AI At Work

1. Context explains the situation, the business, the audience, and the material the tool should use. Context prevents the tool from filling gaps with guesses. For example, write, “I run a residential plumbing company in Columbus, Ohio. The message goes to homeowners who requested a quote but have not scheduled service.”

2. Limits set boundaries for length, tone, facts, format, and content. Limits make the answer easier to review. You might write, “Use a friendly, direct tone. Keep the email under 150 words. Do not promise same-day service or invent prices. Include one subject line and one call to action.”

3. Examples show what good work looks like. A sample email, approved phrase, previous post, or short list of preferred terms gives the tool a pattern to follow. Examples work better than labels such as “make it sound like us,” because the tool can observe your actual wording.

4. Ask states the exact job and desired output. Use an action verb: draft, summarize, compare, classify, rewrite, extract, or create. Specify the format, such as a table, email, checklist, or three options. A precise ask tells the tool when it has completed the task.

5. Review tells the tool how to check its own work and gives you a way to improve the next response. Ask it to list missing information, flag unsupported claims, or provide a short quality check. Review does not replace your judgment; it helps you spot problems before you copy the result into a customer email or business document.

Here is a complete prompt for a follow-up email:

> Context: I run a residential plumbing company in Columbus, Ohio. A homeowner requested a water-heater replacement quote two days ago but has not scheduled an appointment. > Limits: Write under 150 words. Use a helpful, professional tone. Do not pressure the customer, mention a discount, promise an installation date, or invent pricing. > Examples: Our usual closing is, “Reply to this email or call our office at [PHONE] to choose a time.” > Ask: Draft one follow-up email with a subject line, a brief reminder of the quote, and one clear next step. > Review: After the email, list any detail I should confirm before sending.

Notice how the prompt supplies facts instead of asking the tool to guess. It also separates the draft from the review. That structure makes errors easier to find.

Use the CLEAR parts in order when a task matters. Start with the smallest useful amount of context, then add a limit when the first answer misses the mark. If the tone still feels wrong, add an example rather than repeating “make it better.” If the output contains unsupported claims, strengthen the limits and ask for a review. This adjustment process works because each change addresses a specific problem.

You can also ask the tool to handle source material carefully: “Use only the information below. If the information does not answer a question, write ‘Not provided’ rather than guessing.” That instruction matters when you paste meeting notes, product details, or customer messages. The practical takeaway: a strong prompt tells the tool what is happening, what it must avoid, what good looks like, what to produce, and how to check it.

Applying CLEAR to a Real Work Task

A small landscaping company needs a weekly email for customers with upcoming spring cleanups. The owner has a service list, a scheduling rule, and a preferred tone. The goal is not to ask AI for “a nice customer email.” The goal is to create a usable draft without inventing services or dates.

1. Gather the source details. The owner collects the actual facts: spring cleanup includes leaf removal, bed edging, and a final lawn blow-off; crews arrive between 8:00 a.m. and 10:00 a.m.; customers should move outdoor furniture before arrival; the company does not haul large branches unless the customer adds that service. These details give the tool something reliable to use.

2. Write the context. The owner states, “We send a weekly email to residential landscaping customers scheduled for spring cleanup.” This identifies the audience and purpose. Without it, the tool might write an advertisement for new customers instead of instructions for existing customers.

3. Add limits and examples. The owner writes, “Keep the email under 180 words. Use plain language and a calm, helpful tone. Do not promise an exact arrival time. Do not say large branches are included. Use this approved phrase: ‘Please make sure our crew can reach all areas listed on your service agreement.’” These limits protect accuracy and match the company’s normal voice.

4. State the ask and review. The owner requests, “Draft one subject line and one email with a short preparation checklist and a reply option for questions. Then identify any statement that requires confirmation.” The review request directs attention to claims that could cause a customer complaint.

5. Check the result against the source. The expected outcome is a draft that takes less than five minutes to edit. The owner checks the service list, arrival window, preparation instructions, and contact details before sending. If the tool adds branch hauling or says “your crew will arrive at 8:00 a.m.,” the owner removes the claim and updates the prompt.

