How to Use an AI Coach: Eight Patterns, Prompts and Limits
Recognize eight AI-coaching patterns, write prompts that expose missing information, and follow a fictional MBTI-to-ChatGPT case through an error, correction and two small work trials.
Start with one concrete task: planning a week, preparing a conversation or trying a small work activity. Tell the AI what you want to produce, give it relevant context, and ask for missing information and options. Check its answer before choosing an action.
What You’ll Take Away
Recognize eight ways people use an AI coach, from making a plan to rehearsing a conversation or seeking reassurance.
Compare prompts that invite checking with prompts that invite unsupported certainty.
Follow an MBTI-to-ChatGPT example from a selected result excerpt through an error, correction and practical action plan.
Know when to seek evidence, real-world feedback or qualified support.
You do not need an assessment result to use this guide. Start with your own non-sensitive account or practise with the fictional examples. If you bring a result, share only material you are entitled to view and comfortable disclosing.
The Same Question Can Lead to Very Different AI Conversations
“I’ve been feeling scattered lately” might become a decision tree, a place to describe emotions, a rehearsal for a difficult conversation or a request for reassurance. The words can look similar while the person is asking AI to do very different work.
A warm response can feel like understanding, and a tidy outline can feel like an answer. Neither shows that the model has interpreted your situation correctly. Notice when “this helps me think” starts to become “this must be true.”
AI can help you clarify a question, name a feeling, organize options and rehearse an explanation. The practical distinction is between using it to prepare your thinking and giving it authority to decide what is true, what matters or what you should do.
In this article, “AI coach” describes that use of a conversational tool. It does not imply a professional qualification or transfer responsibility for a decision.
The FermatMind Framework: Four Positions Where Use Can Go Wrong
These four positions are an editorial framework for examining a conversation. They are not validated personality categories, dependence scores or clinical assessments.
Position
What it can help with
What happens when it crosses the line
Draft board
Put feelings, thoughts and questions into words.
You keep improving the draft while delaying a useful reality check.
Validator — a checking aid
List comparison dimensions, possible blind spots and gaps in a plan.
Its output starts to count as proof, or it chooses your values and weights for you.
Mirror
Help you describe an emotion or notice a possible repeated pattern.
Being responded to begins to feel like being fully understood.
Avoider
Make it easier to pause and organize a response before acting.
Conversation replaces wanted human feedback, necessary action or engagement with consequences.
For decisions, keep AI’s authority at the level of drafting and checking. A reflective conversation may still be useful, but emotional relief does not verify an interpretation. Ask what evidence or action will follow the conversation.
Eight Common Use Patterns: Which One Sounds Familiar?
These patterns are examples of situations, not research-validated MBTI risk groups. Anyone can use several of them. Choose a row by what you actually did recently, not by your type or score.
Pattern
Common AI use
Potential benefit
Main risk to check
Planning and prioritizing
Decision trees, priorities and step-by-step plans
Turns a confusing problem into possible actions
Mistakes structure for correctness
Exploring possibilities
Brainstorming, counterexamples and more options
Opens up alternatives
Keeps expanding without choosing anything to test
Clarifying values
Values, relationship questions and long-term meaning
Helps put important concerns into words
Mistakes a warm response for sound interpretation
Putting feelings into words
Describing feelings, rehearsing talks and writing a personal account
Makes expression easier
Stops at reassurance before clarifying what is needed
Building a process
Steps, checklists, templates and procedures
Makes a task easier to organize
Treats a standard process as suitable for every exception
Solving an immediate problem
Quick fixes and responses to surprises
Produces a practical first move
Mistakes short-term usefulness for long-term suitability
Preparing a difficult conversation
Expressing needs and boundaries
Makes wording clearer
Remains in preparation instead of taking a safe, wanted next step
Seeking reassurance or companionship
Asking for understanding, confirmation or company
May provide short-term comfort
Lets low-friction conversation replace other wanted support or action
Recognizing a row is not a diagnosis. The useful question is:what would reveal that this way of using AI is failing at the task I brought to it?
Before You Prompt: Protect Your Inputs and Set a Stop
Use the minimum information the task needs. Remove names, contact details, private identifiers and organizational material that you are not authorized to disclose. Do not paste another person’s answers, records or private messages into a coaching conversation. Where possible, describe a fictional or de-identified situation instead.
