Artificial intelligence: finish your first useful task

Artificial intelligence is a broad name for systems that use patterns in data to generate content, organize information, and perform tasks. This guide starts with something you can finish: a checked customer email you can save and reuse. You do not need to understand model architecture or buy a subscription before trying your first useful task.

After this page, you will be able to:

• Choose a starting tool and draft your first customer email.
• Adapt a ready-to-copy instruction to your own job.
• Improve a weak answer with a specific correction.
• Remove confidential details and check unsupported claims.
• Save a reusable template and measure the complete task.

Content updated: October 14, 2026. Short on time? Complete the first section before reading the rest.

Your first 15 minutes: can you finish a useful customer email?

Your goal is one checked response, not a complete automation system. Use a browser such as Chrome, Safari, or Edge on a computer or phone. You will not need to upload a file, connect your email account, or enter a payment card. If you already have an account, skip the registration step. If not, allow extra time for verification.

The timings below are a target, not a measured guarantee. Account setup, accessibility needs, and a slow connection can make the session longer. Finishing means you have a saved response whose facts you have checked. Sending it is optional. For the exercise, use the supplied training facts; for a real customer, replace them with your actual policy first.

Step 1: open the official service, minutes 0–2

Type chatgpt.com into your browser's address bar, the field where the website address appears. ChatGPT has a free plan with usage limits. Start there for this short writing exercise. Do not click an advertisement promising a special version, and do not buy a subscription simply to follow this guide. If you prefer Google, open gemini.google.com and check the plan offered to your account.

Expected result: the official chat screen or an account access screen. Outside Azerbaijan, check the provider's country list for your location. Being listed does not guarantee that a particular workplace account or every feature will be available. If neither service opens, check your connection before changing account settings or purchasing anything.

Step 2: sign in or create an account, minutes 2–5

Choose “Log in” or “Sign in” if you already have an account. Choose “Sign up” to register if that option is shown. Follow the email or existing-account method offered on the official screen. If you receive a verification email, open your inbox and complete that service's verification. Button placement and wording can change; the official domain is the important check.

For Gemini, use the Google account flow. Work and school accounts may be restricted by an administrator. Never paste your password or verification code into the AI conversation. If a payment screen appears, return to the free choice for this exercise. If registration stalls, ask a trusted person to help locate the right button without handing over your password.

Expected result: a text field where you can enter a message. You do not need an advanced model setting, an extension, or an email integration. Leave those options alone until the basic task works.

Step 3: copy a complete instruction, minutes 5–8

A prompt is the instruction you type into the tool. Select the text in the block below and copy it with Ctrl+C on Windows or Command+C on a Mac. Click the chat input field and paste with Ctrl+V or Command+V. On a phone, press and hold to select text, then choose Copy and Paste. Use the screen's send button.

Training scenario: this is an illustrative appointment business, not a reported customer case. The times and service conditions are exercise inputs, not claims about a real company. An international reader can later replace the dates, time zone, and cancellation policy with verified local details.

Write a friendly customer email in plain English.
Customer question: “Can I move my appointment to Friday?”
Use only these verified facts:
- We can offer Friday at 10:00 or 14:00.
- Both times are in the business's local time zone.
- The appointment is not changed until the customer confirms.
Keep the response to three short sentences.
Do not invent a fee, discount, cancellation policy, or guarantee.
Ask which time the customer prefers.

Expected result: a short draft. An author-written example is: “We can offer Friday at 10:00 or 14:00, in our local time zone. Please let us know which time you prefer. Your appointment will stay unchanged until you confirm.” This example demonstrates the requested output; it is not a recorded model result or a claim of successful customer service.

Step 4: check and correct it, minutes 8–12

Compare the draft with the three facts. Are both times correct? Is the time zone clear? Does the response avoid changing the appointment before confirmation? Remove any new fee or policy. If Friday could refer to different dates, add the actual date before sending. A language model cannot resolve a calendar ambiguity from information you never supplied.

Revise the email in natural, plain English.
Keep the two appointment times unchanged.
Remove any policy, fee, or promise I did not provide.
Make clear that the customer must confirm a time.
Use three short sentences and no sales language.

