Research Reports: From Data Collection to a Visualized Report in One Delivery
💡 One-sentence summary: Use WorkBuddy to build a research report pipeline—first collect data into a structured table, then generate a visualized report that management can read directly.
Searching for "AI research report" mostly turns up two kinds of results: lengthy methodology articles teaching you to "pick a topic, search the literature, and draft an outline," and online tools where you "enter a topic and generate a 10,000-word report with one click." After reading the former, you still don't know where the data comes from; the numbers generated by the latter have no sources, so you don't feel confident handing them to your boss. This lesson walks through a complete deep research delivery workflow: with one sentence, have WorkBuddy collect market information from the past 1 year and organize it into structured data, then, after verification, generate a visualized analysis report based on that data—collection, analysis, charting, and layout all done in one task.
First, let's be clear about the factual boundaries: the two-step instructions in this lesson (first collect and organize structured data, then generate a visualized report for management) and the three usage tips—"clarify the analysis question first," "avoid making too many demands at once," and "state data credibility requirements"—come from the official practice case "Practice 3: Data Analysis and Visualization"; the self-driven approach of "state the goal clearly, allow autonomous decomposition, and self-check before delivery" comes from the official practice case "Practice 9: AI Self-Driven." The official homepage lists "Research · Deep Investigation" as one of its four major scenarios, with the tagline "From gathering information to delivering conclusions, a report in 15 minutes"—this is marketing language, not a guaranteed timeframe; actual time depends on the scope of the research.
In the previous lesson, B2 Excel Data Analysis, we analyzed tables you already had in hand; in this lesson, the data doesn't exist yet—let WorkBuddy find it. The official example checklist for creating task documents even includes an item: "Research AI industry trends for 2024 and write a report"—one sentence can start a task, but to deliver a report you'd dare show your boss, you need to follow the two steps in this lesson. This is lesson 3 of Route 2 in "Scenario Practice," based on WorkBuddy v5.9.3.
- How credits are deducted and how much quota the trial version has: 2.6 Credits and Pricing
- The full playbook for follow-ups, rollback, and citations: 2.1 Conversation Advanced
What You'll Be Able to Do After This Lesson
- Have WorkBuddy collect market information from the past 1 year with one sentence and organize it into structured data (the official example wording can be copied directly)
- Generate a visualized analysis report based on verified data: core conclusions, key data charts, trend judgments, and action recommendations all in one output
- Have WorkBuddy self-check before output: complete structure, no duplicate information, concise and clear conclusions (official self-check instruction wording)
- State four things clearly at once in your report requirements: report format, section structure, chart preferences, and target audience
Why Use WorkBuddy Instead of Web-Based AI?
| Capability | Web-Based AI | WorkBuddy |
|---|---|---|
| Finding data | ❌ Answers based on model memory; numbers and timeliness can't be verified | ✅ Describe the topic, scope, and source preferences, then organize the collected information into structured data (official practice case) |
| Producing the report | ❌ Returns a long block of text; you handle charts and layout yourself | ✅ Collection, analysis, charting, and report layout completed in one task, outputting a complete result file (official practice case) |
| Data sources | ❌ Can't provide links when asked about sources, and the ones given may not even exist | ✅ You can specify requirements for "public sources, source links, and time ranges" in advance (official usage advice) |
| Quality control | ❌ Before handing it in, you can only read it through yourself | ✅ You can require a self-check before output, confirming complete structure and no duplicate information (official self-check instruction) |
| Round-by-round revision | ❌ Once the context is broken, you have to restate the requirements all over again | ✅ Produce a basic version first, then supplement charts and conclusions round by round (official usage advice) |
Your Current Struggles
- You get a research assignment out of the blue: "Give me an industry analysis report by next Monday"—no idea where the data is or where to start searching
- You open ten tabs in a search engine, copy and paste into a document, with conflicting definitions and messy formatting—and two days later you can't find the sources anymore
- You have AI write the research report, but the numbers come out of thin air with no sources—they look real, and you'd write them but wouldn't dare submit them
- The data is finally complete, but you still have to do the charts, layout, and conclusions yourself, burning an entire afternoon on beautification
When to Use This Technique
When you need to:
- Produce a research report from scratch: market research, business reviews, operations analysis, project reporting (applicable scenarios named in the official practice case)
- Data is scattered across the web and multiple sources, and hasn't been organized into a file yet
- The output is meant for people to read: with conclusions, charts, and recommendations that management can pick up and read directly
And you don't want to:
- Open dozens of tabs and copy and paste manually
- Write a report using numbers with no sources
- Learn chart creation and report layout yourself
🎒 Preparation Before Starting
- Completed 1.4 Your First Task, know how to create a task and where to find deliverables in the results area
- Think through the questions the research should answer: what topic, what time range, and who the report is for
- Have credits in your balance: all conversations consume credits, and research is a long multi-round task—see the rules in 2.6 Credits and Pricing
- 15 minutes
Follow Along With Me
Step 1: Collect information with one sentence and organize it into structured data
Why: The most time-consuming part of a research report isn't writing—it's looking things up. When the data hasn't been organized into a file yet, the official practice case approach is to directly describe the data topic, scope, and source preferences, letting WorkBuddy first gather the scattered information into structured data.
