About this article
This article was created using an automated generation workflow powered by generative AI. We reviewed official Microsoft support and update information as of October 27, 2026, and compiled concrete examples and precautions to help beginners use the tool without confusion.Actual login and screen operations for Copilot are not performed.

Verification Status: 📘 Microsoft official specifications confirmed – Copilot physical device and WordPress display unverified

Supported Environments, Capabilities, and Limitations (Confirmed as of October 27, 2026)

10 Practical Exercises to Make Copilot Chat Usable for Work: A Learning Course from Prompting to Deliverables, Verification, and Approval

Reading operational UI descriptions alone leaves users uncertain when actually starting work with Copilot, because real workflows continuously involve formulating prompts, reading source documents, checking dates and numbers, and reviewing before sharing. Therefore, we will sequentially execute ten exercises assuming the planning of a fictional "Citizen's Reading Group." Most exercises can be tried using the free chat, while alternatives are provided for scenarios requiring specific plans such as images or Pages.

The readers of this course are individuals who want to self-grade their independent study results and supervisors who need to verify participant proficiency levels. Rather than simply reading through feature introductions, the objective is to score using the same fictional data on whether participants could "detect conditional mismatches," "return to the source evidence to make corrections," and "avoid treating unexecuted operations as completed."

Open a new chat at copilot.com for personal accounts or the permitted Copilot Chat for work and school, paste the fictional conditions from the end of the body text, and then send the prompt for each exercise. Even if the model or UI differs per participant, the expected conditions below remain the same. If Pages or attachments are not displayed, grading can be continued by pasting the text and saving it to your own notes.

Scoring Method Proposed by the Editor

Record each exercise with 0 to 2 points. A score of 0 indicates overlooking mismatches with the input; 1 indicates finding a mismatch or unverified item but failing to document the reason for correction or withholding; 2 indicates being able to explain adoption or withholding based on evidence from the original text, numerical values, or screen. Although the 10 exercises total up to 20 points, this does not represent Microsoft certification or an official exam score. Do not grant submission permission based on the total score alone; if there are any confidential inputs or unapproved external operations, stop the practice and investigate the cause. Record unexecuted exercises as "Not Performed" rather than mixing them with a score of 0.

What problems the feature is designed to solve

The practical example covered in this article is a fictional study and reading group for citizens. Practice is conducted under the conditions of a capacity of 10 participants, a duration of 60 minutes, and undecided date, time, venue, and cost. Even if Copilot's response sounds plausible,verifying whether undecided conditions have been arbitrarily decidedis the passing criterion.

Treating Copilot as a "feature that makes work decisions before humans do" makes failure more likely. Instead, treat what was input, what was suggested, and what was actually executed as separate elements. The procedures in this article are intended for readers to test on their own screens, and do not represent successful results obtained by the author experimenting on a target account.

Figure 1: Decision Flow to Prevent Confusion

flowchart TD
 A["1 条件を整理"] --> B["2 案を広げる"]
 B --> C["3 比較する"]
 C --> D["4 事実を調べる"]
 D --> E["5 文書から要約"]
 E --> F["6 案内文を書く"]
 F --> G["7 数字を検算"]
 G --> H["8 誤りを修正"]
 H --> I["9 保存先を決める"]
 I --> J["10 公開承認を判断"]

How to read the diagram: This is a conceptual diagram to help beginners safely determine the next operation, rather than Microsoft's internal implementation. Actual button names mayCopilot app update notificationPlease verify using resources such as.

1. Exercise 1: Organizing Conditions

The single phrase "book club" does not convey the target audience or duration. First, input hypothetical conditions: a capacity of 10 people, a required duration of 60 minutes, and undecided date, venue, and fee, thereby separating confirmed and pending items. If Copilot arbitrarily decides dates or prices, it is considered a failure.

Specific operations to try

Open a new chat and provide the conditions. Verify whether the meaning of "undecided" is retained.

Prompt that can be entered as-is

架空の市民向け読書会を企画します。定員10名、60分、日時・会場・費用は未定。確定した条件と未決事項を表にしてください。

What constitutes a successful outcome

Check whether two confirmed items and three pending items are separated, and whether placeholder amounts are omitted from blank fields.

