued. The brain supposedly records a strategy and reproduces it everywhere—with every manager and in every company.
A name is then placed on the whole sequence—let’s call it “primary work attachment.”
The label completes the job.
It retrospectively creates the impression that an object has been identified, theorized, and explained, when all that has happened is that several familiar propositions have been placed in sequence because they seem as though they belong together.
The pun is intentional. There had to be at least one tiny bit of fun.
In a few lines, a complete causal chain has been pseudo-constructed:
deep need → behavior → reward → brain learning → generalization → repetition across contexts.
Every link is recognizable.
We know that behaviors can be reinforced. We know that recognition matters. We know that certain behaviors repeat. We know that a person’s history does not disappear when they change companies.
But these general truths are not enough to construct the specific mechanism being asserted.
Why would a particular need produce precisely that strategy? How do we know that the behavior is unconscious? What allows us to claim that the brain has recorded it as a global response? Why would it be reproduced in every company? Which conditions reinforce the repetition, which modify it, and which interrupt it?
None of this has been established.
The pieces are brought together. The label does the rest.
The standard excuse is that everything cannot be explained in a single post or article.
Fine.
Except it is never explained anywhere else either.
From a Valid Critique to an Automatically Effective Solution
The same leap appears in discussions about work organization.
For example:
“Employees are told to regulate themselves better. Nobody asks the organization to reduce the burden placed on them.”
The line is excellent.
It powerfully shows how a collective constraint can be transformed into an individual responsibility. It makes it possible to criticize cosmetic responses, personal stress management, and programs designed to help workers tolerate unchanged conditions.
You already know the mechanism if you follow these subjects on social media.
But the validity of the critique does not automatically validate the solution that follows—even when the solution draws on concepts and vocabulary borrowed from schools of thought or academic disciplines that themselves deserve scrutiny.
The next claim is that we simply need to examine real work, listen to teams, organize spaces for discussion, and ensure that decisions follow the conversations.
Then comes the conclusion:
“These are the real solutions.”
What demonstrates that?
Under what conditions do these spaces actually change the work? Who defines which problems may be discussed? What power is attached to the statements being collected? Which targets, deadlines, or decisions can genuinely be challenged? What happens when the demands expressed conflict with profitability, staffing levels, or management decisions?
A workplace discussion space can transform an organization.
It can also collect complaints, reframe them, produce indicators, recommend a few peripheral adjustments, and leave the decisive constraints untouched.
The passage from expression to transformation cannot simply be asserted.
Yet the solution appears obvious because it mobilizes concepts that are already valued: real work, the collective, listening, dialogue, participation, and primary prevention.
The borrowed respectability and moral desirability of the intervention stand in for evidence of effectiveness.
Whenever a substantive question is asked, an evasive answer often appears:
“That is what I encounter in the field.”
“That is what I observe every day with my clients.”
“It works in my practice.”
They answer questions of principle with individual cases, even though reproducibility is precisely what underpins the very idea of prevention.
That is already a first indicator that they do not master the subjects on which they present themselves as expert-advocates.
Field experience can generate a hypothesis, illustrate a mechanism, or signal a possible pattern.
On its own, it does not demonstrate that the mechanism is causal, generalizable, and reproducible, or that the proposed intervention actually prevents what it claims to prevent.
Answering a general question with a few individual cases therefore means quietly changing the required level of evidence.
Prevention requires us to establish, beyond the situations selected by a practitioner, that an exposure regularly produces certain effects and that an intervention can reduce them.
This systematic retreat into “my field experience” is therefore already revealing: the person may master the professional narrative surrounding the subject on which they claim expertise or advocacy, while understanding far less about the conditions required to establish what they are asserting.
Familiarity Replaces Demonstration
This is where AI introduces a major change.
It does not need to invent an entirely false theory.
It draws from an enormous stock of already-legitimized formulations:
nervous system, internal alert, deep needs, recognition, relationship to work, toxic management, loss of meaning, risk factors, systemic approach, real work, employee listening, primary prevention, sustainable performance.
It then combines them around virtually any situation.
Difficulty resting becomes physiological activation.
A repeated workplace pattern becomes an unconscious strategy recorded by the brain.
Distress becomes the effect of a system.
A listening exercise becomes prevention.
A desirable intervention becomes an effective solution.
And a heat wave can apparently cause “thermal burnout.”
None of the sentences appears absurd, and that is precisely what makes the whole thing difficult to challenge.
The reader recognizes the concepts before examining the relationships between them, and that recognition creates an impression of depth.
The more expected the phrases are, the less they are questioned.
The Loop Feeds Itself
This stock of formulations is not stable. It becomes increasingly concentrated.
AI learns from what circulates. Publishers post its outputs. Those outputs join the body of material on which the next generation learns.
With every cycle, the frequency of expected formulations increases—and therefore their familiarity, their apparent legitimacy, and the likelihood that they will go unquestioned.
What is published because it is expected becomes even more expected because it has been published.
A causal relationship that was never demonstrated can therefore acquire greater apparent obviousness every time it is repeated. Its credibility is increasingly measured by circulation rather than examination.
The system does not merely produce superficial content.
It makes superficiality cumulative.
And the human chain extends the machine-generated loop.
The AI produces.
The person publishing the content, who does not understand its architecture, posts it.
Another person reposts it.
The reader reads it.
