What's the AI tip? Teach you to write perfect Prompt

PromptWhat is it

Many people think that “Prompt = Directive to AI” ignores Prompt as a set of people working with AI. It is not only about “doing what”, it is about defining “why”, “do what”, “do what”, and what”, which is essentially about translating people's vague needs into AI's understandable and implementable structural tasks。

From a technical logic, Prompt ' s central role is to guide LLM in the mobilization of pre-training knowledge through input sequences, to focus on specific mission scenarios and to generate outputs that meet expectations。

LLM itself has no experience in the world of work, nor can it interpret the meaning of it, and the leadership’s offer to his subordinates to make a sea trip may imply that he was more interested in the South-East Asian market than his previous boss had mentioned, but AI would only produce generic content literally. So Prompt's core value is to eliminate poor information (both between human needs and AI understanding and between mission objectives and implementation standards). I'm not sure

Some errors on Prompt

Mistake 1: The more complex, the more professional, the more overstretched the terminology and format

It was felt by many that Prompt needed to include industry standards, professional terms, complex labels to reflect professionalism. For example, AI is required to “manufacture the enforcement system with an ISO 8402 terminological analysis, following the ISA-95 standard”, but the actual result is often to export the obscure theoretical text of AI, which is completely different from the actual use scene。

Truth:Prompt's professionalism is not in complexity, but in precision matching. The current LLM (e.g. Claude 4, GPT-4) has a strong understanding of the natural language and can achieve the same or even better results without the need for XML labels, specialized terminology and clear titles, segments, and guides. The real professional Prompt is to fit the content of the AI output with the scene, not to make Prompt itself as obscurant。

Mistake 2: Just say what to do and ignore hidden demands and constraints

USERS CONSIDER IT CLEAR ENOUGH FOR AI TO WRITE AN OUTREACH PROJECT AND FOR AI TO DO A DATA ANALYSIS, BUT IGNORE KEY MESSAGES SUCH AS WHO THE TARGET AUDIENCE IS, WHAT THE CORE POINTS ARE, AND WHAT THE OUTPUT FORMAT IS。

Truth:LLM HAS NO SO-CALLED INDUSTRY COMMON SENSE AND NO DEFAULT SETUP. FOR EXAMPLE, A PROGRAMME MAY BE A STRATEGIC VERSION FOR CEOS, OR A LAND-BASED VERSION FOR SALES; A TWEET MAY BE DIRECTED AT TECHNOLOGY DECISION MAKERS OR FRONT-LINE OPERATORS ... IF THESE HIDDEN NEEDS ARE NOT CLEAR, AI CAN ONLY GUESS BLINDLY, AND THE OUTPUT WILL CERTAINLY NOT MEET YOUR PSYCHOLOGICAL EXPECTATIONS。

Error 3: One key generation is the end point, rejecting iterative optimization

Many expect a perfect result from one input, and if the output does not match expectations, they will assume that AI is no good and will not spend time optimizing Prompt. In fact, Prompt's optimization process is essentially a process of gradual clarity of needs, not only to make AI more aware of you, but also to make your core claims more clear to you。

Truth:In some work scenarios, such as client programmes, industry reports, the process itself needs to be repeated. AI is just a tool for you to speed up. For example, when you wrote the ERP programme with AI, Prompt was only given a generic framework for the first time, and then added “the target customer is a food-processing enterprise, with a focus on a batch-retroactive module”, the alignment was increased and the programme was further strengthened by adding “citing three client case data”。

Bottom logic of Prompt capabilities: structured thinking + precision expression。

The essence of Prompt capabilities requires users to have:

• Demand for dismantling capacity:Dismantling vague targets into specific, implementable sub-tasks
• Structured expression:(A) ORGANIZE INFORMATION WITH CLEAR LOGIC TO ENABLE AI TO QUICKLY CAPTURE THE CORE
• The power of the scene:Define the standards and style of the output from the point of view of using the scene and the audience
• Reciprocal optimization:The instructions are continuously adjusted through test feedback to achieve optimal results。

It also explains why it is difficult to use AI when it comes to leadership that is not even clear about instructions to subordinates, because the quality of Prompt ultimately depends on the depth of the user ' s thinking and expression。

Bottom logic for building efficient Prompt

1.Roles, needs, scenes, targets

Prompt's first task is to make it clear who you are, who you are, what you're doing, what you're doing, what you're going to do, what you're going to achieve

• Role:
• (a) Clearly define the role of the input (e.g., you are a sales member of HR SaaS) and determine your position
• Audience alignment:IDENTIFYING RECIPIENTS OF OUTPUTS (E.G. MANUFACTURING HRD, INTERNET PRODUCT MANAGERS, GOVERNMENT DECISION MAKERS) AND DETERMINING THE PROFESSIONAL DEPTH AND LANGUAGE STYLE OF CONTENT
• scene alignment:(a) Clearly use the scenes (e.g., client selection presentations, internal wrap-up sessions, industry community outreach) to determine the focus and presentation of content
• Target alignment:Clarify core objectives (e.g., convey product values, address specific issues, provide a basis for decision-making) and determine core logic and key messages of content。

The error demonstration:An article about HR SaaS products。

Error point:WITHOUT CLEARLY IDENTIFIED AUDIENCES, SCENES AND TARGETS, AI CAN ONLY OUTPUT A GENERIC PRODUCT DESCRIPTION AND CANNOT BE USED DIRECTLY。

The correct demonstration:Now that you are a highly professional HR SaaS marketing firm, please write a HR SaaS product promotion for manufacturing HRD for industry community orientation, with the core objective of highlighting the function of "task data and production scheduling" and attracting target users to a free trial link。

Analysis:AUDIENCES, SCENES, OBJECTIVES ARE CLEAR AND AI OUTPUTS ARE MORE TARGETED。

2.STRUCTURED EXPRESSION TO LOWER AI'S UNDERSTANDING COSTS

LLM processes structured information much more efficiently than fragmentation information. When constructing Prompt, use the logical framework of total-point-total-dimensional dismantling to enable AI to quickly capture the core command:

• Core command prefix:(a) To make it clear at the beginning what to do and avoid redundant information interference
• Here's an idea:Distinguishing " core tasks " , " background information " and " output format " with headings, serial numbers and subparagraphs
• The logic is clear:The logical chain of the mission is clearly presented with the words “because ...” “first ... and last ...” (or 1, 2 and 3 ...)。