Check the tool’s current data settings and any relevant organizational policy before sending. Chat history, memory and model-improvement settings are different matters, and available controls may depend on the account or workspace. Review the current ChatGPT Data Controls and Temporary Chat guidance rather than assuming that a particular mode makes every disclosure appropriate.
Decide what you want help with, what information you can share, what an error could cost, how you will check the answer and when you will stop. For example:
Help outline a public presentation using these non-confidential points. Mark claims that need a source. I will verify them before sharing the presentation.
Stop if a response is distressing, presses you toward an unsafe action, requests unnecessary sensitive information or becomes your sole basis for a consequential decision. You can pause or seek support without confronting someone who threatens you.
Better Prompts and Riskier Prompts
Pattern
Better prompt direction
Riskier prompt direction
Planning and prioritizing
“List what information I am still missing, then give me three options I can verify.”
“Tell me the optimal answer directly.”
Exploring possibilities
“Narrow these ideas to three options I can test.”
“Give me twenty more possibilities.”
Clarifying values
“Help me separate value conflict, practical constraints and emotional reaction.”
“Do you think this is more true to my authentic self?”
Putting feelings into words
“Help me describe this feeling in words I could say to a real person.”
“Do you think I have already done enough?”
Building a process
“Make a process, then mark the points that require human judgment.”
“Give me a standard answer template.”
Solving an immediate problem
“Separate short-term damage control from long-term repair.”
“Give me the fastest way to get it done.”
Preparing a difficult conversation
“Help me express my actual boundary more clearly.”
“Write something nobody could be upset by.”
Seeking reassurance or companionship
“Name the emotion, but do not draw the conclusion for me.”
“You think they actually love me, right?”
The better prompts make room for missing information, verification and a choice you still have to make. The riskier versions invite certainty or reassurance without enough support.
Prompt wording can help bound a task; it cannot guarantee a safe or accurate response. OpenAI’s prompting guidance and introduction to goals, context and output provide a useful basis for stating what you need and refining it over successive turns.
Watch Where Agreement Feels Most Convincing
Notice what makes a response convincing to you. Reassurance can end a question too soon; a neat outline can hide weak evidence; more possibilities can postpone a choice; and repeated rehearsal can delay a safe next step. These are habits to examine in your own conversations, not predictions from a personality label.
In preregistered experiments on interpersonal advice, Cheng and colleagues (2026) found that sycophantic AI could leave participants more convinced they were right and less willing to repair a conflict, even as they trusted and preferred its responses. These findings concern the interactions studied; they do not identify a vulnerable MBTI type or show that the prompts in this guide prevent that effect.
Try asking:
Identify the weakest assumption in my account. Give another explanation for the same observations, and tell me what evidence would change your interpretation.
Check that answer too. Disagreement is not automatically better supported than agreement. Open citations and see whether the source supports the nearby claim; a plausible title or citation marker is not enough. OpenAI’s guidance on mistakes and factual accuracy explains why confident answers still need checking.
Bring a Result to AI: Select, Preview and Check
Start with a result you are allowed to view. Read its explanation of the questionnaire, units and quality information before selecting an excerpt for one task.
Keep the reading in context
Record the provider, form and question count, result-material language, actual answer language, date, available versions and original units. Leave missing information as “not provided” or Unknown. A translated screen or a file name does not establish the language in which the questions were answered.
Different assessments answer different questions. MBTI-style preferences are not abilities. Big Five dimensions are not personality types. Enneagram candidates need reflection on motives. RIASEC interests do not establish job competence. Reasoning-test performance and EQ self-report are also different kinds of information.
Discuss a percentile only when the report explicitly provides the relevant reference. Do not rename answer accuracy or a questionnaire-scale score as a population rank. Keep any actual quality warning or flag that affects interpretation, even when you select only a small excerpt.
Use the output controls actually available on your result page
On a result page you are allowed to view, open the export area and choose from the purposes and data options it offers. Look for controls such as:
Goal and experience (your own words, optional)
Full export preview
Copy prompt and selected results
TXT
Markdown
Use the visible controls if the wording differs. Do not assume that every result supports an individual checkbox for every field. If the page exports a group, you can select a smaller excerpt locally and label what you omitted.
Select or omit assessment fields using the controls provided. Add your goals, experiences and preferred response style in the separate editable text area. If you reword an assessment statement yourself, label it as your interpretation so it cannot be mistaken for the original result.
