How to work with ChatGPT

The first session: from opening the page to an answer worth using

Here is what newcomers do first and should not: they type a one-line question, get a generic answer, and conclude the tool is no use for their work. You got a generic answer because you gave a generic question. The model sees nothing except what you wrote in the conversation.

Start the first session with a real job, not a test. Something you already owed this week: a services blurb, a reply to a customer who has asked three times, a list of what is missing from a quote. A real job shows you where the tool helps and where it wastes your time; a test only produces pretty answers.

In that first message write four things and nothing else: what the job is, who it is for, what constraint applies, and the format you want back. For example: "A services blurb for laptop repair, for the front page of a website. The reader owns a small business and has no time for technical detail. Maximum 120 words. Promise nothing about delivery time. Output: two versions in different tones." Those four lines are the whole distance between a useless answer and an editable draft.

After the first answer, edit and keep the conversation open. Say the second version was better but its third sentence is a slogan, and ask for that one rewritten. Every correction you write stays inside the conversation and the later answers can see it. Someone who opens ten separate chats has started from zero ten times.

One habit worth having from day one: keep the output somewhere outside the chat. A conversation is history, not your filing system. The moment a piece of text is useful to you, paste it into your own file.

Four things that stay put every time the interface changes

Any tutorial that explains the buttons is wrong six months later. What does not go wrong is the four concepts below; every chat-based tool gives them different names and they do the same job. Understand these and any interface change is just a button that moved.

The conversation is the container for a job. Each conversation is a separate file and knows nothing about another one. Working rule: one job, one conversation. If, halfway through writing your website copy, you ask an unrelated question about an Excel formula, that question becomes part of the file too and affects the answers after it.

Context is what the model re-reads from the beginning every turn. Contrary to the common picture, the model does not "remember" anything; the whole conversation is placed in front of it again. Two practical consequences: a very long conversation gets slow and expensive and the model starts ignoring things you said at the top, and anything you once wrote into the conversation stays with you until the end of it. The full explanation is in the large language model lesson.

An attachment is a file you hand to the conversation. The file gets read, but it does not become permanent knowledge for the model: in the next conversation it does not exist. More importantly, an attachment is exactly where people absent-mindedly send customer information out of the building; a contract file and a list of phone numbers are the two common cases.

Standing instructions are the setting pinned to every conversation: who you are, who you write for, what it must never say. This is the one place you write something once and it always applies, which is exactly why it pays best. Lower down this page, in the fast path, there is a ready template for it.

Five jobs you can hand over today, with ready prompts

We picked these five because they repeat several times a week in the work of an Iranian freelancer or small business, and none of them replaces your professional judgment. Swap what is inside the braces for your own.

One. Turn a customer voice note or chat into a written scope. Business here happens inside WhatsApp, and later the argument is about what "we agreed". This prompt turns scattered text into a scope and a list of open questions.

The text below is my conversation with a customer. Turn it into two parts:
1) Scope: only what the customer explicitly asked for, one bullet each.
2) Open questions: what is needed to start and is not in the text.
Do not add anything to the scope that the customer did not say. Where
something is ambiguous, put it in the questions list instead of guessing.
Conversation:
{paste the conversation here}

Two. A product description from a spec sheet. For a shop with a hundred items this is the difference between a week and an afternoon.

Write a product description for a shop page from the specs below.
Reader: {a non-technical buyer solving one specific problem}
Tone: {plain, no hype}
Cap: {80} words.
Rules: state no number that is not in the specs. Make no claim about
warranty or shipping time. Do not open the first sentence with the
product name.
Specs:
{the spec sheet}

Three. A canned-answer set built from your own answers. Paste the last twenty replies you sent customers and let it extract the pattern. The output comes back in your voice rather than the model voice, and that is what separates it from a template found online.

