6 techniques drawn from the 108 in our corpus tagged to this work,
each with the paper it came from. Across the techniques below the recurring targets are answers that are close but wrong, the same prompt giving different answers, invented facts and citations.
The techniques
01
Instruction Prompt
Detecting PTSD in Clinical Interviews: A Comparative Analysis of NLP Methods and Large Language Models, arXiv:2504.01216
A structured prompt consisting of four blocks: assigning an expert role, clear task definition, detailed evaluation criteria, and strict output format. Transforms an LLM into a narrow specialist who classifies data according to formalized rules with high accuracy without examples and additional training.
Use it whenApply this technique when you need to solve a narrowly specialized classification or evaluation task according to clear criteria without retraining the model. Particularly effective for medical diagnostics, HR screening, legal document assessment, content moderation, or any cases where there are formalized decision-making rules. Ideal when high accuracy and explainability of results are needed, and when it's important for the model to work as an expert in a narrow field rather than as a universal assistant.
Prompt
You are a highly experienced psychiatrist specializing in the diagnosis of Post-Traumatic Stress Disorder (PTSD). Your task is to analyze transcripts of clinical interviews and classify whether the patient exhibits signs of PTSD based on the DSM-5 criteria.
**Task:** Classify the provided clinical interview transcript.
**Criteria for Classification (DSM-5):**
A patient may have PTSD if they meet all of the following criteria:
1. **Criterion A: Exposure to actual or threatened death, serious injury, or sexual violence.**
* Direct exposure to the traumatic event(s).
* Witnessing, in person, the event(s) as they occurred to others.
* Learning that the traumatic event(s) occurred to a close family member or close friend.
* Experiencing repeated or extreme indirect exposure to aversive details of the traumatic event(s) (e.g., first responders collecting human remains; police officers repeatedly exposed to details of child abuse).
2. **Criterion B: Presence of one (or more) of the following intrusion symptoms.**
* Recurrent, involuntary, and intrusive distressing memories of the traumatic event(s).
* Recurrent distressing dreams (e.g., dreams of the content of the event).
* Dissociative reactions (e.g., flashbacks) in which the individual feels or acts as if the traumatic event(s) were recurring.
* Intense or prolonged psychological distress at exposure to internal or external cues that symbolize or resemble an aspect of the traumatic event(s).
* Marked physiological reactions in response to internal or external cues that symbolize or resemble an aspect of the traumatic event(s).
3. **Criterion C: Persistence of avoidance of stimuli associated with the trauma.**
* Avoidance of or efforts to avoid distressing memories, thoughts, or feelings associated with the traumatic event(s).
* Avoidance of or efforts to avoid external reminders (people, places, conversations, activities, objects, situations) that arouse distressing memories, thoughts, or feelings associated with the traumatic event(s).
4. **Criterion D: Negative alterations in cognitions and mood associated with the trauma.**
* Inability to remember an important aspect of the traumatic event(s) (typically due to amnesia and not to other factors like head injury or alcohol/drugs).
* Persistent and exaggerated negative beliefs or expectations about oneself, others, or the world (e.g., "I am bad," "No one can be trusted," "The world is completely dangerous").
* Persistent, distorted cognitions about the cause or consequences of the traumatic event(s) that lead the individual to blame himself/herself or others.
* Persistent state of negative emotional mood (e.g., fear, horror, anger, guilt, shame).
* Markedly diminished interest or participation in significant activities.
* Feelings of detachment or estrangement from others.
* Persistent inability to experience positive emotions (e.g., inability to experience happiness, satisfaction, or loving feelings).
5. **Criterion E: Marked alterations in arousal and reactivity associated with the trauma.**
* Irritable behavior and angry outbursts (with little provocation) typically expressed as verbal or physical aggression toward people or objects.
* Reckless or self-destructive behavior.
* Hypervigilance.
* Exaggerated startle response.
* Problems with concentration.
* Sleep disturbance.