The error demonstration:HELP ME ORGANIZE THE Q3 CLIENT FEEDBACK, SEE IF THERE'S ANY PROBLEM AND ADVISE THE PRODUCT DEPARTMENT, PREFERABLY WITH DATA SUPPORT。

Error point:INFORMATION FRAGMENTATION MAKES IT DIFFICULT FOR AI TO COMBE PRIORITIES。

The correct demonstration:

• MANDATE: COLLATE CLIENT FEEDBACK ON Q3 CRM PRODUCTS AND GENERATE DEMAND ANALYSIS REPORTS。

BACKGROUND: USED FOR THE PRIORITIZATION OF THE REQUIREMENTS OF THE PRODUCT DEPARTMENT VERSION V.2.3
Core logic: first, a breakdown of feedback by “functional optimization, system stability, service support”, then a breakdown of the percentage of clients per type of feedback, followed by recommendations for improvements to Top3 distress points
Data requirements: Number and percentage of client feedback for each requirement。

Analysis:THE STRUCTURE IS CLEAR, AND AI CAN BE IMPLEMENTED IN A LOGICAL STEP-BY-STEP MANNER。

3.The scene fits the actual context

Okay, Prompt not only finishes the job, but also adapts to the scene. For example, also in the case of product presentations, content for technology decision makers needs to highlight technical architecture and integration capabilities, content for business decision makers needs to highlight ROI and business value, and content for front-line operators needs to highlight ease of use and efficiency improvements

At the heart of the fit-for-the-scenes is the ability to "replace":

• Think about the core demands of using the scene:In the case of client selection, there is a need to highlight the “discretionary advantage”; in-house training is a need to highlight the “operational steps”
• Thinking about the level of awareness of the audience:In the case of non-skilled persons, avoiding professional terminology; in the case of industry experts, the professional details could be deepened as appropriate
• How to use content:FOR PPT PRESENTATIONS, THEY NEED TO BE CONCISE AND REFINED; FOR WRITTEN REPORTS, THEY NEED TO BE DETAILED AND RIGOROUS。

Case in point:

THE MES SYSTEM PRESENTATION FOR THE INSPECTOR GENERAL OF MANUFACTURING, WHICH FOCUSES ON THREE CORE DEMANDS FOR PRODUCTION EFFICIENCY IMPROVEMENT, QUALITY RISK CONTROL, EQUIPMENT OPTIMIZATION, AND VAGUE FORMULATIONS SUCH AS A REDUCTION IN THE TIME SPENT ON EQUIPMENT BY 201 TP3T AND A REDUCTION IN THE BAD RATE 151 TP3T TO REPLACE THE STRONG FUNCTIONALITY OF QUANTITATIVE DATA, ALIGNS THE WORK FOCUS OF THE CHIEF PRODUCTION OFFICER。

4.Over and over again

Prompt, which was written for the first time, is almost impossible to perfect, and the efficient use of AI requires the establishment of a test-feedback-optimization loop:

• Preliminary test:The first version of the output is generated by simple Prompt to determine whether AI ' s understanding of core needs is accurate; (testing)
• Problem positioning:Analyse the deviations of the output (e.g., failure to highlight core functions, inconsistent language styles, lack of data support) and locate missing information from Prompt
• Precision optimization:TARGETED ADDITIONAL INFORMATION (E.G., HIGHLIGHTING XX FUNCTIONALITY, MORE FORMAL LANGUAGE STYLE, CITING DATA FROM THREE CLIENT CASES) RATHER THAN REWRITING IT AS A WHOLE
• Version comparison:Keep multiple versions of Prompt and output and see which formulation is more in line with AI ' s understanding logic。

Practising layers: basic to progressive Prompt

1.Basic approach

At the heart of the basic skills are “clear things” and apply to most simple tasks (e.g., writing short papers, collating data, answering questions) without complex logical design to produce the results quickly。

Skills 1: Demand dismantling - translating vague needs into specific tasks

Core logic:DISMANTLING “DOING WHAT” TO “VERB + OBJECT + CONSTRAINT” TO ALLOW AI TO SPECIFY “WHAT ACTIONS ARE CARRIED OUT, WHAT CONTENT IS ADDRESSED AND WHAT RULES ARE FOLLOWED”:

• Verb:(b) Clarify core actions (e.g., analysis, writing, collating, comparison, generation, etc.) and avoid the expression of vague words
• Object:CLEAR CORE CONTENT (E.G. Q3 SALES DATA, DIGITAL TRANSFORMATION PAIN POINTS FOR MANUFACTURING, “ERP PRODUCT CORE FUNCTIONS, ETC.)
• Binding:Clear border conditions (e.g. region-by-region, highlighting three core pain points, not involving technical details, etc.)。

hands-on template: [verb] [object]

Practising template: [verb] [object]

Binding:

[dimensional 1, e.g. audience/scenario]

[Drive 2, such as core elements]

[dimensional 3, e.g. format/long]

Case in point:This post is part of our special coverage Global Development 2011

Binding:

1. Audience: business owners who do not understand technology

2. Core elements: highlighting the cost advantages and difficulties of digital transformation

Format: 3 paragraphs, each not exceeding 150 words, with a link to the transformation self-assessment questionnaire at the end。

SKILLS 2: CONTEXT COMPLETION - PROVIDE KEY BACKGROUND FOR AI

Core logic:THE QUALITY OF LLM OUTPUT DEPENDS ON THE BACKGROUND INFORMATION ENTERED, AND IT IS NECESSARY TO CLARIFY “WHY”, “WHAT LIMITS”, “WHAT REFERENCES”, TO HELP AI UNDERSTAND THE OPERATIONAL VALUE AND BOUNDARIES OF THE MISSION:

• Operational background:Describe the origin of the mandate (e.g., the need for new marketing strategies to respond to competitive shocks)
• Conditions of restraint:DESCRIBE THE LIMITS OF THE MANDATE (E.G. A BUDGET OF NOT MORE THAN HALF A MILLION, A THREE-MONTH IMPLEMENTATION CYCLE, COMPLIANCE REQUIREMENTS IN LINE WITH GDP R)
• References:
• PROVIDE RELEVANT DATA, CASES, HISTORICAL RESULTS (E.G. REFERENCE TO Q3 SALES DATA, USE OF XX CLIENT SUCCESS CASES)。