These are twenty real replies I sent to customers. Do three things:
1) Extract the five questions that repeat most.
2) Write a canned answer for each, in the tone and the sentence length
   you see in my own texts.
3) Wherever an answer needs a number, a date or a price, do not write
   one: leave a placeholder in braces.
Texts:
{twenty replies}

Four. Read a contract or a quote for what is ambiguous. This is not legal advice and must not stand in for it; what it does is put the gaps in front of you so you can decide.

The text below is a {quote}. Your role is reviewer, not legal adviser.
Give three lists:
1) Things that are written but can be read two ways.
2) Things usually present in such a document and missing here.
3) Places where my obligation is clear and the other side is not.
Do not comment on whether anything is legally right or wrong.
Text:
{the contract text}

Five. Turn a finished job into a portfolio entry. Most freelancers have no portfolio because writing one is tedious, not because they have done no work.

This describes a project I finished. Turn it into a portfolio entry
under four headings: problem, what I did, result, limit.
Under "result" write only what appears in my text. If I gave no number,
do not write "a significant increase"; write what changed.
Under "limit" write one real thing this work did not solve.
Project text:
{the project description}

Notice the shared shape of all five: a role, an input, an output format, and one explicit prohibition. That prohibition is the most important line. When a model does not know something it fills the gap; if you have not said in advance where it may not fill, it will, and six months later you are arguing with a customer about an invented number.

How far does the free plan go, and why this page prints no number

Any Persian page that prints an exact number for the ChatGPT free plan either lifted it from another page or read it months ago and never went back. We do not do that, and we write down why, because the reason itself teaches you something.

On 6 September 2026 we measured from this site own server: openai.com/robots.txt answers with 200 and forbids no crawling, yet the ChatGPT pricing page, the help centre and chatgpt.com itself all return 403. So the obstacle is an anti-bot layer. We do not write what we have not read, so this lesson carries no number for free message counts, upload caps or reset windows.

One thing can be read, and we read it. The OpenAI developer documentation, which does open from this server, publishes a supported-countries list, and Iran is not on it. That same page states plainly that accessing the services from outside the list may lead to an account being blocked or suspended. Note that the list concerns the API services and not the consumer app; we do not blur the two and we make no claim about the consumer app. The vendor situation is written up with the same directness on the OpenAI page in our AI reference.

So how do you learn your own ceiling? From inside the product, at the moment it stops you. Write down three things: what it stopped, at what time, and how many minutes later it let you through again. Two of those notes and you have the real pattern for your own account, which beats any number quoted in any article, because these ceilings differ between accounts and move over time.

And if you decide to pay, look first at whether a payment route exists at all. We wrote that layer up separately and with sources in the guide to buying AI tools. We do not sell accounts and we do not suggest ways around anything.

What not to paste, and which setting to check today

One simple rule that never goes stale: do not put anything into a chat that you would not hand to a stranger. Not because something bad always happens, but because once a thing leaves your hands the decision about it is no longer yours.

Specifically these four: passwords, API keys and tokens; anything a client entrusted to you that you have no permission to publish (a contract file, a customer list, medical or legal text); identity and banking details; and code or a server configuration file with a secret inside it. If you need a text reviewed, replace the names and numbers with placeholders first; the model does not need your customer real name to understand the structure of a contract.

On the ChatGPT data-training setting, this page makes no claim: OpenAI consumer help pages cannot be read from this server and we do not quote what we have not read. What you can do is go into your own account settings, find the data controls section, and read what it says there rather than what a Persian article quotes. For comparison, two vendors whose pages do open from here publish their policy precisely, and we wrote them up with sources: the Claude lesson and the Gemini lesson, both in this same learning track. Reading those two, even as a ChatGPT user, shows you what kind of setting you should be looking for at all.

Account security is three jobs and all three take ten minutes: a unique password used nowhere else, two-step login, and a periodic look at the list of devices and connected apps. Your chat account today is more than a hobby account; inside it sit contract drafts, negotiation text and possibly client files. Treat it as seriously as you treat your work email.

One point that gets said less often: pressing the thumbs up or thumbs down button means, in some products, that the whole conversation is kept for review. Before pressing it on a conversation with something sensitive in it, know what that product policy is.