How to do it: Create a new task and send the official example instruction verbatim (replace the topic, scope, and focus areas with yours):
Please collect market information on domestic AI office tools from the past 1 year, focusing on user scale, application scenarios, and representative products, and organize it into structured data.Breaking this sentence down: topic = domestic AI office tool market information; scope = the past 1 year; focus areas = user scale, application scenarios, representative products; output = structured data.
You should see: WorkBuddy start gathering information, with the execution process scrolling in the conversation area, and finally deliver a well-organized set of structured data, usually in table form—not a long block of prose.
- What you get is structured data, not a block of text
💡 State credibility requirements in advance
The official usage advice, quoted verbatim: "State data credibility requirements: if you need public sources, source links, or a time range, it's recommended to specify them in advance." Add a sentence at the end of your instruction such as "Data must come from public sources, with a source link for each entry," which makes later verification and citation much easier.
Step 2: Verify data sources and mark the gaps
Why: The credibility ceiling of the report is the credibility of this data. The information the AI gathers may contain duplicates, be outdated, or use inconsistent definitions—spot-check first, and don't let problematic data slip into the report.
What to do: have it lay out the data for you to review:
List all the data you have gathered in full, with a source link and time for each entry, so I can verify it.Check three things: whether the source links open, whether the times fall within the range you want, and whether figures for the same metric from different sources conflict. If you find a gap, add a sentence like "Information on XX is missing; please gather more."
You should see: a complete data list with source links. Spot-check 2-3 entries and if they check out, this data can serve as the foundation of the report.
- Spot-checked 2-3 entries: sources open, time ranges match
Step 3: Spell out four things and generate a visualized report
Why: With the data verified, it's time for the report. The official practice examples recommend stating four things at once: report format, section structure, chart preferences, and the audience—saying clearly who it's for matters more than how many words to write.
What to do: the official example instruction, quoted verbatim (replace the audience and sections with your own):
Please generate a visualized analysis report based on the gathered data, including core conclusions, key data charts, trend judgments, and action recommendations, suitable for management reading.You should see: WorkBuddy complete the analysis, charting, and report layout, and output a complete result file—core conclusions first, charts embedded in the corresponding places, trend judgments and action recommendations at the end, with the report file visible in the result area.
💡 This is the unique value of WorkBuddy
A web-based AI writing a research report gives you back a long block of text: you look up the data yourself, draw the charts yourself, do the layout yourself. WorkBuddy puts gathering, analysis, charting, and report layout into a single task and directly delivers a complete result file—the official site's copy for the "Research · Deep Research" scenario is "From gathering information to delivering conclusions, a report in 15 minutes." The marketing claim is for reference only; "delivering a complete file in one go" is its real differentiator.
- The report file has been generated, with all four parts present: conclusions, charts, trend judgments, action recommendations
Step 4: Have the AI self-check before delivery
Why: A report destined for management deserves an extra quality check. The official practice example "AI self-driven" advises: when task quality requirements are high, explicitly request a review so it goes over its own output before delivering.
What to do: the official self-check instruction, quoted verbatim:
Please run a self-check before output, confirming the structure is complete, there is no duplicated information, and the conclusions are concise and clear.Add more check items if needed, for example "Every key data point is annotated with its source."
You should see: WorkBuddy run a self-review, remove duplicate entries and redundant wording, then deliver the revised report.
- After the self-check there are no duplicate entries, and the conclusions section is concise and clear
Step 5: Refine round by round until the AI research report is ready to hand in
Why: The first version of a report is rarely final. The official usage advice: "Avoid making too many requests at once: it's recommended to produce a basic version first, then add charts and conclusions round by round." The context is still there—just keep talking.