Changing just one condition

Copy the conditions to a new practice chat and change only the fee to "500 yen (provisional practice value)." Verify that the date and venue statuses remain undecided without unauthorized confirmation, and do not modify the undecided conditions in the master document.

Decision-making when encountering issues

If the capacity, time, and fee conditions do not match the input, do not adopt the result and return to the original condition table for correction.

Primary Source (Exercise 1: Organizing Conditions)：Microsoft Official: Target Features

2. Exercise 2: Generating 10 Ideas

Instead of jumping to the first proposal, freely generate candidates. At the generation stage, precise cost estimation is not required, but ensure that the candidates are not merely variations in phrasing. For book club formats, ideas can be expanded to include self-introductions, theme selection, short readings, and question cards.

Specific operations to try

In the same chat, convey the objective and participants, and have it generate 10 ideas along with a brief objective for each.

Prompt that can be entered as-is

初心者10名の読書会で参加しやすくなる進め方を10案作ってください。似た案はまとめ、必要な準備も簡潔に示してください。

What constitutes a successful outcome

Are the 10 proposals independent, and is this not too difficult for first-time participants?

Try changing only one condition

Merge two similar candidates out of the 10, supplement with an alternative proposal, and identify issues where the count is merely met by paraphrasing.

Decision-making when stuck

If there is overlapping content among the 10 proposals, do not assume the count has been met; recount only the independent options.

Primary source (Exercise 2: Generate 10 ideas)：Official Microsoft Documentation: Target Feature

3. Exercise 3: Compare options

Even if there are ten promising proposals, there is only 60 minutes. Therefore, compare them based on participant burden, preparation costs, required equipment, and facilitation difficulty. If conditions are unknown, leave them marked as 'requires confirmation'. Even if the AI outputs a total score, do not treat an unsubstantiated ranking as fact.

Specific operations to try

Select 3 proposals from the 10 and tabulate them using the same comparison axes. Leave necessary items unverified.

Prompt that can be entered directly

候補を3案に絞り、準備時間・参加しやすさ・費用・リスクを比較表にしてください。データが不明な項目は推測しないでください。

Criteria for passing

Is there a reason for blank spaces in the comparison items, and are the decision axes of the selected proposals clear?

Try changing only one condition

Change just one evaluation axis from cost-oriented to participant-burden-oriented, and check whether the comparison conclusion changes without filling in unverified conditions.

Decision-making when stuck

If arbitrary figures are generated when costs or equipment counts are unverified, mark that evaluation row as requiring confirmation and review the comparison results.

Primary source (Exercise 3: Compare options)：Official Microsoft Documentation: Target Feature

4. Exercise 4: Checking Sources

Facts such as the actual opening hours of a library are not guaranteed by Copilot's general memory. For a hypothetical planning exercise, avoid using real facility names and instead replace them with practice checking Microsoft's official operation specifications. By separating the publication date, update date, and verification date into different columns, you will learn the differences between news articles and current specifications.

Specific operations to try

Ask Copilot to verify Copilot's file attachment limit using only official Microsoft documentation, and then actually open and cross-reference the sources.

Ready-to-use prompt

Microsoft Copilotのファイル添付について、1会話の件数と1件の容量をMicrosoft公式だけで調べ、一次URLと確認日を示してください。

What constitutes success

Check whether the figures of 20 items and 50MB are documented for the official product in question.

Try changing just one condition

Limit the sources exclusively to official Microsoft documents, and verify the correspondence between the upper limit description in the reference text and the links, rather than relying on a summary of search results.

Troubleshooting decisions

If a link does not open or describes the limits of a different product, consider the source insufficient and do not transcribe the correct numbers into the article.

Primary source (Exercise 4: Checking Sources)：Microsoft Official: Target Feature

5. Exercise 5: Summarizing a Hypothetical Memo

Copy and paste a short text you wrote yourself and have Copilot summarize it. If your subscription allows attaching separate files, you may attach it as a TXT file. Ensure the text includes 'Determine on the day in case of rain' and verify that this condition is not omitted from the summary. Any suggestions outside the document should be requested separately.