Expected result: a draft that matches the supplied facts and sounds like your business. Read it aloud. If you would not say the sentence to a customer, edit it yourself. Asking the model whether its answer is correct is not an independent check; compare against your appointment record or policy.

Step 5: save and reopen the response, minutes 12–15

One specific Windows route: select the Start button, type “Notepad” in search, and open the text application. Copy your checked response with Ctrl+C and paste it into the blank document with Ctrl+V. Press Ctrl+S. In the save window, choose Desktop, enter “Appointment-response.txt” as the file name, and select Save. Close Notepad, find that file on your desktop, and double-click it. If your text reappears, it has been saved. If you cannot find the file, search for “Appointment-response.txt” from Start.

Phone alternative: open your Notes application, choose a new note, press and hold in the blank text area, and select Paste. Give the note the title “Appointment change response.” Return to the notes list, find that title, and reopen it. Check that the text is still there. If your application shows a separate Save or Done button, select it before leaving.

Expected result: you can find and reopen the checked response, not just see it inside the chat. The 15-minute target assumes an existing working account; first-time registration and verification may take longer.

Copy the checked draft into a notes app or a text document. Name it “Appointment change response.” Under it, add “Verify date, availability, and time zone before use.” You now have a reusable work asset. Before a real message, replace the training facts, insert the actual date, and check that the offered slots are still available.

You do not need to grant access to your inbox. Paste the finished text into your email application when you are ready and review the recipient and subject separately. The first useful result is a checked, saved response, not an automatic message. For additional examples, use the email writing prompt guide or the customer service templates.

What does artificial intelligence do, and what should you not assume?

A text model learns patterns and relationships during training and generates a response to your instruction. For practical work, treat it as a drafting assistant rather than a source that is always right. Fluent writing does not establish that a price, date, quotation, or legal requirement is correct. Content generation and fact verification are separate tasks.

The distinction matters when you ask for work that depends on your business. “Write a reply to a customer” supplies almost no context. “Offer these two available appointments and ask for confirmation” supplies a task with boundaries. The system does not automatically know your live calendar, stock position, or approved policy. If those facts appear without being supplied or checked through an authorized source, do not use them.

A useful prompt contains the situation, the task, the permitted facts, and the output format. A role can help, but “act as an expert” does not replace missing information. Add an instruction to ask about missing details instead of guessing. That instruction is helpful, not a guarantee; you still need to review the result.

Try this now: remove the appointment times from your exercise and ask the tool what it needs before drafting. A useful response asks for availability. If it invents times, reject those facts and add the missing information yourself. This small test reveals whether your prompt gives enough context.

Avoid turning the first session into a discussion about which model is strongest. For a simple email, the important question is whether the output follows your facts and needs little editing. If you need a more demanding business workflow, start with the guide to AI-supported growth systems after your first task is finished.

Six job playbooks: two useful tasks for each role

Every example below is a training scenario with an author-written sample output, not an invented success story. Replace the bracketed fields with permitted information and remove the brackets. Complete one task before trying another. You should leave with a draft, checklist, study card, or formula you can actually review and use.

Do not put confidential documents into a consumer chat simply because the prompt asks for context. Give the minimum authorized information needed. For people-related work, use roles or anonymous labels. For regulated work, keep professional judgment and approval outside the model. Those boundaries are part of the task, not an optional step after it.

Small business owner: answer a question and organize the day

Start with a repeated customer question. Write the actual policy separately from what the customer wants. A request for a refund is not evidence that your policy offers one. Give the model the rules it may use and ask it to flag missing information. Check every commitment before the response becomes a template.

Draft a plain-English response for my small business.
Customer question, with personal details removed: [question].
Approved policy and verified facts: [facts].
Give two versions: two sentences and four sentences.
Do not promise anything outside the supplied policy.
If a key fact is missing, ask me one question first.

Sample output: “We will check availability and get back to you. Which size do you need?” This fits only when availability is unknown. If stock is confirmed, replace uncertainty with that verified fact. Do not keep a vague template when your business already has a clear answer.

Your second task is a practical daily plan. Supply the work, its deadlines, and your available time. A useful plan identifies an action and an owner rather than making everything sound urgent. Move the first action into your actual calendar or task list; otherwise the output remains an unused suggestion.