Inspect the complete selected preview
“Full export preview” means the full set of material you selected, not the complete private report.
Check for names, contact details, report identifiers, private URLs, recovery credentials, questions, answers and information you did not intend to disclose. Also check that the units, relevant warnings and missing-information labels are still present.
After copying, check the pasted text before you send it. If copying fails, select the full preview manually or use TXT. Open downloaded files to confirm that the expected text is there; a download or print request alone does not confirm that a file was saved.
The MBTI TXT and Markdown options described here contain a summary of selected results, not a complete report PDF. Use a PDF option only if your result page offers it, and open the file to check what it contains. If the PDF is unavailable, use TXT, Markdown or a manual excerpt from material you are entitled to view. Recovery and sharing have their own access conditions; you do not need to make a private report link public.
Open ChatGPT yourself and decide whether to send
Copying or downloading does not send the result to an external model. Open the tool you choose, paste or attach the selected material, review the scope and decide whether to send it.
This is a manual transfer, not a native connection to ChatGPT, Claude or Notion. A pasted prompt supplies instructions and context for the chat; it does not become a system instruction or guarantee memory in future conversations.
Before asking for advice, have the model separate:
supplied assessment information;
your account, goals and constraints;
its own inferences;
missing information.
It should discuss the material you supplied without claiming to have read an unseen report. The broader habit is to inspect and revise the output, as described in OpenAI’s guide to using ChatGPT and checking results.
How to Use MBTI Results with ChatGPT: A Worked Career Example
This example concerns a small work trial, not a decision to resign or change careers.
This is a fictional teaching case. The person, scores, prompts, responses and plan are invented; they are not a real report, recorded AI conversation or evidence that the trials work. The Chinese and English articles use the same case, not two separate experiments.
The example uses a hand-picked result excerpt with the field meanings explained below, rather than a direct export from a result page.
What the selected result does and does not contain
FermatMind offers 93- and 144-question MBTI-style forms, estimated at about 10 and 15 minutes. Both currently allow free completion and corresponding results, with the available detail governed by the selected result page. More questions do not establish greater accuracy, and the forms are not the official MBTI® instrument.
E/I, S/N, T/F and J/P are the four core preference pairs. Additional A/T is separate; its T is not the T in T/F. A result cannot supply unmeasured cognitive functions, abilities or career suitability.
The teaching excerpt below selects only E/I: E 70% and I 30%. These are preference percentages within the questionnaire. Their 40-percentage-point difference is arithmetic, not reliability or a probability that the type is correct.
When using your own material, keep any actual version, reference or quality information it provides. Do not replace a real warning with this example’s Unknown. Leaving a field out does not mean that it is zero or that the product lacks it.
First prompt: give the excerpt, your account and one bounded task
You can copy the whole prompt below to practise the teaching case. For personal use, replace the synthetic content with material you are entitled to disclose, while keeping measured information separate from your own account.
This is a synthetic teaching exercise, not a real user record. Help me discuss a low-risk work trial; do not choose a career for me.
SELECTED RESULT EXCERPT
Provider: FermatMind, an MBTI-style assessment, not the official MBTI® instrument.
Form: mbti_144; 144 questions.
Result-material language: English.
Answer language, date, questionnaire build version, scoring version, reference and quality: not supplied in this teaching material; leave Unknown.
Selected E/I: E 70%, I 30%. Unit: preference percentages within this questionnaire.
Other three pairs, full type and additional A/T: not selected here. Do not infer them; omission does not mean zero or that the product lacks them.
No ability score, career-fit score, percentile, type-accuracy figure or success probability was supplied.
A difference of 40 percentage points between the two displayed values is arithmetic, not reliability.
MY ACCOUNT, NOT MEASURED FACTS
I am employed and am not considering resignation now. I want to investigate whether I would like more discussion and explanation tasks in my current work.
A brief one-to-one alignment conversation often helps me start writing. A six-person meeting without an agenda leaves me tired.
Last week I also independently completed a one-page explanation successfully. Keep this counterexample.
I prefer a short answer first, then steps. That is my own editable communication preference, not something inferred from E/I.
This week I have at most two hours and no budget. I will use the same public or explicitly authorized non-confidential material, without another person's private records or company secrets.
TASK
First separate supplied result information, my account/goals/constraints, your inferences and missing information.
Confirm the original units. Leave Unknowns unknown; do not invent a full type, A/T or quality, or claim to have read an unseen report.