The error demonstration:Collating client feedback and generating reports。

Error point:THE BACKGROUND IS NOT SPECIFIED, AND AI MAY SEQUENCE FEEDBACK IN TIME AND HAVE NO REFERENCE VALUE。

The correct demonstration:Collating client feedback on Q3 HR SaaS products to generate demand analysis reports. Operating background: used for the prioritization of needs for version V2.3 of the Product Department; binding conditions: focus only on three modules: attendance management, pay accounting, and staff training; reference information: Q3 collects 120 feedbacks, of which 58 are for attendance, 32 for pay module and 30 for training module (* Note: these feedbacks are uploaded here)。

Skills 3: Format Definition - Clarify the Structure and Presentation of Output

Core logic:PRE-DEFINED OUTPUT FORMATS TO AVOID THE CONTENT GENERATED BY AI REQUIRES A SECONDARY SORTING PROCESS TO ENHANCE EFFICIENCY。

Common formats include:

• Text format:e.g. Markdown, breaklist, paragraph, dialogue
• Structure format:e.g. tables, JSON, PPT outline, flow chart (in Mermaid syntax)
• Special format:SUCH AS MAIL, REPORTS, PROGRAMMES, CASES, FAQ。

hands-on template:

Output format requirements:

Structure: 4 parts by Pain Point - Program - Case - Impact

Format: Markdown, 1 TP5T#, 1 TP5T1T#

Data presentation: Core data with rough labels and cases with tables (customer name - industry - effects)。

Case in point:ANALYSIS OF THE COMPETITION DIFFERENCES IN THE MAINSTREAM RM PRODUCTS IN PARAGRAPH 3, OUTPUT FORMAT:

1. Form, listing: product name - core function - price range - applicable scenario - competitive advantage

2. A 300-word summary is added below the table to recommend options for small and medium-sized manufacturing enterprises。

Skills 4: Example Guidance - Solve style/format puzzles with "Small Sample Tips"

Core logic:FOR TASKS FOR WHICH THE FORMAT REQUIRES A COMPLEX AND SPECIFIC STYLE (E.G., CASE WRITING, QUESTIONNAIRE DESIGN, CODE GENERATION), IT IS MORE EFFICIENT TO DIRECT AI WITH “1-3 EXAMPLES” THAN SIMPLY DESCRIBE THE FORMAT。

• Example requirements:Simplicity, with emphasis on “structure/style/ logic” rather than full content
• Example fit:The examples need to be consistent with the target mission scenario (e.g., the goal is to write retail trade cases, and the examples are also, to the extent possible, retail trade)
• Quantity control:one example would solve most of the problems and would increase the number of complex missions to two to three, avoiding excessive token。

The error demonstration:Write a client case of HR SaaS products for the retail trade。

Error point:WITHOUT A CLEAR FORMAT, AI OUTPUT CASES MAY BE STRUCTURALLY CONFUSED AND LACK DATA SUPPORT。

The correct demonstration:Writing of the retail trade HR SaaS product customer case (customers: supermarket chains, 10 stores, 300 employees) following the following example。

Example: Client case: a car spare parts business (500 people, 200 million per year)

PAINFUL: PRODUCTION IS NOT SYNCHRONIZED WITH HR SCHEDULE DATA, WITH A VACANCY RATE OF 151 TP3T; MANUAL ACCOUNTING ATTENDANCE TAKES 3 DAYS PER MONTH. SOLUTIONS: THE HR SYSTEM IS USED TO ENABLE SCHEDULING AND ATTENDANCE TO AUTOMATICALLY GENERATE REPORTS。

EFFECTS: DISMISSAL RATE REDUCED TO 51 TP3T, ATTENDANCE TIME REDUCED TO ONE DAY AND ERROR RATE TO ZERO。

THE CORE MESSAGE OF THE NEW CASE IS THAT THE PAIN IS THAT “DOOR SHOP STAFF ARE FREQUENTLY SCHEDULED AND HR COMMUNICATION COSTS ARE HIGH; OVERTIME ACCOUNTING IN THE FRESH SECTOR IS COMPLEX”, AND THAT THE SOLUTION IS “DOOR SHOP SELF-SCHEDULED + AUTOMATIC OVERTIME ACCOUNTING”。

2.Progress Policy

For complex tasks (e.g., preparation of industry white papers, multi-dimensional analysis, annual programming), basic skills alone are difficult to meet, and a more advanced Prompt strategy is required to guide AI to in-depth thinking and logic。

STRATEGY 1: THE CHAIN OF THOUGHT - LET AI REASON

Core logic:THE COMPLEX TASK IS ESSENTIALLY A “MULTI-STEP PROCESS OF REASONING”, AND THE CHAIN OF THOUGHT LEADS TO A LOGICAL STEP-BY-STEP ANALYSIS BY AI BY MEANS OF A CLEAR “DOCTIVE STEP”, AVOIDING THE OUTPUT OF PARTIAL OR LEAPFROGGING CONCLUSIONS。

• Step design(i) Designing the logical steps of the business (e.g., competitive analysis: identifying functional differences in the matching of user needs to collect feedback from clients to summarize competitive advantages)
• Step constraints:EACH STEP CLEARLY DEFINES “WHAT TO DO”, “OUTPUT WHAT” AND AVOIDS AI SKIPPING CRITICAL LINKS
• Direct completion:IF THE AI REASONING IS INCOMPLETE, IT CAN BE SUPPLEMENTED BY GUIDANCE (E.G., IN THE SECOND STEP, THE DIFFERENCE BETWEEN THE PARTIES IN THE IMPLEMENTATION CYCLE)。

For example:

FIRST OF ALL, GIVE A BASIC DESCRIPTION OF THE PRODUCTS OF YOUR COMPANY ' S ERP (OR UPLOAD THEM DIRECTLY FOR FEED)