Specific operations to try

Input the hypothetical memo 'Capacity 10 people, 60 minutes, cost undecided, rain decision on the day' and separate it into decided items and conditional items.

Ready-to-use prompt

次の架空メモを要約してください。『定員10名、時間60分、参加費未定、雨天は当日判断』。確定、条件付き、未決を分け、書かれていないことを補わないでください。

What constitutes success

Verify that the rain condition is not omitted and that the cost is not incorrectly stated as free.

Change only one condition

Change one item in the fictitious memo from 'Decision on rain is made on the day' to 'Canceled in case of rain', and compare whether the conditional field in the summary updates correctly.

Troubleshooting judgment

If cancellation in case of rain and decision on the day are mixed, the meaning of the condition has changed. Do not use the summary as an official notice.

Primary Source (Exercise 5: Summarize a Fictitious Memo)：Microsoft Official: Target Features

6. Exercise 6: Draft an Information Email

Draft only decided items into the text, and do not omit unresolved items. Pay attention to whether the AI generates recipients, contact information, or reservation links. Generating an email body in the chat is different from actually sending it.

Specific operations to test

Generate an information message under the same conditions. Do not send it, and verify whether the reader can understand what is undecided.

Prompt that can be entered as-is

架空の読書会の案内メールを作ってください。定員10名、所要60分、日時・場所・費用は未定。丁寧な文面で、送信はしないでください。

Criteria for success

Ensure undecided items do not look like confirmed information and that it does not state the email has been sent.

Change only one condition

Change the recipient from general participants to management staff, and check whether the level of politeness and required items change. Neither will be sent.

Troubleshooting judgment

Even if the response contains the phrase 'already sent', it is not evidence that the email was actually sent. No sending operation is performed in this exercise.

Primary Source (Exercise 6: Draft an Information Email)：Microsoft Official: Target Features

7. Exercise 7: Verify Time Allocation

When you have AI generate a 60-minute program, individual items may look natural while the total adds up to 65 minutes. Get into the habit of performing your own addition before trusting numerical tables. Factoring in preparation, transition, and cleanup activities helps make the schedule realistic.

Specific steps for testing

Test segments such as 5, 10, 30, 10, and 5 minutes as candidates, and calculate whether the total equals 60 minutes.

Prompt that can be entered as-is

読書会60分の進行表を、開会・自己紹介・読書・意見交換・閉会で作ってください。各項目を分で示し合計60分にしてください。

Criteria for success

Verify manually whether the total is 60 minutes and whether adequate time is allocated for closing.

Try changing a single condition

Change "10 minutes of discussion" in the schedule to "15 minutes of discussion," and determine which sections should be adjusted to maintain the 60-minute total.

Troubleshooting guidelines

If the total comes to 65 minutes, the result is incorrect. Do not trust the AI's "total 60 minutes" display; sum up the values yourself.

Primary Reference (Exercise 7: Verifying Time Allocation)：Microsoft Official: Target Features

8. Exercise 8: Fixing a Single Error

When an error is intentionally found, rewriting the entire text introduces new errors. Specifying the error location, the correct value, and the items that must remain unchanged makes it easier to review differences.

Specific steps for testing

If it says "Participation fee: 0 yen," revert only that part to "To be determined." Compare whether other conditions remain unchanged.

Prompt that can be entered as-is

前の回答の『参加費0円』は根拠がありません。正しくは『未定』です。その箇所だけ修正し、変更点を明示してください。ほかの条件は変えないでください。

Criteria for success

Check whether corrections are limited to the specific area and whether dates, times, or venues have been arbitrarily finalized.

Try changing a single condition

Limit the error to a single cost location and verify from the diff before and after the correction that the capacity of 10 attendees and the 60-minute duration are maintained.

Troubleshooting decisions

If correcting the cost also changes the venue or capacity, recheck to revert differences outside the requested scope.