At every link, confidence increases—it has been published, repeated, and seen everywhere—without verification occurring at any point.
Confidence accumulates along the chain while actual mastery remains at the exact level where the AI left it:
nowhere.
A Post Can Simplify. An Entire Body of Work Should Go Deeper.
A short post obviously cannot demonstrate everything.
It simplifies, condenses, and selects. It would be absurd to demand that every social-media publication achieve the precision of a scientific article.
The problem begins when a simplified post is followed by dozens of other publications that supposedly deepen the argument, while each one merely rearranges the same stock of ready-made causalities.
One day, the nervous system explains difficulty resting.
The next, recognition explains overinvestment.
Then the brain explains repetition.
The organization explains suffering.
Listening explains prevention.
The continuity of the publications creates the appearance of a body of thought being progressively developed.
Closer examination often reveals something else: the same concepts circulate, change order, and are used to dress up new situations without their definitions, relationships, or limitations ever being seriously developed.
Quantity creates the illusion of depth.
A long string of explanations is not a theory.
A theory must be able to specify what produces what, under which conditions, through which mechanisms, with which limitations, and against which counterexamples.
Without that, each new publication merely confirms the previous one through repetition.
Expert Identity Becomes Something You Can Put On
AI does not only supply sentences.
It supplies the social signs of expertise: vocabulary, tone, expected distinctions, possible references, causes recognized by the professional environment, and solutions regarded as reasonable.
It creates an immediately available kit.
The apparent expert knows what to say—which concepts to use, which causes to denounce, which solutions to praise, and which oppositions to produce:
individual versus systemic, symptom versus cause, reaction versus prevention.
The discourse becomes acceptable because it matches exactly what the surrounding environment expects an expert on the subject to say.
But this identity, built on visible expertise, eventually reaches a limit.
Sometimes, all it takes is asking why one proposition leads to another.
What exactly are you measuring?
Why are you claiming causality?
How do you distinguish the objects involved?
What demonstrates the effect of the proposed intervention?
When the architecture does not exist, the response changes register.
Evidence becomes conviction.
An assertion becomes field experience.
Effectiveness becomes a constructive intention.
The request for precision becomes jargon or unnecessary complication.
This shift is logical.
Once a public expert position has been adopted, acknowledging a gap becomes costly. The person is no longer protecting an idea. They are protecting the status that the idea allowed them to occupy.
AI can therefore supply the initial appearance of mastery, and then help defend it.
Because it does not understand the subject either, when the AI runs out of exits and must concede the argument, only one escape remains: retreat into morality or personal experience.
Why This Becomes Dangerous in Mental Health
This superficiality would already be problematic in any field. It already exists in medicine, for example among people who claim that homeopathy can cure cancer.
It becomes especially dangerous in medicine, health, and mental health because the reader is not merely receiving content.
The reader may use it to interpret an experience, identify a cause, name their distress, select a form of support, accept a method, or assign authority to the person speaking.
A statement about the nervous system can become a personal explanation.
A list of symptoms can become a self-identification.
An organizational causal claim can become a certainty.
A category such as burnout, trauma, hypersensitivity, neurodivergence, or a toxic relationship can suddenly give complete coherence to a complex experience—even when that category has been profoundly distorted through repeated long-term exposure to the same framing.
That coherence often produces relief.
But the relief produced by an explanation does not demonstrate its validity.
And we must be precise about the nature of the danger: it does not arise only from claims that are false.
Some of these explanations may be valid.
The problem is that it becomes increasingly difficult to distinguish well-supported explanations from merely assembled ones, because both have exactly the same surface: the same vocabulary, the same fluidity, and the same confidence.
Partially true propositions, assembled in a plausible order, reinforced by recognized vocabulary, and presented with a confidence their author might never have been able to produce alone—this is what circulates.
And in these fields, it has real consequences for decisions, identities, and care pathways.
A surgeon must master both the field and the procedure.
The same should be true of anyone claiming expertise in mental health.
We Will Have to Learn to Examine the Links
Trying to detect whether a text was written using AI will solve nothing.
A competent person can use AI to clarify, popularize, simplify, or smooth out a genuine line of thought. A person with no real expertise can write alone using other sources. No stylistic analysis will reliably distinguish between the two every time.
The real criterion lies elsewhere.
We must stop evaluating expertise through the fluidity of a text, the presence of concepts, or the overall coherence of a narrative.
We must examine the links.
And in mental health specifically:
How do we move from an observation to a cause?
From a presumed cause to a measurement?
From a measurement to an intervention?
From an intervention to the announced effect?
A reader alone in front of a piece of content can already perform three checks.
First, reverse the hinge statement. If the opposite claim would sound equally plausible, the sentence explains nothing.
Second, ask what the explanation rules out. An explanation compatible with every possible case explains no particular case.
Third, ask for the path. At exactly which point does the argument move from observation to cause, and what authorizes that move?
These three checks require no expertise.
They require only a refusal to let familiarity stand in for demonstration.
For readers interested in the question of expertise, my open-access academic article develops the issue further:
A Typology of Contemporary Expertise in Context: Producers, Accumulators, and the Self-Appointed
In the age of AI, the question will soon no longer be:
Can this person produce expert-sounding discourse?
It will become:
What remains of their expertise when they are asked to defend, one by one, the links in their reasoning?