ANALYSIS OF THE DIFFERENCE BETWEEN OUR ERP PRODUCTS AND THE COMPETITION OF K/3 WISE IN SMALL AND MEDIUM-SIZED MANUFACTURING ENTERPRISES, WITH GRADUAL REASONING IN THE LINE OF THOUGHT: FIRST STEP: IDENTIFICATION OF THE CORE NEEDS OF SMALL AND MEDIUM-SIZED MANUFACTURING ENTERPRISES (ADDITIONAL INFORMATION: BASED ON Q3 STUDIES, WITH A FOCUS ON COST ACCOUNTING AND MES INTEGRATION AND IMPLEMENTATION CYCLES); SECOND STEP: COMPARISON OF THE FUNCTIONAL DIFFERENCES BETWEEN THE THREE CORE NEEDS OF EACH PRODUCT (E.G., COST ACCOUNTING: WE SUPPORT BATCH ACCOUNTING AND THE MONTHLY ACCOUNTING OF THE BUTTERFLIES); THIRD STEP: REFERENCE TO Q3 CUSTOMER FEEDBACK (COMPLEMENTARY SUMMARY OF CUSTOMER FEEDBACK) TO VERIFY THE PRACTICAL IMPACT OF FUNCTIONAL DIFFERENCES; FOURTH STEP: SUMMARY OF COMPETITIVE ADVANTAGES AND IDENTIFICATION OF FIT SCENARIOS FOR MY PRODUCT。

Final output: presented according to the table " Demand dimension -- our function -- butterfly function -- client feedback -- competitive advantage " 。

Strategy 2: Distinction of tasks - Dismantling complex tasks into children

Core logic:COMPLEX TASKS OFTEN INVOLVE “INFORMATION GATHERING AND ANALYSIS OF OUTPUTS OPTIMIZED”, AFTER WHICH EACH LINK FOCUSES ON A SINGLE OBJECTIVE, AND AI OUTPUTS ARE OF HIGHER QUALITY AND ALLOW USERS TO CONTROL THE RESULTS OF EACH LINK。

• Disaggregation principle:Disaggregation by business process progression (e.g., white paper: Data collection and analysis of pain and pain design framework to optimize language)
• Association logic:Clear linkages between sub-missions (e.g., the second step pain point analysis based on first step industry data)
• Step-wise optimization:Optimizing Prompt to move to the next step after the result of each sub-task to avoid an accumulation of errors。

Case in point:Preparation of a White Paper on the Digital Transformation of Small and Medium-sized Manufacturing Enterprises in 2025 (3,000 words)

Sub-Task 1 (data collection): collection of digital transformation data from 2024-2025 for small and medium-sized manufacturing enterprises, with emphasis on digital penetration, core input orientation, key factors for success in transformation, labelling of data sources (IDC/McK) and output of Markdown tables

Sub-mission 2 (analytical pain points): based on sub-mission 1 industry data, analysis of Top3 pain points for the digital transformation of small and medium-sized manufacturing enterprises, with data support for each pain point (e.g. 30% enterprises providing feedback on inefficient production schedules), list of output points

Sub-Target 3 (design framework): Based on the pain analysis of sub-task 2, design the White Paper framework, which contains section 5 on “The state of the industry — core pain points — transformation pathways — case references — selection recommendations”, each of which sets out the core elements

Sub-Task 4 (filling): prepare the full text of the White Paper in the framework of Sub-Task 3 with a language style professional and common, and the case highlights in part the application of our product (which needs to be supplemented by some specific information on your company ' s products) and output the Markdown format

Sub-mission 5 (optimal language): Optimization of the full text of sub-mission 4, correction of grammatical errors, adjustment of the paragraph logic to ensure fluidity and output of the final version。

STRATEGY 3: ROLE EMPOWERMENT - PUT AI IN A PROFESSIONAL PERSPECTIVE

Core logic:SET AI “SPECIFIC ROLE + INDUSTRY EXPERIENCE + CORE COMPETENCIES” TO GUIDE IT IN THINKING FROM A PROFESSIONAL POINT OF VIEW AND TO OUTPUT MORE RELEVANT CONTENT. ROLE-SETTING REQUIRES “PRECISION RATHER THAN GENERALIZATION” AND AVOID VAGUE EXPRESSIONS SUCH AS “THE WORLD'S TOP EXPERT”。

• Role composition:INDUSTRY BACKGROUND (E.G. FIVE-YEAR MANUFACTURING MES SALES EXPERIENCE) + PROFESSIONAL COMPETENCIES (E.G. FAMILIARIZATION WITH THE AUTO INDUSTRY PRODUCTION PROCESS) + CORE PERSPECTIVES (E.G. FOCUS ON CLIENTS ROI)
• Perspective guide:(a) Clarifying the core focus of the role (e.g. the Director of Production is concerned with efficiency and quality, the Director of Finance is concerned with cost and compliance)
• Avoid over-set:No unrelated capacity is stacked (e.g., preparation of a financial programme does not require a "proficient programming" requirement)。

The error demonstration:ASSUMING YOU'RE AN EXPERT IN THE INDUSTRY, WRITE AN INTRODUCTION TO THE MES SYSTEM。

Error point:Role blurry, output broad。

The correct demonstration:NOW THAT YOU ARE A CONSULTANT WITH FIVE YEARS OF PRE-SALE EXPERIENCE IN THE AUTOMOBILE INDUSTRY, WHO IS FAMILIAR WITH NEW ENERGY VEHICLE PRODUCTION PROCESSES, PLEASE WRITE TO THE DIRECTOR-GENERAL OF AUTOMOBILE PRODUCTION AN INTRODUCTION TO THE MES SYSTEM, FOCUSING ON “IMPROVING PRODUCTION EFFICIENCY, RETROACTIVITY OF QUALITY, OPTIMIZATION OF EQUIPMENT USE” FROM THREE ANGLES, WITH LANGUAGE STYLES THAT REPLACE VAGUE EXPRESSION WITH QUANTITATIVE DATA。

Policy 4: Pre-fill responses - Forced output structured format

Core logic:FOR TASKS REQUIRING A FIXED FORMAT (E.G. JSON DATA, FORMS, DEMAND LIST), PREFILL A PARTIAL CONTENT FRAMEWORK THAT ALLOWS AI TO FILL IN KEY INFORMATION DIRECTLY, AVOID REDUNDANT REPRESENTATIONS AND CAN BE IMPORTED DIRECTLY INTO THE SYSTEM。

