Using an AI Conversation

Why this matters

Most beginners use a chat tool the way they use search: one short question, one answer, done. That works for trivia and fails for building, because building is a sequence: describe, draft, inspect, correct, refine. The chat holds the thread across that sequence, but only if you feed it well. A vague first message produces a generic answer; three precise follow-ups produce something you can use. The skill is not clever wording. It is giving the tool the same things you would give a helpful colleague: what you want, what you already have, and what the result should look like.

Concepts

A conversation has four working parts:

• Message. Your turn. A good first message has three ingredients: the

goal ("a one-page site for my uncle's bakery"), the context ("he sells bread and cakes, customers call to order, no online payment"), and the output shape ("give me the page sections as a list, then the text for each").

• Answer. The tool's turn. Read it as a draft: note what is usable, what

is generic, and what is missing.

• Follow-up. Your correction turn. Point at the specific part to change

and say what to change it into: "Keep the three sections, but rewrite the second one for phone orders instead of walk-ins." Follow-ups beat restarts: they keep everything that already works.

• Context. What the chat remembers from earlier messages. Context is

short and fragile. After a long chat the tool starts forgetting the beginning, mixing up decisions, and repeating suggestions. That is your signal to act.

Starting a fresh chat is a deliberate move, not a failure. Start fresh when the task changes ("now the flyer, not the site"), when answers start contradicting earlier ones, or when you catch yourself arguing with the tool for the third time. Copy the two or three sentences that still matter into the new chat as your first message. Lesson M06L23 will turn this into a reusable project-summary habit.

Files and screenshots. Many tools accept an attached file or image. What helps most is small and specific: the exact file you are working on, or a screenshot of the exact problem with a sentence saying where to look ("the button at the bottom overlaps the text on my phone"). A twenty-file dump with "fix it" is the chat equivalent of shouting. One page, one pointer, one question.

Worked example

Vague first message: "I need a website for a bakery." Likely answer: generic advice plus generic text you could have found anywhere. Nothing to build on.

Better first message: "My uncle runs a neighborhood bakery. Customers currently order by phone. I want a one-page site with three sections: what they sell, opening hours, and how to order by phone. List the three sections first, then write two sentences for each. Simple language, no online shop."

Now the follow-ups do the real work: "Change the hours section: closed on Mondays." Then: "Rewrite the order section so it tells people to call before 18:00 for next-day pickup." Each answer inherits the previous decisions, and after three turns you hold text that is actually yours in shape, even though the tool drafted the sentences.

Expected result: a short conversation, saved or copied somewhere you keep notes, that ends with usable text and a clear next action, not with more questions.

The common wrong move

Two opposite ones. First: accepting the first answer untouched, which produces the same generic page as ten thousand other beginners. Second: restarting the chat every time something is slightly off, which throws away the good parts along with the bad. Default to follow-up; restart only when the task or the thread is genuinely stale.

Lab and next step

Lab A02 gives you a half-written conversation with the context in the wrong places; you will move goal, current state, limits and output shape where they belong. Next, lesson 3 maps the tool landscape: chat tools, coding assistants and app builders, and which one to reach for when.