Primary Source (Exercise 8: Fix a single error)：Microsoft Official: Target Feature

9. Exercise 9: Preserve deliverables

If the created proposal cannot be found later, you will have to create it from scratch again. In addition to the chat history, if Pages is available, move the content there and verify the save destination. If Pages is not available, you may copy it to a document you manage. Understand the difference between the chat response and the saved document.

Specific actions to try

Decide on the official notice to retain, and try editing it in Pages if available. If not available, copy it to a local note.

Prompt that can be entered as-is

この架空の企画について、確定事項だけを転記するための一覧を作ってください。保存先を勝手に決めず、私が指定するまで外部ファイルを変更しないでください。

Criteria for success

Whether you can explain which draft is the official version and whether the original undecided items remain.

Try changing only one condition

Compare moving the deliverable to Pages versus saving it to a locally managed text document, and determine a single location for the official version.

Troubleshooting decisions

Even if Copilot replies that it has saved, do not record it as successfully saved unless the actual file or page can be opened.

Primary Source (Exercise 9: Preserve deliverables)：Microsoft Official: Target Feature

10. Exercise 10: Approve before publishing and sending

The final exercise teaches human responsibility rather than operating Copilot. Determine whether the content is correct, whether there are factual sources, whether the information is safe to show to others, and who approves it. Even if the AI's response is well-crafted, do not mark a yet-unapproved proposal as 'published'.

Specific operations to try

Create a dummy approval checklist. Evaluate the system assuming all fields include evidence, without actually sending emails or sharing externally.

Prompt that can be entered directly

この架空の読書会案内を公開する前のチェックリストを、内容・日時・費用・宛先・個人情報・承認者・保存先で作ってください。公開や送信は行わないでください。

Conditions for a successful outcome

Verify whether fields lacking confirmation are treated as passed.

Try changing only a single condition

Add a condition where the approver is undetermined, and verify that external transmission cannot be performed while the date, cost, and publication destination remain undetermined.

Decision-making when issues arise

If necessary approvers or publication dates are undecided, distribution is placed on hold even if the text is complete. AI quality and publication eligibility are separate matters.

Primary source material (Exercise 10: Approval prior to publication and transmission)：Microsoft Official: Target features

How to manage this in actual operations

When introducing new features to work, avoid rolling them out to everyone all at once simply because they are convenient. Test with a small amount of dummy data, and record the current account, contract, displayed menus, and actually saved results. If you are standardizing tasks, clearly define who enters data, who verifies it, and which document is designated as the official version.

In actual work, it is also important to accurately document what could not be done rather than just what was successfully done. Examples include being unable to use a feature due to lack of a contract, missing buttons, unknown source data locations, or a lack of permission for information management. In such cases, it is necessary to leave a record of the reason for non-action and make the decision to revert to standard documents or manual work. Maintaining a stance of not making AI processing the objective itself leads to secure operations.

Simplified evaluation table for practical implementation

Common beginner mistakes

Q. If Copilot says "Saved," has the file actually been saved?

A. Text confirmation alone is not enough. Navigate to the actual save location and verify that the file exists and that the title and content are correct.

Q. I do not see the same button on my screen.

A. Check personal and work accounts, app updates, previews, agreements, and administrator settings. Feature names may be renamed or in the process of rolling out. Prioritize current official Microsoft documentation over guessing button locations.

Q. Between the free and paid versions, is the paid version always completely accurate?

A. Even if there are differences in available features, usage limits, and models, fact-checking and approval are necessary. Unsupported dates and amounts, as well as incorrect sharing destinations, must be manually reviewed even with a paid subscription.

Q. Can AI-generated text be used directly as an official company record?

A. Automatically generated content is a suggestion. Establish your company's storage locations, reviewers, permissions, and source documents, and obtain the necessary approvals before adopting it as an official record.

Key points of this article

The most important thing, rather than memorizing many feature names, is to be able to explain:The intended purpose, the information handled, the necessary permissions, and the expected results.Regardless of any new feature and its convenient-sounding name, do not grant production permissions without first testing it on a small scale. This approach applies equally to beginners and experts.