Turn these tasks into a realistic daily plan: [tasks and deadlines].
Available working time today: [minutes].
Use a table: task, priority, first action, responsible role.
Mark unknown duration as “estimate needed.”
Do not assume every task can be finished today.

Sample output: “Confirm open orders | high | open the order list | owner.” Check that the owner has time and authority. For restaurant-specific work, the restaurant prompt collection offers a more focused starting point.

Sales and marketing: product copy and a respectful objection response

For product copy, provide verified features and one clear action you want the reader to take. Do not add unsupported superlatives, invented testimonials, medical outcomes, or scarcity claims. A product photo may not establish its material, durability, or compatibility. Use a specification you have checked rather than asking the model to infer it.

Write three short social posts in plain English.
Product: [name]. Verified features: [list].
Audience: [audience]. Desired action: [action].
Each post must stay under 50 words.
Do not invent prices, promotions, reviews, or outcomes.
Finish each version with one clear next step.

Sample output: “Designed for [verified use]. Message us to check the available sizes.” Replace the bracket before publishing. Check that the stated use is supported by the product, not merely plausible. For a specific platform, try the Instagram writing prompts and the ad copy workflow.

The second task handles a price objection without pressuring the customer. Explain verified value and ask what they are comparing. Do not authorize a discount or create a deadline you cannot honor. Your sales policy should control the answer.

A prospect says our offer is too expensive.
Price: [amount and currency]. Included items: [verified list].
Write three respectful replies in plain English.
No pressure, invented urgency, or unauthorized discount.
End each reply with one clarifying question.

Sample output: “Installation is included in this price. Are you comparing the product alone or the installed service?” Use that only if installation really is included. Continue with the sales prompt guide if this is a recurring task.

Teacher: build a short exercise and check understanding

Provide the learning objective and level before requesting questions. Use material you are allowed to share. Do not include class lists, individual grades, health information, or student identifiers. Solve the generated questions yourself before using them; an answer key is another model output, not independent proof.

Create a classroom exercise in plain English.
Level and topic: [level, topic]. Learning goal: [skill].
Use this permitted source excerpt: [short excerpt].
Write five questions: two basic, two applied, one explanation.
Put the answer key in a separate section with reasoning.
Do not require facts outside the supplied material.

Sample output: “What is half of 12? Answer: 6, because 12 is split into two equal parts.” Check both the mathematics and the language level. If a question requires knowledge not in your lesson, either provide that background or remove the question.

Your second task is an understanding check, not automatic grading. Ask for questions that require learners to explain a reason. Supply your assessment criterion, then decide whether each question genuinely checks that criterion. Keep the decision about a student's performance with the teacher.

Write three understanding questions about [topic].
Each question should ask the learner to explain a reason.
Assessment criterion: [my criterion].
State the skill each question checks.
Use no student names or personal information.

Sample output: “Why do two equal halves make a whole? Skill: explain the relationship between parts and a whole.” Adjust the wording to your class. Test the question aloud before putting it into a worksheet.

Student: make study cards and improve an outline

Turn your own notes into questions you can answer without looking. Do not ask for quotations or page numbers from a book you have not checked. Separate questions from answers during practice. The useful result is finding what you cannot yet explain, not producing more text to memorize.

Turn my notes into six study cards: [my notes].
Format: question | short answer.
Use only information contained in the notes.
List unclear points under “ask the teacher.”
Do not invent citations, quotations, or page numbers.

Sample output: “What is the central term? | The definition supplied in the notes.” Hide the answer, write your response, and compare. Mark the cards you miss and review those again later. Use the student study prompts for more learning-focused tasks.

Your second task is improving the structure of your own writing. Check your course's AI policy first. If assistance is permitted, ask for structural feedback and missing evidence, not an essay to submit as your own. Keep a record of assistance when your institution requires it.

Review my outline: [outline].
Thesis: [my claim]. Assignment requirements: [requirements].
Do not write the finished essay.
Suggest three structural improvements.
For each, ask what evidence is missing. Do not invent references.

Sample output: “The second paragraph explains the claim but does not support it. Which source you have read provides an example?” Find that source yourself and write the paragraph in your own words. A plausible citation is not a checked reference.