What to do: follow up as needed, for example:
Add a trend chart to the user-scale sectionExpand the action recommendations to 3 items, each annotated with the data it's based onYou should see: the report is updated on top of the original, with charts and conclusions filled in round by round, without needing to re-describe the research background. For the full playbook on follow-ups (sorting, rollback, citation), see 2.1 Conversation Advanced.
- After the follow-up the context isn't lost, and the report updates accordingly
📋 Magic Prompt
Turn the two-step method into a copyable template: step one asks for data, step two asks for the report, plus a self-driven one-shot version. Placeholders are marked with【】; replace them with your own and it's ready to run.
📝 Step 1: Gather data and structure it
Expected result: gather and structure in one sentence—the template version of the official practice example.
## Role
You are an industry researcher, skilled at gathering and organizing market data from public information.
## Task
Please gather market information related to【the past 1 year(s)】【domestic AI office tools】, focusing on【user scale, application scenarios, and representative products】, and organize it into structured data.
## Input information
### Required fields
- Research topic:【e.g., the domestic AI office tool market】
- Time range:【e.g., the past 1 year(s)】
- Metrics of interest:【e.g., user scale, application scenarios, representative products】
### Optional fields
- Source preferences:【e.g., prioritize industry associations and official reports?】
- Exclusions:【e.g., excluding overseas markets?】
## Output specification
- Structured data table: one entry per row, with metric, value, time, and source link
- Entries that cannot be gathered are explicitly marked "not found"
## Constraints
- ✅ Every data point includes a public source link
- ✅ Only use information within the specified time range
- ❌ Do not use figures without sources
- ❌ Do not fabricate missing data📝 Step 2: Visualized report for management
Expected result: a report based on verified data—the template version of the official practice example; just fill in the four things and it's ready to use.
## Role
You are a business analyst, skilled at turning research data into conclusions management can use directly.
## Task
Please generate a visualized analysis report based on the gathered data, including【core conclusions, key data charts, trend judgments, and action recommendations】, suitable for【management】reading.
## Input information
### Required fields
- Audience:【e.g., management / department heads】
- Core sections:【e.g., core conclusions, key data charts, trend judgments, action recommendations】
### Optional fields
- Report format:【e.g., Word / PDF?】
- Chart preferences:【e.g., line charts for trends, bar charts for comparisons?】
- Length limit:【e.g., main text no more than 5 pages?】
## Output specification
- A complete report file, with charts embedded in the corresponding sections
- Core conclusions first, supporting data attached after
- Every key data point annotated with its source
## Constraints
- ✅ Conclusions based only on the gathered data
- ✅ Action recommendations are actionable and no more than 3 items
- ❌ Do not present speculation beyond the data as conclusions
- ❌ Do not pile on background material irrelevant to the topic📝 Self-driven one-shot version
Expected result: use this when you don't want to split it into two steps and monitor the process—the recommended approach from the official practice example "AI self-driven": state the target outcome, constraints, output format, and completion criteria all at once, and let it break down the steps itself and deliver after a self-check.
## Role
You are an industry researcher, capable of autonomously completing the full research workflow from gathering to delivery.
## Task
Please complete this task autonomously: gather information on【the past 1 year(s)】【the domestic AI office tool market】, organize it into a research report including core conclusions, key data charts, trend judgments, and action recommendations, attach source links, and check for duplicated information yourself before output.
## Input information
### Required fields
- Research topic and scope:【topic + time range】
- Completion criteria:【e.g., complete structure, no duplicated information, concise and clear conclusions】
### Optional fields
- Audience:【e.g., management?】
- Report format:【e.g., Word?】
## Output specification
- Structured data first, then the complete report file
- Source links attached throughout
## Constraints
- ✅ Self-check before output: complete structure, no duplicated information, concise and clear conclusions
- ✅ Information gaps are honestly marked
- ❌ Do not fabricate data or sources
- ❌ Do not skip the self-check and deliver directlyCheckpoints ✅
- Got structured data in one sentence, not a block of prose
- Spot-checked 2-3 data entries: sources open, time ranges match
- The report's four things are spelled out: format, section structure, chart preferences, audience
- The report has been generated, with core conclusions, charts, trend judgments, and action recommendations all present
- The pre-output self-check has been run: no duplicated information, concise conclusions
Pitfall warnings
❌ Asking for data and the report all at once
Throwing over a single sentence like "research the XX market and write me a report"—gathering, verification, and report writing crammed into one round; if a number is wrong you can't tell which step it went wrong in, and you have to redo everything from scratch.