• Format frame:DESIGN ON A PHYSICAL BASIS (E.G. JSON FOR SYSTEM IMPORT, TABLES FOR DOCUMENT COLLABORATION)
• Field definition:IDENTIFY THE FORMAT REQUIREMENTS FOR EACH FIELD (E.G., PRIORITY: HIGH/MEDIUM/LOW, MODULE NAME: MES - PRODUCTION SCHEDULE)
• Remove redundancy:Specifically require only output of filled format without any mattress or explanation。

Case in point:DRAWING DEMAND FROM CLIENT FEEDBACK AND GENERATING DEMAND LISTS IN JSON FORMAT (FOR THE PRODUCT DEPARTMENT DEMAND MANAGEMENT SYSTEM)

Prompt: Draw core needs from the following client feedback, fill them in the pre-filled JSON format, and only export JSON without any mattress. Client feedback: We are e-manufacturing enterprises, using your MES system, to export only Excel, hoping to support PDF formats; moreover, equipment alarms can only be alerted in the system, hoping to be synchronized to company micro-mail。

ENTER JSON:

OUTPUT JSON:

Strategy 5: Uncertainty Management - Enhancing export credibility

Core logic:AI IS EXPLICITLY INFORMED THAT IT “DOES NOT KNOW ABOUT, DOES NOT FABRICATE INFORMATION” ESPECIALLY FOR DATA CLASSES, DE FACTO TASKS (E.G., INDUSTRY REPORTS, COMPETITION ANALYSIS) AND AVOIDS THE GENERATION OF FALSE DATA OR PARTIAL FINDINGS BY AI TO ENHANCE THE CREDIBILITY OF THE OUTPUT。

• Rule:(a) Clarify the manner in which “uncertain information” is labelled (e.g., for additional research, for insufficient data and for further validation)
• Data constraints:Emphasizing that “the generation of content based on the information provided does not create unreferenced data/cases”
• Distinction recommendations:Clear “data-supported recommendations” and “valid recommendations” to avoid misleading decision-making。

Case in point:ANALYSIS OF CAUSES OF Q3 CLIENT LOSS

Prompt: Analysis of the causes of the loss of Q3 CRM products, requesting:

1. Based on data provided (32 lost clients: 18 feedback services were slow to respond, 8 were overpriced and 6 were not explained)
2. (a) The form is organized by “cause of loss - number of clients - percentage”
3. Six families that did not state their reasons were identified as “to be supplemented” and did not speculate about the specific reasons
4. Part of the proposal distinguishes between “data-supported” (e.g., optimization of service response processes) and “validation required” (e.g., research does not indicate the real needs of clients for reasons)。

3.High-level skills

For super-complex tasks (e.g., multi-modular content generation, cross-cutting programme design, professional mapping), more innovative Prompt techniques need to be used to break conventional output limits, taking into account AI ' s capacity characteristics。

Skills 1: Cross-mode-linking - consolidate text, chart, data

Core logic:LLM not only produces text, but also charts by specific syntax (e.g. Mermaid, LaTeX) that combine text and chart to make the output more persuasive。

• Chart Generation:Generate flow charts, architecture charts, time-series charts (e.g., product module flow charts, business process diagrams) in Mermaid syntax
• Data visualization:Comparative data are presented in tables and hierarchical relationships in lists
• Cross-state collaboration:The text part explains what the chart means, the chart part supports the text view and forms an integrated output of the text + chart。

Case in point:DESIGN FUNCTIONAL ARCHITECTURE FOR ERP PRODUCTS

Prompt: Designing an ERP product functional framework for small and medium-sized manufacturing enterprises that requires:

1. Text section: Sub-finance module, production module, supply chain module, human resources module, describing core functions and solutions
2. Chart section: Generate functional architecture in Mermaid syntax to demonstrate the relationship between modules
3. DATA SUPPORT: THE QUANTITATIVE EFFECT OF REDUCING XX COSTS AND INCREASING XX EFFICIENCY IS INDICATED FOR EACH MODULE。

Skills 2: Knowledge Injection in the Field - Supplementary Information in the Professional Field

Core logic:For highly specialized areas (e.g., medical, legal, industrial), LLM pre-training knowledge may be lagging or insufficient, requiring the injection of field knowledge (e.g., industry standards, business processes, professional terms) in Prompt to enhance the professionalism of output。

• Type of knowledge:(A) INDUSTRY STANDARDS (E.G. COMPLIANCE WITH THE GSP, COMPLIANCE WITH ISO 9001), BUSINESS PROCESSES (E.G., BACKTRACKING OF PHARMACEUTICALS IN PHARMACEUTICAL ENTERPRISES, ETC.), PROFESSIONAL TERMS (E.G., BATCH RETROACTIVE, COMPLIANCE STATEMENTS, ETC.)
• Injection mode:Briefly describe in the background section, avoid stacking and focus on mandate-related knowledge
• Validation logic:AI IS REQUIRED TO QUOTE THE INFUSION OF KNOWLEDGE IN THE OUTPUT TO ENSURE PROFESSIONALISM。

Case in point:PREPARATION OF AN ERP PRODUCTS PROGRAMME FOR PHARMACEUTICAL ENTERPRISES

Prompt: Preparation of an ERP product programme for pharmaceutical enterprises

COMPLIANCE REQUIREMENTS: COMPLIANCE WITH THE GSP AND SUPPORT FOR A FULL LIFE CYCLE OF THE DRUG

(c) Core pain points: batch backtracking, complex reporting of compliance, slow inventory turnaround。

REQUEST: THE PROGRAMME HIGHLIGHTS HOW TO MEET THE GSP NORMS, ADDRESSES CORE PAIN POINTS, AND REFERS TO SPECIALIZED TERMS SUCH AS DRUG TRACEABILITY, COMPLIANCE STATEMENTS, ATTACHED TO THREE MEDICAL ENTERPRISE CLIENT CASES。

SKILLS 3: REMITTANCE LOOP EMBEDDED - OPTIMIZING AI

Core logic:Embedding a “self-checking” link in Prompt to allow AI to generate output, self-assessed and optimized according to pre-set criteria, and to reduce the number of artificial overlaps。

