How to use Microsoft Copilot – the new Bing AI version?

Microsoft Copilot (the new version of Bing AI) is a chat based on Bing search that allows you to quickly summarise, compare information and prepare content drafts, but it requires a deliberate setup of context and source checking. In practice, what matters most is whether you use Copilot in a browser, in Edge or in Windows, as well as whether you log in with a personal or work account. This determines access to chat history, privacy settings and, in a corporate environment, the level of Commercial Data Protection. Copilot can clearly speed up research and option comparisons, but it can be prone to errors and sometimes “fills in” facts, so do not treat its answers as a primary source. In this guide, you will learn the distinction between Copilot versions and the most important steps regarding access, sign-in and configuration. This will make it easier to choose the right usage mode and reduce common problems with features, language and stability.

What is Microsoft Copilot and how to distinguish it from other Copilots

Microsoft Copilot is the continuation of Bing Chat under a unified brand, available via copilot.microsoft.com and integrations (Edge/Windows/mobile apps). If you are wondering whether it is “still Bing”, the answer is yes — Copilot still uses Bing search to provide answers based on content from the Internet, and the biggest changes concern the interface and the development of features under the Copilot name. The easiest way to tell it apart is by where it works and what data it works with: online, it relies on sources from the Internet and on what you paste into the chat. Key rule: treat Copilot’s answers as a synthesis with references, not as final “proof” without verification in the cited sources.

Individual “Copilots” also differ in the scope of access to organisational data and privacy rules. Copilot in Microsoft 365 (Word/Excel/PowerPoint/Outlook/Teams) works on your files and organisational context (Microsoft Graph), but requires a separate licence and permissions, whereas Copilot in the browser answers based on the Internet and the content you provide. In turn, when you sign in with a Microsoft Entra ID work account, you can use Commercial Data Protection, which changes the rules for processing prompts and logs. If you work with sensitive company information, the account type and the data protection mode are the first criterion for choosing the environment.

Copilot works best for creating drafts, summaries, option comparisons and checklists, as well as for preparing interview questions and summaries of articles with citations. In practice, it can also help you write short scripts in Python/JS, provided you describe precisely what should be created. At the same time, Copilot can make mistakes or “fill in” facts when it lacks certain data, so for important decisions always verify the linked sources. If you plan to cite information, record the URLs, access dates and context, and check the details directly on the source pages.

The quality of answers in Polish is generally good, but it can depend on the domain and on how you ask for sources. For topics such as law or medicine, it is better to require citations and caveats more often, and in technology, ask for examples and tool versions when relevant. If you see that answers in Polish are sometimes less precise than in English, ask for an analysis of English-language sources and a translation of key excerpts into Polish. It is also worth remembering that Copilot may use different models (in practice: the GPT family) and search tools, so similarly phrased questions may produce different results.

1. 01Continuation of Bing ChatUnified brand, based on Bing
2. 02Synthesis from the InternetAnswers with links to sources
3. 03Scope and PrivacyDifferent access to organisational data

Key rule: Treat Copilot’s answers as a synthesis to be verified, not as final proof.

Access, sign-in and configuration of Microsoft Copilot

The easiest way to launch Microsoft Copilot is to go to copilot.microsoft.com and sign in with a Microsoft account, which gives you chat history and consistent settings. A limited mode without sign-in often also works, but then some features are unavailable, such as history, personalisation and more stable access. Depending on the device, you can choose the most convenient way of working: the side panel in Edge, the app in Windows 11 or the mobile app. It is a good idea to decide straight away whether it is more important to work “alongside” the page in the browser, or to have quick questions and voice dictation on your phone.

• Web: copilot.microsoft.com — good for working with chat history after signing in.
• Microsoft Edge: Copilot in the side panel — convenient when you are reading a page and want summaries or conclusions.
• Windows 11: Copilot app as a separate chat — practical for multitasking alongside other applications.
• Android/iOS: Microsoft Copilot app — good for quick questions, voice dictation and generating images on the go.

Copilot in Edge can usually refer to the context of the currently open page, but to be sure it is worth pasting a passage or asking for a summary with quotes. In Windows 11, you use the Copilot app as a separate chat, which makes parallel work easier, although current deployments more often move towards apps and integrations with online services than towards controlling system settings. On mobile, chat, content creation and graphics features are usually available, whereas working with long documents tends to be less comfortable than on a PC. If you care about repeatable results, signing in and a consistent working environment help maintain coherent settings and history.

If Copilot replies in another language, set Polish as the interface language and start the conversation with a clear request such as “Reply in Polish, use PL terminology”. In some regions, access to some features (e.g. image generation or selected integrations) is limited, so it is worth reviewing your account settings and, if necessary, changing the region in your Microsoft profile. If you use both a personal and a work account at the same time, make sure which one you are signed into, because this affects privacy, access to history and any corporate security controls. Differences in functionality in a work environment may result from organisational policies or from the fact that the commercial data protection mode works differently from the consumer version.