6. Save the improved prompt. The owner stores the final version in a document named “Spring Cleanup Customer Email.” Next week, the owner can replace the service details and reuse the structure. This turns one successful request into a repeatable work asset rather than a one-time experiment.

A useful prompt might look like this:

> Context: We send a weekly email to residential customers scheduled for spring cleanup. Services included are leaf removal, bed edging, and a final lawn blow-off. Crews arrive between 8:00 a.m. and 10:00 a.m. > Limits: Keep the email under 180 words. Use plain, calm language. Do not promise an exact arrival time. Do not say large-branch hauling is included. > Examples: Use this approved phrase: “Please make sure our crew can reach all areas listed on your service agreement.” > Ask: Draft one subject line and one email. Include three preparation steps and a reply option for questions. > Review: List any statement that goes beyond the information provided.

Quick checklist

• State who you are and what task you need completed. - Identify the audience and the purpose. - Provide the facts the tool must use. - Set a word count, tone, format, and “do not” rules. - Include a real example when style matters. - Ask for one specific output. - Request a review of missing or unsupported details. - Compare the answer with your source material before using it. - Save prompts that work for recurring tasks.

The owner can measure whether the method helps by tracking editing time for three weekly emails. If each draft needs only a few factual corrections instead of a full rewrite, the prompt supplies enough direction. The takeaway: use CLEAR to make AI produce a controlled first draft, not an unchecked final answer.

Common Prompt Problems and Practical Fixes

The request has no useful context

“Write a reminder for a customer” leaves out the customer’s situation, the business, and the desired action. The tool must guess whether the reminder concerns an appointment, payment, renewal, or delivery.

Do this: State the situation, audience, purpose, and relevant facts. Not this: “Write a professional reminder.”

For example, write, “Draft a 100-word reminder for a customer whose annual equipment inspection is due next month. Ask them to call our office to schedule. Do not mention a fee because the service agreement covers the inspection.” The added details guide both the content and the boundaries.

The prompt asks for a tone but gives no example

Words such as “friendly,” “professional,” and “modern” mean different things to different businesses. A tax preparer may want a calm, precise message, while a children’s dance studio may use warmer language. The tool cannot infer your normal style from a tone label alone.

Do this: Paste a short approved message and explain what to preserve, such as sentence length, greeting, or closing. Not this: “Make it sound more like our brand.”

Remove private customer information before sharing an example. Replace names, phone numbers, account numbers, and addresses with placeholders. You can preserve the writing style without exposing sensitive details.

The output sounds confident but contains invented details

AI tools can present guesses as facts, especially when a prompt asks them to fill gaps. A draft might add a warranty period, claim that inventory is available, or promise a delivery date that you never supplied.

Do this: Tell the tool to use only the provided information, mark missing details, and list claims for review. Then verify every customer-facing fact. Not this: Copy the draft directly into an email because it sounds polished.

For a product description, write, “Use only the specifications below. If a specification is missing, write ‘details to confirm.’ Do not infer materials, delivery times, certifications, or warranty terms.” This instruction reduces unsupported claims, but your final check still matters.

The prompt combines several jobs and produces a mixed result

A request such as “Summarize this meeting, write a follow-up email, create a task list, and suggest a sales plan” asks for four different outputs. The tool may give shallow answers or blend them together.

Do this: Separate the work into labeled outputs or run one prompt at a time. Ask for a meeting summary first, then use the confirmed summary to draft the email. Not this: Accept a long response that makes it difficult to tell which statements came from the meeting and which statements came from the tool.

Clear prompts make review faster because you can compare each output with a specific purpose. Start with context, set the limits, show an example, make the ask, and request a review. When you repeat that pattern, AI becomes easier to direct across emails, documents, customer messages, and routine business tasks. The better your instructions, the less time you spend correcting avoidable guesses - and the more useful automation becomes.

End of chapter one. 4 more chapters and about 9,333 words in the full book.