• Inspection criteria:Task-based targeting (e.g., whether core functions are highlighted, format requirements are met, data support is included)
• Optimizing command:Clarity on how to modify “if they do not meet the criteria” (e.g., by adding one relevant customer case if no case is included)
• Number of cycles:Sets a cycle of one to two and avoids excessive overlaps leading to redundancy of content。

Case in point:Writing HR SaaS Product Promotion Program

Prompt: Preparation of HR SaaS product promotions for the manufacturing of HRD. Core requirements:

1. (a) Highlighting of attendance and production scheduling
2. Including 1 client case
3. 400 words
4. End with a free trial link。

Self-censorship and optimization: Once a case has been produced, check whether the four above-mentioned requirements have been met, if not met, with targeted modifications (in the absence of a case, supplemented by a reduced number of words), and ultimately output the optimized text。

Site site: Prompt field template for four core operations

In order to put the above-mentioned approach to a specific business scene, I have assembled a field template for the four core scenarios of the To B industry, “Content creation, data analysis, programme planning, customer service”, which can be used either directly or adjusted。

Scenario 1: Content creation — industry white paper

Template core logic: Data support + pain point analysis + case validation + selection recommendations。

Mission: preparation of White Paper on Selection of Small and Medium Enterprises [Themes] in 2025 [industry] (3,000 words)

1. Basic information:

– TARGET AUDIENCE: [INDUSTRY] SMALL AND MEDIUM-SIZED ENTERPRISES [DECISION-MAKING ROLES SUCH AS IT DIRECTORS/FINANCE DIRECTORS]

– Core objective: to help target audiences to identify selection criteria and highlight our product strengths

– Style requirements: professional and practical, including data support and client cases, and avoiding a multiplicity of technical terms。

2. Mind chain steps:

ANALYSIS OF THE CURRENT STATE OF THE INDUSTRY: THE 2024 REPORT OF THE [AUTHORITY BODY, SUCH AS IDC/MACKENZIE] PRESENTS THE STATUS OF THE DIGITAL TRANSFORMATION OF THE [INDUSTRY] (E.G. PENETRATION RATE, INPUT SIZE) WITH DATA TABLES

2 SELECT PAIN POINT DISASSEMBLY: ANALYSIS OF THE CORE PAIN POINT OF THE [INDUSTRY] SMALL AND MEDIUM-SIZED ENTERPRISE SELECTION BY 4 DIMENSIONS OF “FUNCTIONAL MATCHING, INTEGRATION CAPACITY, IMPLEMENTATION CYCLE, COST INPUT” WITH DATA SUPPORT FOR EACH PAIN POINT (E.G. “60% ENTERPRISE FEEDBACK `INTEGRATION DIFFICULTY'”)

THREE SELECTION CRITERIA: FOR PAIN POINTS, FOUR CORE SELECTION CRITERIA FOR "FUNCTIONAL SUITABILITY, TECHNOLOGICAL MATURITY, SPEED OF SERVICE RESPONSE, ROI" ARE PROPOSED, EACH WITH SPECIFIC MEASURES (E.G. "ROI: ONE YEAR BACK")

4 Comparative analysis of competitions: highlighting our strengths (e.g., “We have a three-month implementation cycle and six-month competitions”) over differences in the selection criteria between our products and [competition 1/competition 2]

5 Client case validation: 3 [industry] customer selection cases, presented as “customer background - selected pain points - solutions - use effects”, highlighting the actual value of our product

Six alternative hands-on advice: four phases of “demand combing - product testing - implementation planning - impact assessment” with concrete operational steps。

3. Output format:

– Markdown format, which contains the “title - Catalogue - Text (Part 6) - Final (with entry for the probationary application)”

– Each part of the text is summed up with “[core view]” at the beginning, and key data is thicker

– Cases are presented in tables in part and competitions in tables。

4. Attention:

– Uncertain data (e.g. the latest price of the bid) are labelled “to be updated” and are not made up

– Avoid undermining the competition and focus on “How our products work out the pain points for their clients”。

Scenario 2: Data analysis — sales data redisposal report

Template Core Logic: Data Dismantling + Reason Analysis + Strategy Proposal + Target Adjustment。

• Mandate: to generate a sales duplicate report based on [time period] [product] sales data

1. Data background:

– DATA RANGE: [TIME PERIOD] [REGIONAL/TEAM] SALES DATA, CORE INDICATORS: NUMBER OF NEW CLIENTS (TARGET X, ACTUAL X), UNIT RATE (TARGET X%, ACTUAL X%), UNIT PRICE (TARGET X0.000, ACTUAL X0.000), NUMBER OF LOST CUSTOMERS (X HOUSEHOLDS)

- Use: Rewinding [the decision-making role, such as the sales director] for [the next cycle] adjustments to the sales strategy。

2. Analysis logicPrompt wordSerial:

1 DISMANTLING OF DATA: DISMANTLING OF CORE INDICATORS BY “REGIONAL, SALES LEVEL, CUSTOMER INDUSTRY” TO IDENTIFY THE MOST PERFORMING/LOWEST DISAGGREGATED DIMENSIONS (E.G. “35% IN EAST CHINA, 20% IN NORTH CHINA”), PRESENTED IN TABLES

2 GAP ANALYSIS: ANALYSIS OF THE CAUSES IN RELATION TO SPECIFIC CASES FROM THE PERSPECTIVE OF “CUSTOMS DEMAND MATCHING, MARKETING CAPACITY, MARKET COMPETITION, EXTERNAL ENVIRONMENT” (E.G. “CUSTOMS SINGLE RATE 15%, BELOW 35% FOR OLDER PERSONS, DUE TO LACK OF KNOWLEDGE OF PRODUCTS”) FOR THE INDICATOR OF UNDERACHIEVEMENT (E.G. SINGLE RATE)

3 LOSS ANALYSIS: ANALYSIS OF LOST CUSTOMERS BY “INDUSTRY, CAUSE OF LOSS, POSSIBILITY OF RECOVERY”, LABELLING OF “POSSIBLE CUSTOMERS” AND RECOVERY STRATEGIES

4 Strategy recommendations: based on the results of the analysis, concrete land-defeating measures are given in four areas: “resource-sorting, staff training, customer maintenance, market outreach” (e.g. “In October, new-person product training is conducted once a week, with instruction in Top sales”)