Chat history after signing in is usually saved automatically, which makes it easier to return to earlier threads, but you can also delete individual conversations and check your Microsoft account privacy settings. In the privacy and personalisation section, you can limit personalisation and control how Copilot uses account data, which matters with sensitive topics. The safest assumption is that everything you type may be analysed for the purposes of providing the service, so unless absolutely necessary, do not paste passwords, API keys or special category data. When Copilot does not load or gets “stuck” generating, disabling script blockers (uBlock/NoScript), clearing the cache or running it in InPrivate/Incognito often helps, and in Edge it often works more reliably thanks to deeper Microsoft integration.

Conversation modes and prompt writing in Microsoft Copilot

The conversation mode in Microsoft Copilot is selected to match the style of response to the task: Precise for facts and short answers, Balanced for most everyday uses, and Creative for ideas and copywriting. Differences in results often come precisely from changing the mode, because it affects the tendency to digress and the “boldness” in proposing solutions. If you care about consistency, keep one mode throughout the thread and refine the requirements instead of asking one very broad question.

The best prompts usually have a fixed structure: role + goal + constraints + answer format, so Copilot knows what perspective to adopt and what to prioritise. When you get vague responses, it is usually because constraints such as budget, market (e.g. Poland), date or preferences are missing. In practice, it is more effective to run the conversation in 2–4 steps: first a list of options, then clarification of the criteria, and finally a decision with justification and a request to confirm the assumptions.

• Prompt template: “You are [role]. Do [goal]. Constraints: [budget/market/date/priorities]. Return in the format: [table/JSON/checklist/SOP]”.
• Forcing the format: “Return a table with the columns …” or “Return JSON only, no commentary” (with a given field schema).
• Length control: “max. 180 words”, “3 paragraphs” or “10 points”, plus “First 3 most important takeaways, the rest optionally”.

When the topic is complex, ask Copilot to start by asking you 5 clarifying questions, because this reduces the risk that the recommendations will miss your situation. When working with your own materials, it is better to paste the key excerpts (e.g. SLA, pricing, limitations) and ask for a summary, risks and questions for the supplier, because longer content may be cut off or omitted. If you want to reduce “hallucinations”, add the condition: “If you are not sure, write ‘I don’t know’ and suggest how to check it” and “Provide links to sources for every number”.

1. 01Choose the right modePrecise (facts, brief), Balanced (everyday), Creative (ideas).
2. 02Consistency and disciplineKeep one mode for repeatable results; refine the requirements.
3. 03Structure of an effective promptRole + Goal + Constraints + Format (clear guidelines).

Key to efficiency: Match the mode and use a consistent prompt structure to get concrete and valuable answers.

Working with information from the Internet: sources and verification

Treat sources in Copilot’s answers as leads to verify, not as proof that every detail comes exactly from the linked page. When Copilot provides links, open them and check whether they confirm the specific claim or number, rather than just describing the topic in broad terms. If a link “does not say the same thing”, it usually means the answer is a synthesis from multiple places and needs manual checking. In practice, it also helps to search for a specific sentence on the open page (Ctrl+F) and compare at least two independent sources.

The freshness of the data depends on the topic, so for prices, regulations and software versions, ask explicitly for sources from the last 6–12 months and for the publication dates. If you want to assess discrepancies, ask for a table such as “source → claim → date → reliability”, and then for the contradictions and the most likely version. This approach makes it easier to separate facts from opinions and spot places where, for peace of mind, you need additional confirmation.

You can improve local search by adding the context “Poland, PLN, availability in PL, PL law, .pl sources if possible”, because without this Copilot often chooses popular English-language content. When you are looking for one specific document (e.g. a manufacturer’s manual or a regulation in PDF), a classic search engine is sometimes faster than the chat, and in Copilot it is worth being more precise: “provide 5 direct links to PDF documents, without summaries”. For academic work, Copilot can format a bibliography (APA/MLA/Chicago), but you must verify the metadata (author, year, title, URL) before using it.

When summarising pages, ask not only for a summary, but for “7 key theses and a source (URL) for each one”, so you can quickly check the basis for the conclusions. When analysing reviews, remember that rankings can be affected by affiliates, so it is worth asking about quality signals (methodology, measurements, number of tests), “partner/affiliate” labels, and a comparison with independent tests. In medical, legal and financial topics, treat the answers as initial research: ask for caveats, a list of questions for a specialist and pointers to official sources, because Copilot is not responsible for decisions.

Practical tasks in Copilot: writing, planning, coding

Microsoft Copilot works best for quickly creating content drafts, outlines and support with coding, provided you give specific context and the expected output format. When writing emails and messages, the key remains clarifying the recipient, relationship, objective, length and tone (e.g. “polite, concise, without corporate jargon”). When you need several versions, ask explicitly for a short, longer and more assertive variant, so it is easier to match the style to the situation. The more clearly you describe the goal and constraints (who, to whom, why, how long), the less time you will spend on revisions.