Document pass/fail results and failure examples to prove that the 10 exercises were actually practiced

Copying 10 prompts does not mean completing the 10 exercises. The purpose of practice is not just to submit questions to Copilot, but to be able to adopt results based on your own judgment. Therefore, for all 10 exercises, handle the following on a single verification sheet:Input conditions, expected specific results, and corrections upon failure.This is not actual measurement data obtained from Copilot during article writing, but an evaluation example designed by the editor so that readers can test it themselves.

Input conditions standardized with a fictional event

【架空の市民向け読書会】
定員：10名
時間：60分
対象：初心者
日時：未定
会場：未定
参加費：未定
雨天時：当日判断
※これは練習用データであり、実在の催しではありません。

Copy this block into your notes before starting the first exercise. This is so you can verify which input was the authoritative source if different values, such as a capacity limit of 12 people or free participation, appear in subsequent chats. Even if Copilot adds that "it is usually free," do not assume the confirmed conditions have changed.

Evaluation table for the 10 exercises

Especially useful is Exercise 7: Numerical Verification

When having Copilot create a schedule, verify the sum of the numbers yourself. For a 60-minute hypothetical schedule, also verify whether each activity is independent as follows.

The total of 60 minutes is an example scenario where a human manually summed the table.Scenario exampleThis does not mean Copilot itself actually outputted this result. If the initial output is 65 minutes, you should ask it to adjust so that the total is 60 minutes, andrecalculate it manually even after the revision. Leaving it at 65 minutes and accepting it as "roughly correct" trains you to overlook operational errors.

Exercise 8: Why is it necessary to correct only one place?

For example, if an announcement states "Admission free," but the original confirmed conditions state "Admission fee to be determined." If you give the instruction to "redo everything," even unaffected parts will be rewritten in different phrasing, making it difficult to review the modified sections later. Limiting changes as follows makes verification easier.

修正対象：案内文にある「参加費は無料です」の一文だけ
元データ：参加費は未定
正しい表現：「参加費は決まり次第お知らせします」
変更禁止：定員10名、所要60分、日時・会場未定
出力：修正した文と変更点一覧

After making corrections, verify whether not only the fee but also the capacity and required time are preserved. If even the date and time have been arbitrarily decided, reject the output and request corrections again. It is important not to be satisfied just because the AI writes "I have corrected it."

Exercise 10: Determine whether the completed article or email is ready to be published or sent

In the final exercise, verify not only the quality but also the conditions for publication and sharing. Are the capacity and date/time correct? Has the contact information been officially decided? Is the information safe to disclose to the target audience? Who approved the text? Even if you have Copilot generate a checklist, it is not considered "approved" unless a human fills it out.

This is the conclusion of the exercise: Even after finishing up to the tenth step, the announcement for the fictional event cannot be officially published because the date, time, and venue have not been determined, and no one has given final approval. This is not a failure, but rather an opportunity to practice making the most critical judgment when introducing AI into practical operations. It enables you to handle the ability to generate text and the evidence required to declare it complete as two separate matters.

Exercise Log Template

演習番号：
使用したCopilot（個人用/職場用）：
使用日時：
入力した架空データ：
結果の主要部分：
期待結果との一致・不一致：
出典または数値の照合：
修正した条件：
画面で操作した事実（なければ「なし」）：
採用/保留の判定：

By keeping ten results using this template, you will be able to compare what has changed compared to previously, even if the model or interface changes. Record only the experimental results and do not enter unexecuted items as successful.

Official Information / Primary Sources (October 27, 2026)

• Microsoft Official Documentation 1 — update

• Microsoft Official Documentation 2 — license

• Microsoft Official Documentation 3 — limits

• Microsoft Official Documentation 4 — prompts

• Microsoft Official Documentation 5 — files

• Microsoft Official Documentation 6 — pages

Verification and Publication Limitations: The prompts, diagrams, and verification tables in this article are examples designed for the manuscript and do not represent actual Microsoft Copilot operation logs. Operations on user accounts, independent AI reviews, test displays on WordPress, and duplicate checks across all items including unpublished articles remain uncompleted.