5 TARGET ALIGNMENT: RECOMMENDATIONS FOR ADJUSTMENTS TO THE CORE INDICATORS [FOR THE NEXT CYCLE] ARE MADE IN THE CONTEXT OF THE RESULTS OF THE EXERCISE (E.G. “THE NUMBER OF ADDITIONAL CLIENTS IS ADJUSTED TO X, THE UNIT RATIO GOAL X%”)。

3. Output format:

– Markdown table for data dismantling, loss analysis

– Gap analysis, strategy proposal with a “problem point - concrete performance - solution” structure

- At the end, add “the core objectives of the [next cycle] and the timetable for implementation”。

4. Attention:

– Accurate calculation of the data (e.g. unit = unit/business) and clarification of the formula when it is not clear

– It is suggested that the section should clarify “responsibles, time nodes, measures” and avoid generalities。

Scenario 3: Programme Planning - Annual Market Extension Programme

Template core logic: Target dismantling + channel strategy + budget allocation + implementation monitoring + risk response。

Mission: development of [product] annual market extension programme for 2025

1. Operational context:

- PRODUCT POSITIONING: [PRODUCT TYPE, E.G. MES], CORE ADVANTAGES: [E.G. “IMPROVING PRODUCTION EFFICIENCY” “QUALITY RETROACTIVE”), TARGET CLIENTS: [INDUSTRY + SIZE, E.G. “SMALL AND MEDIUM MANUFACTURING ENTERPRISES]

– STATUS 2024: CLIENT X, MARKET OCCUPANCY X%, MAIN ACCESS: [E.G. INDUSTRY FAIR/OLD CUSTOMER RECOMMENDATION], COST X/HOME

– TARGET 2025: NEW CUSTOMERS X (X% FOR [INDUSTRY BREAKDOWN, E.G., MACHINERY/FOOD/MEDICINE]) WITH MARKET SHARE RISING TO X% AND REDUCED COST TO CUSTOMERS X%。

Programme framework:

TARGET 1 DISMANTLING: DISMANTLING THE ANNUAL TARGET BY “QUARTER, INDUSTRY, CHANNEL”, FOR EXAMPLE: Q1 ADDS X (MECHANIC X, FOOD X) AND RECEIVES A COST OF X DOLLARS/HOME

Channel 2 strategy:

– underline channels: a list of key participants in 2025 (e.g., the shanghai industrial fair in march), with a clear budget for each event, targets for the recipients, and design priorities for booth

– Online channels: content marketing (one white paper, two cases per month, five short videos), distribution channels (industry media/micro-publics/diversities), core themes (e.g. “Performance efficiency cases”)

– OLD CLIENT RECOMMENDATION: DESIGN OF A RECOMMENDED INCENTIVE MECHANISM (E.G., A THREE-MONTH FREE SERVICE RECOMMENDED BY AN OLD CLIENT), WITH TARGET RECOMMENDATION RATIO X%

BUDGET ALLOCATION: THE TOTAL BUDGET OF 100,000, ALLOCATED ACCORDING TO CHANNEL (X0.000 BELOW LINE/X0.000 ON LINE/X0.000 RECOMMENDED BY OLDER CLIENTS) + X0.000 EMERGENCY, WITH A QUARTERLY BUDGET SPLIT TABLE

4 Implementation monitoring: development of a monthly watch sheet (core indicators: number of visitors, cost of clients, rate of channel conversion) identifying those responsible (e.g., “publishing of content”)

5 Risk response: Prejudicing possible risks (e.g., poor performance of fairs, changes in industry needs) and providing responses (e.g., increasing on-line access, adjusting programme focus)。

3. Output format:

– PPT OUTLINE FORMAT, EACH PART OF WHICH CONTAINS “TITLE-CORE CONTENT-RESPONSIBLE-TIME NODES”

– Key data presented in tables (budget distribution table, quarterly target dismantling table)

– THE FEASIBILITY OF SUPPORTING STRATEGIES, CITING SUCCESS STORIES FROM 2024 (E.G., “25 AT THE SHANGHAI INDUSTRIAL FAIR IN 2024, AT A COST OF X DOLLARS/HOME”)。

SCENARIO 4: CLIENT SERVICES - PRODUCT FAQ

Template core logic: HF filter + structured answer + common expression + operational guidance。

MISSION: WRITING [PRODUCTS] CLIENT FAQ (FOR [USER ROLE, E.G. HR COMMISSIONER/PRODUCER])]

Background information:

- Target users: [user roles] using [products], non-technical background

– Core requirements: to address HF issues in day-to-day operations and to reduce the number of passenger service consultations

– Style requirements: language is common, steps are clear, and each issue is accompanied by “Operational guidelines” (word describes the key steps of the screenshot)。

2. Content requirements:

1 Question screening: select Top10 HF based on [time] passenger data (e.g., “How do you export monthly reports?” “How to modify employee information?”)

2 Structure of answer: each question is written by "Question - Reason - Solutions (Step) - Attention", Example:

How do you export monthly attendance reports

Question: Need to export attendance data to financial accounting salaries

Reason: System default report format is Excel and manual export is required

Solutions: 1. Log-in system, click on left " time and attendance management " ; 2. Select " monthly statistics " , set statistical months; 3. Click " Export " , select Excel format, to the top right; 4. Wait 30 seconds for download

ATTENTION: 1 REQUIRES CONFIRMATION THAT DATA HAVE BEEN CLEARED PRIOR TO EXPORT; 2 EXPORT FAILURE CAN CHECK THE NETWORK OR CONTACT THE PASSENGER SERVICE (TELEPHONE XXX)

3 Classification: classified as " Functional modules (e.g., attendance management/staff management/report export) " , each module is sorted in question order。

3. Output format:

– Markdown format, with ## and ##

– Steps are listed in an orderly manner, and attention is given to matters beginning with “Note:”

– At the end, add “other problem feedback channels (guest calls/business tweets)”。

4. Attention:

– AVOID TECHNICAL TERMS (E.G. “API INTERFACE” AND “SYSTEM INTERFACE”)

– Different scenarios (e.g. “Staff Separations Active/Preventive”) need to be described separately。

Five pits that Prompt designed

Even if the above methods are available, in practice they may result in the failure of AI output to meet the standards by negligence in detail. Here are six common mistakes and their corresponding solutions to help you avoid pits and enhance Prompt's effectiveness。