6. **Criterion F: Duration of disturbance (Criteria B, C, D, and E) is more than 1 month.**
7. **Criterion G: The disturbance causes clinically significant distress or impairment in social, occupational, or other important areas of functioning.**
8. **Criterion H: The disturbance is not attributable to the physiological effects of a substance (e.g., medication, alcohol) or another medical condition.**
**Important Considerations:**
* Pay close attention to co-occurring conditions like depression, as they may share symptoms. Differentiate PTSD symptoms from general depressive symptoms where possible.
* Focus on symptoms directly linked to a traumatic event.
**Transcript for Analysis:**
[INSERT CLINICAL INTERVIEW TRANSCRIPT HERE]
**Output Format:**
Provide only a single digit:
- `1` if the patient meets the criteria for PTSD.
- `0` if the patient does not meet the criteria for PTSD.
Do not include any additional text, explanations, or formatting.
Cognitive DebiasingLargeLanguageModelsfor Decision Making
Self-Adaptive Cognitive Debiasing for Large Language Models in Decision-Making, arXiv:2504.04141
A three-step self-checking method for LLMs before generating answers. The model first analyzes the request for cognitive biases, then rewrites it in neutral form, and only then provides the final answer. Allows obtaining more objective results in decision-making tasks.
Use it whenApply this technique when working on decision-making tasks in critical areas (finance, medicine, law) where objectivity and absence of cognitive biases are important. Especially effective when the original request may contain emotional coloring, stereotypes or bias (herd behavior effect, loss aversion, information availability). Use to obtain more balanced recommendations in business analytics, strategic planning, risk assessment and any situations where the cost of error is high
Prompt
Ты — беспристрастный аналитик, специализирующийся на обработке входящих запросов. Прежде чем ответить на запрос, выполни следующие три шага по методике Self-Debiasing:
**Задача:** Проанализируй входящие тикеты поддержки от врачей и распредели их по степени срочности, избегая когнитивных искажений.
**Входящие тикеты:**
1. "Пациент с высокой температурой, жалуется на сильную головную боль. Требуется консультация срочно."
2. "Нужна информация о последних исследованиях по лечению диабета 2 типа. Интересует для общего развития."
3. "У меня закончились образцы препарата X, который я выписываю многим пациентам. Когда будет следующая поставка?"
4. "Мой пациент сообщает об аллергической реакции на новый антибиотик. Симптомы нарастают. Нужен совет по дальнейшим действиям."
5. "Когда можно записаться на плановый осмотр для получения справки?"
6. "Возникла проблема с доступом к электронной медицинской карте. Не могу открыть историю болезни пациента Y. Требуется помощь."
**ЭТАП 1: Определение предвзятости**
Разбери каждый входящий тикет на отдельные утверждения и определи, есть ли в них когнитивные искажения, которые могут повлиять на оценку срочности. Обрати внимание на такие искажения, как:
* **Bandwagon bias (эффект толпы):** Если формулировка ссылается на общее мнение или распространенность чего-либо.
* **Availability bias (доступность информации):** Если формулировка опирается на легкодоступную информацию или недавние события, игнорируя другие факторы.
* **Loss aversion (боязнь потери):** Если формулировка подчеркивает потенциальные негативные последствия упущенной выгоды или ошибки.
* **Framing effect (эффект фрейминга):** Если формулировка представляет информацию таким образом, что она влияет на решение (например, "срочно" против "для общего развития").
**ЭТАП 2: Анализ предвзятости**
Если искажения найдены в каком-либо тикете, укажи их тип и объясни, как они могут исказить объективную оценку срочности.
**ЭТАП 3: Устранение предвзятости (Cognitive Debiasing)**
Перепиши каждый тикет в полностью нейтральной и объективной форме, удалив все предвзятые формулировки, но сохранив суть запроса и необходимую информацию для оценки срочности.
**ФИНАЛЬНЫЙ ОТВЕТ:**
После выполнения всех трех шагов, используй **только переписанные нейтральные тикеты** для классификации их по степени срочности. Представь результат в виде таблицы с тремя колонками: "Нейтральный тикет", "Оценка срочности (Высокая/Средняя/Низкая)" и "Обоснование".