Propose two reversible small trials: discuss briefly before writing a page; write independently before receiving feedback.
Give steps, output, observations, information gaps, a counterexample, editable action, stop conditions and a review point, within two hours total.
Do not recommend an occupation from a type or one preference pair. Keep preferences, interests, abilities and practical feasibility distinct.
Explain which gaps matter before a future career decision but do not prevent this week's small task.
Use an accessible original source for an important external factual claim, or label it unverified. Do not claim to have permanently remembered me.
The request for a short answer followed by steps comes from the person, not from the E/I values. The task also leaves room for both discussion and independent writing rather than assuming one must be the right way to work.
An error to catch
The following wrong answer is a teaching construction, not an observed ChatGPT response:
“E70 puts your social ability above 70% of people. The 40-point gap means the result is only 40% reliable. You should move into a communication-heavy role and leave writing-focused work soon.”
Three claims need to be withdrawn:
Claim
Error
Required correction
Social ability is above 70% of people
A questionnaire preference was converted into an ability percentile without an ability measure or population reference.
Withdraw the ability and percentile claim.
The result is 40% reliable
An arithmetic difference was converted into reliability or statistical confidence.
Withdraw that interpretation and preserve the original numbers.
The person should change roles
One preference pair was converted into a career verdict, ignoring successful independent writing and the stated constraint.
Withdraw the job-change recommendation and return to the small task.
Changing “you should” to “you might” would not fix an unsupported inference. The correction needs to change the reasoning and the proposed action.
Ask the model to retract the error
Withdraw those three unsupported claims, rather than only softening the wording.
E70/I30 are not social ability or population percentiles. A 40-percentage-point difference is not reliability or type accuracy.
The other pairs, full type, A/T, quality and reference were not supplied. Preserve the original values; do not edit the result to fit your advice.
You omitted my successful independent one-page draft and my constraint that I am not considering resignation now.
Separate supplied result, my account, inference and Unknowns again. Explain how fatigue in unstructured meetings and successful independent writing limit your interpretation.
Give two zero-cost trials totaling no more than two hours, using the same authorized non-confidential material. Explain how order and familiarity affect the observations; these informal trials cannot isolate the effect of discussion or validate ability.
Return an editable action card with stops, a review point and the information still needed before a future career decision.
A corrected response and action plan
The response below is also constructed for teaching. It shows the substance a correction should contain, not a guaranteed model response.
Retraction
I withdraw the claims about social-ability percentile, 40% reliability and a need to leave writing-focused work. None is supported by the supplied material. E70/I30 remain preference percentages within this questionnaire. Their numerical difference does not establish reliability, a full type or career suitability.
What can be used
Category
Current information
Supplied result
mbti_144, 144 questions; English result material; E 70% and I 30% as questionnaire preference percentages
Your account and constraints
Brief alignment often helps you begin; an unstructured six-person meeting is tiring; you have also completed a page independently; not considering resignation now; two hours and no budget
Working hypothesis
A short, purposeful conversation may sometimes help you start writing. This comes from your account, not a measured effect of E/I.
Counterexample and alternatives
Successful independent writing shows that discussion is not always necessary. Agenda, group size, familiarity and daily condition may also explain the experience.
Unknowns
Answer language, date, build and scoring versions, reference and quality, plus the unselected pairs and full type. Unknown does not mean normal or passed.
Two small trials
Trial
Time
Output and observation
A: discuss briefly, then draft
10 minutes of voluntary discussion; 30 minutes drafting one page; 20 minutes for a willing reader to restate the main points, give feedback and help you complete a brief record
One page, the reader’s actual points of confusion and a record of assistance needed
B: draft independently, then get feedback
40 minutes independently drafting with the same material; 20 minutes of comparable restatement, feedback and brief recording
One page and the same observation questions
Total
120 minutes; no spending
Notes and the immediate review fit within the feedback periods rather than becoming an assumed extra session.
For this example, you could do A before B and record that order. The second attempt benefits from greater familiarity with the same material. Reversing the order does not automatically remove that effect. Discussion time, different readers and conditions also affect the observations.
These trials can help you decide what to investigate next. They cannot isolate the effect of discussion, measure ability, validate a personality result or demonstrate that the method improves outcomes.