Error 1: Overdesigned

Prompt is long and complex, with vague core needs

Performance:For the purpose of writing a PRD, a product manager has added to Prompt the expression “with 10 years of To B product experience, master of PRD writing specifications, familiar with ISO 9001, IEE standards ...”, Prompt has 500 words and the content of AI output is terminological and lacks real demand。

reason:OVERSTRETCHING OF ROLES, NORMS, CONTEXTS, WHICH MAKES IT IMPOSSIBLE FOR AI TO IDENTIFY CORE NEEDS。

Solutions:

• Core requirements ahead:THE BACKGROUND IS SUPPLEMENTED BY A DIRECT INTRODUCTION TO “WRITING XX MODULE PRD”
• Role-setting precisionINSTEAD OF THE WORDS “EXACT IN 10 INDUSTRIES”, REPLACE, FOR EXAMPLE, “PRODUCT MANAGERS FAMILIAR WITH THE NEEDS OF MS IN THE FOOD INDUSTRY”
• As required:ONLY REFERENCES TO MISSION-RELATED NORMS (E.G. PRD NEEDS TO INCLUDE FUNCTIONAL DESCRIPTIONS, INPUT OUTPUT, ABNORMAL HANDLING)。

Error 2: Base is missing

Relying on advanced skills, ignoring core information

Performance:A MARKETING MANAGER USES A “THINK CHAIN” TO WRITE AN OUTREACH PROGRAMME WITH A DETAILED DESIGN OF THE STEPS, BUT DOES NOT STATE “TARGETING CUSTOMERS, OUTREACH OBJECTIVES”, AND THE AI-OUTPUT PROGRAMME HAS A COMPLETE LOGIC BUT DOES NOT FIT THE BUSINESS。

reason:Core instructions are vague and advanced skills do not compensate for the lack of basic information。

Solutions:

• Prompt looks at the three core elements: Are the objectives clear? Is the context clear? Are the constraints specific
• (a) Complex tasks are first “written with basic skills” and then optimized with advanced skills
• If the output deviates, the basic information is supplemented and the advanced skills adjusted。

ERROR 3: UNDEFINED HIDDEN NEEDS, FREE AI

Performance:A sale allowed AI “to write a follow-up e-mail to a potential customer” and did not state that “the customer is the manufacturing owner, who previously communicated with the product demo, focusing on ROI”, and that the mail sent out by AI was generic and did not respond。

reason:IMPLICIT NEEDS (E.G., CUSTOMER FOCUS, COMMUNICATION HISTORY) ARE NOT CLEAR AND AI CANNOT BE ACCURATELY MATCHED。

Solutions:

• Create NeedChecklist:Ensuring that customer roles, trades, pain points, communication history are included
• We're going to use the "counter-examination" method:IF YOU WERE AI, WHAT WERE THE QUESTIONS? FOCUS ON WHAT
• Identify hidden needs:For example, clients are concerned about long implementation cycles and need to emphasize “three months of implementation” in their mail。

Error 4: Skills stacking

If you need it or not, use all of your advanced skills

Performance:A pre-sale consultant writes a simple client case, using the term “think chain, task split, role set”, which takes one hour, and the result is not much different from the " Example Guide."。

reason:Skills choices are less efficient than “demand-based matching” over-complex approaches。

Solutions:

• Selecting skills by complexity of need:Simple tasks (e.g., case writing, data collation) use basic techniques; complex tasks (e.g., white papers, competition analysis) use progressive strategies
• Guided by the “minimum cost principle”:It can be solved with one technique, not many。

Error 5: Neglecting ethics

Prompt contains sensitive information or inappropriate guidance

Performance:A client service allows AI to analyse client feedback, and Prompt directly carries sensitive information such as a client ' s clear name, telephone, address, and risks privacy disclosure。

reason:Failure to consider data privacy and ethical norms leads to compliance risks。

Solutions:

• Data desensitization:REPLACE PERSONAL INFORMATION WITH ANONYMOUS ID (E.G. CLIENT 001)
• Follow the data minimization principle:(b) Provide only the minimum information necessary to fulfil its mandate
• Avoid sensitive orientation:Prompt, for example, does not design “how to devalue competition”, “how to mislead customers”。

Some reflections on enterprise-level Prompt capacity-building

For enterprises, Prompt capabilities are not only personal skills, but also the core competitiveness of organizational efficiency, and many of them now go to AI “family buckets” to enhance the productivity of their internal staff, so that the scale-up and efficient application of AI tools can be achieved by creating a systematic organizational empowerment system that allows team members to quickly master the Prompt approach。

Create'Prompt Template Library

/ Lower use threshold

• Classification dimensions:Classification by business scene (content/analysis/programme/service), product line, output format
• Template content:Each template contains "Prompt originals, use scenes, optimization techniques, impact cases"
• Maintenance mechanisms:The appointment of a dedicated person to update on a monthly basis to collect quality team cases to be added to the library
• Tool Recommendation:A NUMBER OF ONLINE FILES AND KNOWLEDGE MANAGEMENT TOOLS ARE STORED TO SUPPORT KEYWORD SEARCHES (E.G., THE ERP CLIENT CASE)。

Conduct of a training competition on “Prompt dismantling”

/ Upgrade structured thinking

• Question:Selecting complex tasks in the near future (e.g., the Market Extension Programme 2025)
• Dismantling:Grouping to decomposition tasks using “task split” and design Prompt
• Practice:Each group produces AI output using Prompt
• Rewind:Comparing the outputs of the groups and summarizing the commonality of excellent Prompt
• Frequency:Quarterly 1-2 meetings to avoid theoreticalization in the context of actual tasks。

Establishment of a feedback optimization mechanism

/continuously increasing effect

Feedback dimensions:(a) Designing scorecards, which rate "accompany demand, format accuracy, data credibility, landing"

Feedback process:

• Fill in the score sheet after the use of Prompt by members to indicate problem points
• Weekly wrap-up meetings to share high-scoring and problem cases
• DEVELOP GENERIC SOLUTIONS FOR HIGH-FREQUENCY (HF) PROBLEMS (E.G. AI-GENERATED DATA)
• Incentive mechanisms: scoring performance or learning resources for members who contribute high-quality cases and make valid recommendations。