You can prepare documents such as a CV, cover letter or proposal more efficiently when you paste in the requirements from the job advert and ask for adaptation for ATS and a list of keywords. If you are concerned about “making up” facts, it is worth setting a working rule: “Do not invent facts; if data is missing, leave [TO BE COMPLETED]”. In project planning, Copilot can structure a backlog and sprint plan, but it needs the scope, deadline, resources and definition of “done”. To account for risks sensibly, ask for a risk register with the fields: likelihood (1–5), impact (1–5), mitigation plan and owner.

Coding and debugging in Copilot works most effectively when you provide the code snippet, the error message, the environment version (e.g. Python 3.11, Node 20) and the expected behaviour. Instead of a general “fix it”, ask for a minimal reproducible example (MRE), a list of hypotheses and unit tests (pytest/jest) that confirm the fix. When working with data and Excel, you will get to the result faster if you paste sample columns and describe the outcome precisely (e.g. “7-day moving average, excluding weekends”). Copilot can then suggest specific formulas (SUMIFS, XLOOKUP, LET) and explain them step by step, which makes implementation easier without getting lost.

1. 01Drafts & OutlinesQuick creation of drafts.
2. 02Emails & MessagesDefine tone and format.
3. 03CVs & DocumentsATS adaptation.

Clarifying context and expectations improves results.

Integrations, exporting results and automations with Microsoft Copilot

Copilot integrations make it easier to move an answer from “chat” mode into a real workflow, for example summarising pages in Edge and then further editing them in documents. In Microsoft Edge, the biggest advantage is parallel working: you read the page and immediately ask for conclusions, risks, counterarguments or a list of questions for the author. If you want to prepare a quick note from reading, ask for a bullet-point version with quoted excerpts and a list of actions to take. When exporting to Word or PowerPoint, the “draft → refine” approach usually works best, because final quality requires checking consistency, sources and formatting. The safest approach is to treat Copilot’s output as a draft, not as ready-to-publish material without correction and fact-checking.

In Teams and Outlook, integrations work when you have Copilot in Microsoft 365 as well as the appropriate permissions and organisational context. In Teams, Copilot can summarise meetings, capture decisions and produce a task list, provided the meeting was recorded or transcribed and you have access to it. In Outlook, it can summarise a thread and suggest a reply, but you decide on the tone and verify the figures and dates. If you need an assistant “for company procedures”, Copilot Studio lets you build your own bots based on your knowledge sources and flows, by configuring topics, sources (e.g. SharePoint) and actions, and then publishing them to Teams or a website.

Automations in practice start with standardising what Copilot is meant to return, so the result can be pasted straight into work tools. Copilot can prepare content ready to be moved into Jira/Asana/Trello (title, description, acceptance criteria and a checklist), if you provide the ticket template and required format. Even without direct integrations, you can generate materials in a structure convenient for populating knowledge bases (e.g. the standard sections “Goal”, “Steps”, “Checklist”, “FAQ”, “Source links”). In a team, it is worth establishing a “prompt playbook” and a review process in which a human checks the facts, style and policy compliance, because standardising input and controlling output reduce quality fluctuations.

Security, privacy and quality in Microsoft Copilot

Safe use of Microsoft Copilot starts with a conscious choice about what you paste into the chat. Do not paste passwords, API keys, payment card data, PESEL numbers or full medical data, because you do not control the entire processing chain. If you want to analyse a document, such as a contract, remove the parties’ data, amounts and numbers, and leave only the clauses for risk assessment. For sensitive topics, treat the chat as a supporting tool for working on anonymised fragments, not as a data repository.

Privacy and compliance in an organisation depend mainly on company policies, the type of account and the data protection configuration. If you are covered by an NDA or an internal security policy, use a company account with Commercial Data Protection or a tool approved by IT. There is no single answer to the question “is this GDPR-compliant?” — it depends on the data controller, the legal basis and the service settings. If in doubt, do not process personal data in the chat and escalate the matter to the appropriate people in the organisation.

You will maintain a high quality of Copilot’s answers when you actively reduce the risk of mistakes and thoroughly verify key elements. In areas such as law, finance, medicine or IT security, require sources, document versions and boundary conditions, then confirm them manually in the primary materials. In professional work, the approach “Copilot suggests, human approves” works best, and the decision should be based on a verifiable document or data. To assess results in a repeatable way, use a checklist covering the objective, factual accuracy, completeness, language, sources, freshness and risks, and also ask Copilot for self-checking (a 1–5 rating and an indication of what needs to be checked).

Problems and refusals to answer can usually be resolved by refining the prompt and sticking to legal, safe alternatives. When Copilot refuses because of safety policies or ambiguity, ask for an educational framing and best practices for defence rather than instructions for harmful actions. If sources are missing, ask it to assign links to individual claims, and when the result is cut off — ask for it to be split into parts and continued from the indicated point, without duplicating content. When it comes to code and scripts, do not run anything “blindly”: first read it, test it in a controlled environment and make sure there are no destructive operations (e.g. rm -rf, DROP TABLE), and in prompts require error handling, logging and tests.