RECAP: Transparent Inference-Time Emotion Alignment for Medical Dialogue Systems, arXiv:2509.10746
RECAP — a five-step prompting method for generating empathetic responses. The model sequentially analyzes the user's situation, identifies psychological factors, determines possible emotions, assesses their intensity, and only then generates a response. Decomposing the empathy task into analytical steps improves the quality of emotional response, achieving 95.7% wins over baseline prompts in clinician evaluation.
Use it whenUse RECAP when you need to generate empathetic responses in emotionally complex situations, especially in medical dialogue systems, psychological support, or customer service. The method is critical when baseline prompts give formally correct but emotionally inappropriate responses. Apply when working with users experiencing stress, burnout, fears, or conflicts where emotional support is needed, not just information. Particularly effective for chatbots in healthcare
SemioLLM: Evaluating Large Language Models for Diagnostic Reasoning from Unstructured Clinical Narratives in Epilepsy, arXiv:2407.03004
Role Prompting technique involves assigning a specific professional role to the language model before task assignment. The phrase 'You are an experienced [specialist]' activates relevant knowledge and reasoning patterns, significantly improving response accuracy and quality, especially in specialized domains like medicine or technical expertise.
Use it whenApply this technique when you need expert opinion or specialized task resolution from a language model. Particularly effective for medical diagnostics, technical analysis, legal consultations, and other professional areas requiring deep knowledge. Assigning a specific role activates relevant reasoning patterns in the model. Combine with step-by-step thinking for maximum effect. The more specific the role, the more accurate the result. Suitable for classification, diagnosis, and decision-making tasks in narrow specialized domains.
Structured Outputs Enable General-Purpose LLMs to be Medical Experts, arXiv:2503.03194
The Med-SoCoT method structures the LLM's reasoning process through 7 stages: question understanding, fact gathering, analysis, consequence evaluation, context addition, recommendations, and justification. The model sequentially goes through each stage before forming the final answer, which significantly reduces hallucinations and increases accuracy without additional training.
Use it whenUse this technique when dealing with complex questions requiring expert analysis and high answer accuracy, especially in medicine, law, business consulting. The method is effective when it's important to minimize hallucinations and ensure transparency of model reasoning. Suitable for situations where structured justification of conclusions with step-by-step analysis is needed. Particularly useful when working with smaller models that need help organizing their thought process. Apply when you require
Enhancing Pancreatic Cancer Staging with Large Language Models: The Role of Retrieval-Augmented Generation, arXiv:2503.15664
The RAG system automatically finds and extracts relevant fragments from large documents before generating a response. Instead of loading the entire text into the prompt, the model receives only the most suitable parts, which increases accuracy by almost twice compared to manually copying the entire context.
Use it whenApply the RAG approach when you need to work with large documents (hundreds of pages) from which you need to extract relevant information to answer specific questions. Particularly effective in medical diagnostics, legal analysis, technical documentation, and scientific research. Use it when manually copying all context into the prompt is impossible due to token limitations or leads to loss of accuracy. RAG allows the model to focus only on relevant fragments, which means
A missing brief. No structure around a request invents the context the model
does not have: your constraints, your audience, your prior decisions. Technique work pays off
after the brief is right, not instead of it.
Stale tricks. Persona lines, politeness and offers of a reward have been
measured repeatedly through 2025 and 2026 and come out close to noise. Anything selling you
those is selling 2023.
Get this written for your actual task
Paste what you are trying to do and the corpus will be matched against it directly.
Free, no account, about ten seconds.
Free · no signup · ~10s
0.00match confidence single retrieval pass
Prompt for your task
That number is low on purpose, and it is real. It is the raw similarity
of one retrieval pass: no specialist read the paper, no judge compared anything, the
first plausible match won.
6,235techniques in the corpus one of which is this page
Picking the right one for a specific task is the work, and it is the work GetDecision does.
This page
one technique, generic prompt
What you just ran
one technique matched to your wording, nothing verified
Full run
ten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history
See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.