How to Use the 200-Prompt Library

The library is a shelf of worked examples, grouped by jobs such as research, writing, sales, and planning. Each prompt shows a purpose, a moment to use it, and a way to improve the reply. It is not a promise that the first paste will be finished work. This course is independent of Google, so the words travel even when a product button moves.

Read the purpose line before you copy anything. If the purpose is a customer email and you need a lesson plan, you are in the wrong row. Skim two neighbors in the same category and pick the closer fit. A near miss that you edit is better than a perfect title that solves a different job.

Do not run twenty prompts in one afternoon and call that practice. Pick one prompt that matches work you have this week. Replace the bracketed details with your real context, then delete any detail you are not allowed to share. Run it once, check the draft, and write what you changed. That single loop teaches more than a folder of unused copies.

Some prompts ask for structured output, such as a table or a checklist. Keep that structure on the first run so you can see what the pattern does. Only then loosen it if your reader needs prose. If the library example includes a sample fact, treat the fact as fake and replace it. Never ship an example number, name, or citation that you did not bring.

Start from the job, not from a clever title. Ask what decision, document, or practice session you need in the next hour. Filter the library by that category and ignore the rest. If two prompts both fit, choose the one whose constraints are stricter. Loose prompts feel friendly and then waste a revision.

Read the when line as a gate. A research prompt that assumes you already have sources is the wrong tool on a morning when you have none. Go get the sources, or switch to a planning prompt that lists what to gather. The library will not fetch your files for you. Using the wrong stage of a workflow is a common way to get fluent nonsense.

Look at the expected result before you paste. If you needed a decision memo and the prompt produces ten slogans, stop. Either change the output line or pick another card. Do not hope the model will notice the mismatch. The card is a brief you are responsible for aiming.

When a prompt mentions a product feature, such as a file upload or a connected app, check the current screen. As of late 2026 those extras depend on plan and layout, and this course does not require them. If the feature is missing, do the same job by pasting a short excerpt you are allowed to share. Write that workaround next to the card.

Keep the role, the bans, and the output shape. Swap the topic, the audience, and the facts. If you delete the ban on invented numbers, you have removed the safety rail. Add one local constraint, such as a word limit or a term you must use. Change one thing at a time so you know what helped.

Paste your material under a clear label, such as Notes or Spec sheet. Tell the model to use only that block for factual claims. If a sentence in the library example conflicts with your notes, your notes win. Say that explicitly. Models follow the last specific instruction more reliably than a general vibe.

After the first reply, use the improve line on the card as a second turn, not as decoration. Ask for the critic pass, the claim list, or the shorter version the card suggests. Then do the human check the card cannot do: open the source, confirm the price, or reread the policy. Save the adapted prompt only if the check passed. A saved failure will betray you later at speed.

If the adapted prompt still misses, diagnose before you browse for a new card. Missing context, a vague output, or a leaked instruction from an older chat are the usual causes. Start a fresh chat when the thread is tangled. Check the current screen if you are unsure whether tools are on. Return to the library only when the job itself was wrong.

Build a personal short list of ten to twenty prompts from the two hundred. For each, keep the original title, your edited text, the job, and the verification step. Chat history is a poor library because it mixes drafts with secrets and dies when you switch tools. This course is independent of Google, so do not lock your best briefs inside one product pane.

Review the short list once a month. Delete prompts you did not run. Rewrite any prompt that keeps producing filler or extra facts. Add a note about the audience if your job changed. A living sheet of fifteen beats an archive you do not trust.

When you invent a new prompt that worked twice, give it the same fields as a library card: purpose, when, text, expected result, and how to improve it. That discipline stops you from saving a lucky paragraph. Share it with a teammate only after you remove private details. Invite them to try the check, not just the paste. A prompt without a check is a rumor.

Use the library on real tasks, then bring the result back to the lesson that matches the category. If you cannot explain why the prompt is shaped that way, you are copying. Reread the cheat sheet when your edits keep getting worse. The two hundred examples are raw material for judgment. Judgment stays with you after the tab is closed.