AI Business Toolkit

Write the offer as it exists today: who it is for, what they receive, the price you actually charge, and what you refuse to promise. Ask Gemini to find vague lines, not to invent a new market. If you have not spoken to customers, say so in the prompt. The model may prepare interview questions. It may not play the customer or estimate demand.

Paste anonymized notes from real conversations and ask for themes, contradictions, and questions you still need to ask. Require a quote or a note label under each theme. Drop any theme you cannot point to. Friends who like the idea are not a segment. Keep that distinction in the prompt so the summary does not flatter you.

Turn a real constraint into a small test: one audience, one offer, one channel, one week, and one thing you will count. Ban new features the business cannot deliver this month. Ask what could go wrong in operations, not only in the wording. You choose the test. A toolkit prompt does not replace a week of doing it.

For a product page or menu, paste the spec or the price list and forbid claims that are not on it. Ask for a plain description and a list of adjectives with the line that justifies each one. Delete adjectives with no line. Have a second person try to spot a promise you cannot keep. That read is the verification step.

Build proposals from redacted discovery notes. Ask for situation, goals, in scope, out of scope, assumptions, and open questions. Tell the model not to invent metrics, logos, or testimonials. If a heading is empty, it should say what to ask the prospect. Write the fee only after a human who heard the call checks the outline.

Use a critic pass before you send. Ask where the scope can be misunderstood and where unpaid work might hide. Do not allow the critic to add services. Confirm or delete every smoothed assumption. A clean proposal that invents agreement will cost you later.

After a workshop or a client meeting, sort notes into decisions, owners, and unresolved items. Forbid fake consensus and invented quotes. Read the page aloud to the room or send it for correction the same day. Strike anything the model inferred from who talked the most. The business result is a decision people recognize, not a longer recap.

Keep a small set of business prompts in a document you control: offer test, proposal outline, scope critic, and decision log. Note the check beside each one. This course is independent of Google, so do not store the only copy inside a chat product. When a template mentions a Workspace export, check the current screen. Paste a redacted excerpt if the button is missing.

Use Gemini on recurring writing: reminders, agendas, and status notes. Give it your real hours, prices, and policies. Tell it not to add rules you did not write. A reminder that invents a cancellation fee is worse than no reminder. Read numbers aloud against the notebook before anything is public.

For email, start from one checked public fact or one fact the recipient already knows. Ask for a short note and one question. Ban fake familiarity, invented urgency, and a feature list. Verify every claim, then send it yourself. Do not paste a full contact export into the chat.

For hiring notes, use job criteria you already have. Ask for evidence, the question asked, and what is unknown. Do not ask for a score, a rank, or a hire recommendation. Remove names and personal details before pasting, and follow your company policy. Send the summary back to the interviewer to correct.

For a team calendar, apply a meeting test before you draft an agenda. If there is no decision, ask the model to say so and to format a written update instead. Ban comments about a person's performance. Confirm owners with the people named. The toolkit should remove meetings, not decorate them.

Paste figures you calculated, with the definition of each column. Ask Gemini to spot gaps, odd jumps, and missing comparisons. Ask it not to invent a benchmark. If you want a benchmark, you must supply it. A fluent explanation of a number you do not understand is not analysis.

Treat any forecast as a scenario with assumptions written out. Ask for what would make the scenario fail. Do not ask whether the business will succeed. You cannot outsource that judgment, and the model does not know your street, your kitchen, or your pipeline. Label estimates as estimates when you share them.

Separate a decision memo into context, options, tradeoffs, and the choice you still have to make. The model may fill the tradeoff list from facts you provide. It should not pick the winner unless you are practicing how to argue both sides. Even then, label the pick as practice. High-stakes choices need a human who bears the result.

Once a month, review the toolkit against actual use. Which prompts did you run, which checks did you skip, and which invented claim almost got through? Rewrite the prompt that failed. Drop theater. A business toolkit is a set of habits that survive a busy week, not a binder of unused ideas.