Record only what happens
Observation
Trial A
Trial B
Completed within the planned time; actual time used
To be filled
To be filled
Points the reader misunderstood
To be filled; mark incomplete if feedback was not obtained
To be filled; mark incomplete if feedback was not obtained
Assistance needed
To be filled
To be filled
Fatigue and familiarity with the material
To be filled
To be filled
Willingness to repeat the task the next day
Not yet observed; leave unrecorded if not followed up
Not yet observed; leave unrecorded if not followed up
Do not extend the time budget merely to fill every box. Do not invent reader feedback, a fit score or a percentage improvement. The optional next-day note also needs to fit within any time left in the two-hour budget; leave it unrecorded if you do not follow up. A next-day observation cannot be completed in advance.
Information gaps
This week’s task does not require a final career choice or a complete personality interpretation. It mainly examines a hypothesis drawn from your account. You need suitable material, a feasible time limit and willing participants.
Before a later career decision, you would need to know how often the relevant tasks occur in a target role, what skills or qualifications it requires, what your existing work samples show and what opportunities or constraints apply.
For a concrete look at the work, read what technical writers do: checking the intended audience, confirming product versions, working with technical reviewers and revising instructions. Use those tasks to form questions about the job; they are not a recommendation based on E70. The career library offers other roles to investigate.
Stop or reduce the task
Pause or reduce the scope if the material’s permitted use is unclear, it contains confidential information, no one is willing to participate, or the activity is likely to overrun. If feedback does not happen, record it as incomplete rather than asking AI to manufacture it.
Review
After the trials, choose whether to keep exploring, adjust the task or pause. You can revise the goal, time allocation, communication format and plan. You cannot rewrite the measurement to support the outcome you prefer.
These trials cannot establish career fit, income or hiring prospects. A useful result may simply be a better question about the work you want to do.
Using RIASEC Interests: Another Small Trial
This second case is also fictional. Its person, result, prompts, responses and plan are invented for teaching; no private report is attached and no outcome is claimed.
Synthetic input:
Provider: FermatMind. Language: English. Questionnaire: RIASEC, 60 questions. Date: not provided. Selected result: I, A and S are near the top; the full six-area scores and tie information are not provided, so do not infer the gaps. My account: I enjoy turning public science material into a one-page visual explanation, but put off revising the text. Goal: try one explanation task this week. Budget: one hour, without buying a course.
Try this prompt:
This is a synthetic exercise. First separate the supplied result, my account and your inferences. Leave missing values, dates and conditions unknown. Do not use a three-letter code to decide my occupation, ability or hiring chances. Give me two reversible ways to explain one public science idea in an hour, state what each needs checked, then let me choose. Finish with one action, an observable record and a stop condition.
A reasonable synthetic response:
The result supplies an interest-ordering clue. Enjoying visuals and delaying revision come from this account; neither establishes writing skill. You could draft a text explanation or a visual explanation. Check the source material and permission to reuse it, then ask whether a reader can restate the main point.
This is a proposal to try, not a new measurement of ability.
Corrections to make
If AI says “IAS means research is your ideal career and success is likely,” point out that the input contains no career-fit score or success probability. If it says “I is much higher than A,” remind it that no six-area values were supplied. If it calls a score a population percentile, check the original report and remove that claim unless the relevant reference was provided.
Check any external facts in the response against the original source before using them in your draft.
You might revise the action to:
Spend 45 minutes on Saturday drafting one page and 15 minutes asking a willing friend to restate the main point. Record whether I would repeat the task, what the reader found unclear and one revision for next time. Pause if I cannot verify the material and its permitted use, or if the task exceeds my one-hour budget.
Keep interest, personal observations, work feedback and practical conditions separate. The next decision can be to revise the structure, reduce the task or stop this direction. One trial need not settle a career choice.
For an interest-and-action record, use the career exploration worksheet. It offers a separate synthetic example, your own entries, TXT/Markdown output and local browser printing. Entries stay in page memory without autosave and may be lost when you refresh or leave; check and take away the record first.
The blank PDF, your completed worksheet record and a personal assessment report are different outputs. The worksheet is filled in separately; it does not automatically import this example or your MBTI result.
For other assessments, preserve that report’s own units and limits. Do not fill missing results with an example. More complete input still does not give AI responsibility for your decision.
When Should You Stop and Find a Human?
Situation
Why AI is not enough
A better next step
A relationship involves commitment, breakup, betrayal or broken boundaries
A one-sided account cannot resolve another person’s intentions or the consequences.