Accountant: organize documents and test a simple formula

For month-end preparation, use an anonymous checklist rather than uploading client accounts. Remove bank details, tax identifiers, payroll information, and contracts. Ask the tool to organize your approved process, not create tax obligations or filing deadlines. Your jurisdiction and professional requirements remain separate from the writing task.

Organize this anonymized month-end checklist: [checklist].
Columns: document, responsible role, review step, missing detail.
Do not add legal requirements or filing deadlines.
Turn unclear items into questions for me.
Use plain English and short labels.

Sample output: “Sales records | accounts team | confirm reporting period | missing dates need checking.” This is document organization, not approval of the accounting treatment. Assign an owner to each missing item and confirm the actual deadlines in your approved source.

Your second task is a simple spreadsheet formula. Describe the columns and test with harmless sample values before using a real workbook. Regional settings can affect separators and function names. One correct cell does not establish that an entire spreadsheet is correct.

Column B contains quantity and column C contains unit price.
Write a multiplication formula for D2 and explain it.
Test with B2=3 and C2=4.
Do not assume tax, currency conversion, or rounding rules.
If it fails, ask which spreadsheet application I use.

Sample output: “=B2*C2; 3 multiplied by 4 equals 12.” Compare with a hand calculation and test blank or text cells before extending the formula. Use the spreadsheet prompt guide and keep tax decisions outside this exercise.

HR and office staff: action notes and a vacancy draft

A meeting summary should distinguish a decision from a suggestion. Use permitted notes with names replaced by roles. Do not turn discussion into agreement or assign a deadline that was never stated. Confirm the final action list with the people responsible before treating it as the team's record.

Summarize these anonymized meeting notes: [notes].
Use three sections: decisions, open questions, actions.
Action table: task | responsible role | stated deadline.
If a deadline is missing, write “not assigned.”
Do not treat suggestions as decisions or invent owners.

Sample output: “Review the price list | sales lead | not assigned.” Ask the sales lead to confirm ownership and agree on timing. If the notes do not establish a decision, move the item into open questions instead.

Your second task is drafting a vacancy from approved job information. Keep requirements related to the work. Do not ask the model to rank candidates by personal characteristics or infer sensitive attributes. Missing pay and hours are questions for the hiring team, not spaces the model should fill.

Draft a vacancy in plain English.
Role: [role]. Responsibilities: [list]. Required skills: [skills].
Approved terms: [terms]. Application method: [method].
Sections: the work, requirements, how to apply.
Do not add age, gender, or family-status requirements.
Ask about missing pay and hours instead of inventing them.

Sample output: “Record customer requests and route them to the appropriate team.” Check the draft against actual responsibilities and applicable hiring rules. For related writing tasks, use the resume prompt guide; it is not a candidate-selection system.

Which tool should you choose first?

Use the same safe task to compare tools. Record factual errors, editing effort, and whether you can finish without a subscription. Do not compare one tool's best answer with another's vague prompt. You need a tool that fits your task, not a universal winner. The table covers web access, not every mobile feature or organization account.

The official ChatGPT free-plan explanation establishes the starting option and limits. The OpenAI country list and Gemini web availability list cover the regional claims above. Language output and translated interface menus are different questions.

For an English-speaking international reader, country availability should be checked against your own location. For multilingual work, ask for the target language explicitly and have a competent speaker review important messages. Do not assume an English draft transfers unchanged to another market. The ChatGPT and Gemini comparison and Azerbaijani-language workflow provide next steps.

Adapt the work to your market and keep control of the risks

A local service business can draft an appointment reply from its confirmed calendar. A small online shop can explain verified dimensions. A training center can turn permitted material into a short exercise. These are application scenarios, not evidence that named businesses achieved a measured outcome. You can test them with your own facts without inventing a success story.

Local context means more than adding a city name. Specify currency, date format, time zone, service area, and the tone customers expect. “Friday at 10” can be ambiguous across time zones. A refund phrase may imply a commitment in one market that your policy does not authorize. Check the final wording against the actual service conditions.

A verifiable example of Anar's original published work: the site's Azerbaijani AI prompt library organizes instructions around role, context, task, and output format, with a separate fact and style check. This linked resource is in Azerbaijani. You can apply that observable structure to the English exercises here without assuming a customer revenue result or a measured productivity claim.