The official usage advice, quoted verbatim: "Avoid making too many requests at once: it's recommended to produce a basic version first, then add charts and conclusions round by round."
The right approach: two steps—ask for structured data first (Step 1), and only ask for the report after verification checks out (Step 3).
❌ Not specifying data credibility requirements
Without source requirements written down, the AI may pad the answer with stale numbers from its model memory—the figures look real, but the sources cannot be found anywhere.
Official usage advice: "Specify data credibility requirements: if you need public sources, source links, or a time range, state them in advance."
Correct approach: write it firmly into the instruction—"public sources, each item with a source link, time range within the last 1 year(s)"—so the spot-check in step 2 has a standard to check against.
❌ Submitting the report as soon as you get it
AI self-checks cannot replace your own review: broken source links, two sources with conflicting numbers, conclusions that go beyond what the data supports—these need a human eye.
The official practice case recommends double insurance: the AI self-checks structure and repetition before output, and you spot-check sources and numbers.
Correct approach: spot-check the data in step 2 + read through the conclusions before delivery, and click open the sources to verify key numbers.
❌ Asking time-sensitive questions without going online
Asking for "the latest 2026" market data and relying only on model memory to answer—the information is outdated yet delivered with full confidence. The official Web Search skill documentation says verbatim: "When a question involves time-sensitive information such as recent, latest, or current, relying solely on model memory is often not enough."
Correct approach: install the web-search skill (installation name web-search, from SkillHub, one of the official Top 10 zero-cost skill recommendations), and ask with a time range (e.g., "the last 1 month(s)"). See D2 Selected Zero-Cost Skills.
💡 How many credits will it cost
Every conversation consumes credits. The officials have not given specific consumption figures for research tasks; gathering plus report writing is a long multi-turn task that costs much more than a single-turn Q&A, so use the 500 base credits per month in the trial version sparingly. See 2.6 Credits and Pricing for the rules.
Advanced Tips (Optional)
- For big topics, review the breakdown plan first: when the research scope is especially large (e.g., a full industry analysis), switch to Plan mode and have it produce an execution plan first; confirm the breakdown is reasonable before running, to prevent the direction from going off track. See 2.4 Three Execution Modes.
- When the report needs to be presented on stage: drag the generated report file into a new task and have WorkBuddy convert it into a slide deck; see A4 One-Sentence PPT Generation for the workflow.
- Hand recurring research to scheduled tasks: set up weekly or monthly industry briefings once and have them run automatically on schedule; see 3.3 Scheduled Tasks; for the daily morning news briefing playbook, see C4 Daily News Briefing.
Lesson Summary
In this lesson you completed the full workflow of using WorkBuddy to produce an AI research report:
- ✅ Two-step method: first gather data into a structured table with one sentence, verify it, then generate a visualized report
- ✅ Four things about the report: format, section structure, chart preferences, and audience—state them all at once
- ✅ Double-inspection quality check: AI self-check before output + your spot-check of sources and numbers
- ✅ Round-by-round additions: basic version first, then charts and conclusions
| Key Points | Framing |
|---|---|
| Core method | Two-step instruction: gather and organize structured data → generate a management-ready visualized report (official practice case) |
| Official example | Gather market information on domestic AI office tools from the last 1 year(s) / core conclusions + charts + trend judgments + action recommendations |
| Credibility requirements | Public sources, source links, and time range stated in advance (official usage advice) |
| Self-check instruction | Please self-check before output, confirming the structure is complete, information is not duplicated, and conclusions are concise and clear (official practice case) |
| Official website copy | "From researching the material to delivering conclusions, a report in 15 minutes"—marketing framing, not a committed timeframe |
Next Lesson Preview
The report is finished on your computer, but you're not at your desk—what now? The next lesson connects WorkBuddy to WeCom—C1 WeCom Remote: Connecting WorkBuddy to a WeCom Bot.
You will learn:
- How to create a WeCom smart bot: two paths for admins and regular members
- WebSocket persistent connection binding: Bot ID + Secret, with QR-code quick binding also supported
- How to push scheduled task results to WeCom (persistent connection only)