Discuss only when safe and mutually wanted. If threatened, seek trusted or qualified support; do not confront someone because AI suggests it.
A career choice has high-cost or irreversible consequences
A generated comparison cannot bear the opportunity cost or establish local facts for you.
Check real jobs, consult people with relevant experience and examine your practical conditions.
Low mood, insomnia, panic or hopelessness persists
The issue goes beyond ordinary self-organization.
Seek appropriate professional mental-health support rather than relying on a coaching conversation.
Legal, medical, financial or compliance decisions are involved
These require qualified judgment about the actual circumstances.
Use AI to prepare questions and organize permitted information, not to make the decision.
You feel better after a conversation but are no clearer about the problem
Relief does not establish that an interpretation or plan is sound.
Pause and return to evidence, a safe reality check or wanted human feedback.
When responsibility, irreversible costs or sustained psychological risk enter the picture, AI can remain an organizer. It should not become the sole judge.
This Week’s Action Card: Audit Your AI-Coach Conversations
You can do this audit locally. You do not need to upload private conversations.
Pick three conversations you remember clearly. For each, write what you were trying to solve and whether you gained clarity, comfort or both.
Find one prompt that came closest to handing over judgment. Rewrite it as a request for information gaps, alternatives or ways to verify an option.
If AI often becomes an emotional mirror, consider one safe reality check or a trusted support contact after a few conversations. This is an optional reminder, not a validated quota. You do not owe anyone a disclosure or confrontation.
If you use AI mainly for structure, include one step that checks the plan against real conditions before acting on it.
The point is to keep a connection between the conversation and what actually happens. AI can help you organize a problem, but you still have to judge the evidence and live with the decision.
What to Do Next in FermatMind
If you already have MBTI-style or Big Five results, use them to raise questions rather than settle them. Observe whether your actual AI use leads to checking, action and wanted support. A score does not predict dependence.
If you want a result and do not already have one, the MBTI test and methods explain the available forms. Taking another test is not a prerequisite for using this guide.
For a work question, return to the career library and examine actual tasks and conditions. Do not remain in a conversation about which occupation a type supposedly determines.
If you increasingly rely on AI to stabilize your mood, consider appropriate human support rather than assuming that more frequent AI use will solve the problem. You can also use the non-AI version of this method: write down the facts, your account, missing information and one small action yourself.
Research Notes and References
The research sources below discuss sycophancy, coaching style, disclosure, attitudes, advice or trust in their own settings. They do not validate the four positions or eight editorial patterns in this guide, assign risk by MBTI type, establish a disclosure quota or prove the effectiveness of the proposed actions. Consider each study’s task, participants and limitations before applying it.
Research on sycophancy and coaching style
Cheng, M., et al. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391(6792), eaec8352. This supports the caution about excessive agreement in interpersonal advice, not a claim that a particular MBTI preference predicts dependence. An open 2025 preprint is available; its earlier experiments should not be confused with the final publication's sample.
Terblanche, N. H. D., & Brown, A. B. (2026). Rethinking directiveness in AI coaching chatbots. Frontiers in Psychology, 17, 1822088. This exploratory, single-session study of UK-based, English-first-language graduate millennials compared two coaching styles. Its findings do not validate this guide's patterns or establish effects across populations or long-term use.
Further research on disclosure, advice and trust
【1】Ho, A., Hancock, J., & Miner, A. S. (2018). Psychological, Relational, and Emotional Effects of Self-Disclosure After Conversations With a Chatbot. Journal of Communication, 68(4), 712–733. Study DOI.
【2】Lee, J., & Lee, D. (2023). User Perception and Self-Disclosure towards an AI Psychotherapy Chatbot According to the Anthropomorphism of Its Profile Picture. Telematics and Informatics, 85, 102052. Study DOI.
【3】Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm Appreciation: People Prefer Algorithmic to Human Judgment. Organizational Behavior and Human Decision Processes, 151, 90–103. Study DOI.
【4】Bogert, E., Lauharatanahirun, N., & Schecter, A. (2022). Human Preferences toward Algorithmic Advice in a Word Association Task. Scientific Reports, 12, 14501. Study DOI.
【5】Papneja, H., & Yadav, N. (2025). Self-disclosure to Conversational AI: A Literature Review, Emergent Framework, and Directions for Future Research. Personal and Ubiquitous Computing, 29, 119–151. Study DOI.