Privacy: share less before changing settings

Remove names, phone numbers, addresses, account identifiers, salaries, and contract details. Use labels such as Customer A or Product B. If the task cannot be completed without sensitive information, use an organization-approved system and policy rather than a general consumer conversation. Permission to access a document is not necessarily permission to send it to an external service.

ChatGPT's official data controls explanation describes the “Improve the model for everyone” setting under Settings and Data Controls. Turning it off does not authorize uploading confidential files or remove every risk. Data minimization is still the first practical protection.

Accuracy: compare against the original source

Use your price list for prices, your calendar for appointments, approved learning material for lesson answers, and official sources for regulated requirements. Open a cited link and check that it supports the specific claim. A working URL is not enough. If the claim cannot be established, remove it or mark it unresolved in a private draft.

Do not treat a second model's agreement as independent evidence. Both outputs can be wrong. Ask what information is missing, then supply a checked fact. For content work, use the AI content quality guide and the SEO prompt workflow as separate editorial tasks.

Rights: check permission for the inputs and the intended use

Do not paste another author's full article and ask to publish it under your name. Start with your own facts and material you are permitted to use. For images involving recognizable people, brands, or existing works, check consent and rights before commercial use. Review the tool's terms and the source material's permissions separately; generated output is not a universal rights clearance.

Before publishing, answer four questions: have I removed private details, checked the facts, avoided unauthorized promises, and confirmed permitted use? If one answer is unknown, keep the result as a draft. The AI-generated content rights guide offers a deeper workflow, not a blanket legal guarantee.

Seven beginner mistakes and specific repairs

1. Asking for “a good email” without facts. Add the customer situation, permitted information, length, and desired next action. Missing context cannot be replaced by asking the model to sound expert.
2. Sending the first draft immediately. Compare every time, price, name, and condition with the original record. Save the checked version, not whichever answer appeared first.
3. Uploading a complete document unnecessarily. Select a permitted excerpt and remove identifiers. If confidential information is essential, switch to your approved workflow rather than a consumer chat.
4. Mixing unrelated businesses in one conversation. Start a new conversation for a new task and restate the facts. Do not allow an earlier policy to become the next client's condition.
5. Accepting the wrong language or tone. Specify the target language and provide a short example of your own style. Read the result aloud and edit phrases you would not use.
6. Buying a plan before identifying a limit. Complete a safe task first. Record the actual limitation, frequency, and review effort, then assess whether a paid option solves that particular problem.
7. Using confidence as evidence. “I am sure” is not a source. Open the original document or official page and compare the claim. Remove information that cannot be checked.

If several problems appear together, repair facts first, tone second, and formatting last. You will see which change improved the output. A shorter answer with an incorrect promise is still incorrect. A polished paragraph is useful only after its content matches the task.

Your next action: save a repeatable task, not a pile of prompts

Choose one task for the next week: customer replies, product descriptions, or meeting action notes. Save the instruction and your review checklist together. On the next attempt, update the facts rather than blindly reusing the previous response. A repeated workflow becomes valuable when its owner knows what must be checked.

Copyable first-task worksheet

My first AI-supported task
Task:
Tool:
Verified facts supplied:
Confidential details removed:
Location of the saved result:
Errors I corrected:
Time spent doing the task without AI:
Total time drafting AND reviewing with AI:
Facts to recheck before reuse:
Decision: use / revise / do not use

Record the start and finish rather than guessing time saved. Include account access, prompt preparation, editing, and verification in the AI version. If the draft arrives quickly but checking takes longer than writing from scratch, the workflow is not yet helping. Simplify the task or improve the supplied information before expanding it.

Do not report increased revenue unless you have actually measured it. A saved reply is a work result; it is not proof of additional sales. Compare like-for-like tasks over several attempts and keep rejected outputs in your private notes if they explain why a prompt failed.

For another focused task, choose the content planning prompts, freelancer workflow prompts, or team training guide. If repeated customer work later justifies a system, read how to prepare a website chatbot. Automation comes after a reliable manual task, not before it.

Completion rule: you have one saved output, checked facts, and a note explaining what must change next time. You can stop reading and use that asset. Completing every linked article today is not necessary; applying one reviewed result is the point of this page.