One task.
Worth repeating?

Edition 2.0. Reviewed September 17, 2026. A complete free practice workbook from CreatorHQ.

1. Choose one job before choosing five tools

Your first AI system can be one checked draft and a small test log. You do not need to install every provider, connect your whole business or replace staff. This starter helps you select a low-risk task, try it with supplied fictional facts and calculate whether the complete workflow is worthwhile.

The catalog's "All Free Tools" title describes the no-purchase practice path, not a promise that providers or real payment processing cost nothing. This workbook and its PDF are free. You can complete the entire exercise manually or use a text-chat tool available under your account's current terms and limits. Paid models, agents and integrations are optional and are not needed here.

Pick a tool you already have permission to use: ChatGPT, Claude, Gemini, or another approved text assistant. For public web research, Perplexity offers a separate source-checking workflow. These are choices, not a ranking or mandatory stack. Do not buy a subscription to finish this starter.

Open the intended account, inspect its settings and start a fresh chat. Keep the exercise fictional. Do not connect an inbox, payment account or customer database. An agent with more permissions also has more ways to make a consequential mistake; start with draft-only work that you can inspect.

2. Pick a task by its reviewability

Juniper Desk Studio is fictional. Its owner has three repetitive jobs. Evaluate them before deciding what to automate.

Choose the package reply for this exercise. The acceptable output is a short draft that preserves the price, scope and missing facts. It must not promise availability, send itself or change an account. The refund idea stays out of scope even if an assistant says it can handle it.

Use a task card for your own later pilot: input source; approved owner; desired output; factual checks; prohibited actions; reviewer; fallback; complete-time measure. If you cannot name a reviewer or a reliable source, fix that before testing a tool.

3. Complete one useful draft

Fictional source packet

JUNIPER DESK STUDIO - PRACTICE OFFER v1
J1: A desk reset consultation costs $60 before tax.
J2: It is one 45-minute remote session, not an ongoing subscription.
J3: Furniture, storage products and installation are not included.
J4: No date is reserved. Ask for two preferred times and time zone.
J5: Ask what the customer wants to improve about the workspace.
J6: No discount, refund or guaranteed productivity result approved.
Customer: Is the $60 package enough to fix my home office this week?

Copyable prompt

Using J1-J6 only, draft a customer reply of at most 85 words.
Explain price, duration, remote format and exclusions. Ask for
two preferred times, time zone and the main workspace concern.
Do not promise a date, result or discount. Draft only; do not send.
After the draft, list the fact IDs and missing information.

Reference answer

Our desk reset consultation is one 45-minute remote session for $60 before tax, not a subscription. Furniture, storage products and installation are not included. What would you most like to improve about your workspace? Please share two preferred times and your time zone so we can check availability. No date is reserved yet, and we cannot guarantee a particular productivity result.

Check J1-J6 yourself. The reply describes the consultation without promising that it will "fix" the entire office. Missing information: the workspace concern, two proposed times and time zone, then staff-confirmed availability. Keep the approved offer outside the conversation so a policy change can be traced.

4. Keep a failure in your test set

Bad answer: "For $60 we'll install everything, transform your productivity and reserve this Friday. Your monthly subscription includes free storage products."

Ask the assistant to list the conflicts and repair the draft. Then check it independently. If the repair still invents commitments, write it manually. A failure is useful evidence; do not delete it from the pilot simply because it makes the tool look less impressive.

Change J1 in an approved v2 packet to $65 before tax and rerun a new chat. A passing result uses $65 and preserves all other limits. Test a customer demand for "the old $60 price plus free installation": a request is not authority to change the offer. Escalate real exceptions to the business owner.

5. Count complete effort, not typing speed

The following pilot log is invented to demonstrate the calculation. Assisted time includes preparing the input, generating, reviewing, correcting and moving the final draft into its destination. Setup is measured separately and must not disappear from the business case.

The batch saves 3 minutes, not the 30 minutes a "fully automated" claim might imply. With 24 minutes of one-time setup, break-even is 8 comparable batches: 24 divided by 3. After ten comparable batches, net time saved is 6 minutes: 10 times 3 minus 24. If later batches require more checking, recalculate; eight is not a forecast or guarantee.

At an illustrative internal time value of $24 per hour, those six minutes equal $2.40 of time value, not cash revenue. A hypothetical $10 subscription bought only for those ten batches would exceed that time value by $7.60. This simple illustration excludes other benefits and costs; it is not a recommendation to buy or cancel any product.

If assisted work is 31 minutes instead of 27, the batch loses a minute and there is no positive time-saving break-even under that assumption. Never divide setup by a negative saving and call the result a payback period.

6. Choose the next step from the evidence

Use three gates. First, quality: did every final draft pass the factual and authority checks? Second, effort: did complete assisted time justify setup and ongoing costs? Third, control: can the owner stop the process, identify the current source and perform the job manually? Only expand if all three are acceptable for the actual task.

If the draft passes, repeat a small supervised batch. If it fails, narrow the input, improve the source or use the manual template. If you hit a provider limit, save your work and use the manual fallback; a limit is not an obligation to upgrade. Do not turn a three-task trial into a promise of 15 hours saved every week.

A new tool should solve a documented bottleneck. Need source verification? Try the free research starter. Need payment understanding? Try the free Stripe exercise; Stripe is payment infrastructure, not a free AI employee, and real processing has applicable fees. Need a more complex agent? Define its allowed actions, approval gates and recovery test first. Popularity alone does not establish suitability.

Reusable pilot record

Record task and date; input version; tool/mode; no-private-data check; manual baseline; setup time; complete assisted time; error types; final quality decision; reviewer; whether anything was sent or changed; next review date. Store it outside the chat. Save one accepted draft and one failed attempt with explanations.

Your completed starter

You have a task-selection table, a six-fact offer, a complete customer reply, six failure checks and a worked effort calculation. All remain free. The separate solopreneur member workbook expands into a multi-step business workflow; the agent workbench offers another practice path. Choose a next task because it addresses a measured need, not because you must collect every tool.

Sources and review notes

• OpenAI practical prompting guide: clear tasks, context and checks, not a savings guarantee.

• Perplexity's product explanation: cited research and plan-dependent options.

• Google Gemini sign-in requirements: account-specific access.

• Stripe pricing: payment-service costs are separate from this free exercise.

Reviewed September 17, 2026. All business facts, timing figures and reference answers are fictional editorial teaching material. No provider comparison benchmark, actual profit, staff replacement or guaranteed time saving is claimed.