Expert

Cover of AI for Client Work

AI for Client Work

A scope you can defend.

Anyone can use this page. Copy the prompt and try it today.

$9.99

Buy · $9.99

Everyday AI

AI for Client Work

Scope, status, and the hard email.

You stay the editor. The model drafts. You decide.

$7.99 · 20 pages

Expert shelf · paid engagement · you stay the lead

Everyday AI · Client Work1 / 20

CONTENTS

What's in this book

Read pages 3 through 6. Then copy one cream box.

03The Idea — The draft did not attend kickoff
04What The Labs Found — Editing beats imaginary certainty
05Before You Paste — Say what is settled
06How To Use This — One job. One record. One checked draft.
07Scope — Draw the boundary from the record
08Kickoff — Ask without reopening the deal
09Status — Name the slip without spin
10Change Request — Name the delta before the price
11Estimate — Use real hours and show the math
12Feedback Round — Sort comments without agreeing to them
13Decision Log — Rescue the decisions from the thread
14Client Deck — Build the story from work that exists
15Pause Note — Stop the work without inventing the terms
16Case Study — Build the proof file before the story
17Close-Out — Finish the handoff, not the legend
18Stuck — Correct the draft you have
19Before You Send — The seven-line engagement check
20Sources — Research and guidance used on page 4
Everyday AI · Client Work2 / 20

THE IDEA

The draft did not attend kickoff

Mara opens Monday’s project thread. The signed outline says five pages. A newer message casually asks for seven. The chat did not hear the agreement or watch the work change.

A model can organize the record. It cannot create the record.

The record: signed scope, approved notes, real hours, real messages.

The draft: wording suggested from that record.

The decision: what you agree to, price, pause, or send.

Confusing those three is how polished fiction enters a paid engagement. A useful prompt names a role, task, details, limits, and format. The cream boxes are editing briefs, not permission slips.

You decide what is in. You check every number. You send the message. You sign any agreement.

The short rule: Give it the record. Ask for a draft. Keep the decision.

Everyday AI · Client Work 3 / 20

WHAT THE LABS FOUND

Editing beats imaginary certainty

A client thread can contain a contract excerpt, call notes, comments, and half a decision. The useful research here is about writing from real material—not winning client disputes.

1 · Start from your words. Writing is about 40% of work-related ChatGPT messages. About two-thirds of those writing messages modify user text.

2 · Put the packet first. Anthropic advises placing long documents before the query. In tests, this improved quality by up to 30% on complex, multi-document inputs. It is not a guarantee.

3 · Name the job. Google’s guidance includes objective, persona, constraints, context, and response format.

4 · Inspect the first reply. Anthropic’s AI Fluency Index links iteration with more goal clarification and critical evaluation. This is correlational, not proof of better client outcomes.

The practical lesson: bring the record, specify the job, and inspect the draft.

Everyday AI · Client Work 4 / 20

BEFORE YOU PASTE

Say what is settled

Dev is annoyed. The client called a new dashboard “one small addition.” He is about to paste the message while angry.

What is settled: The signed scope includes two reports, not a dashboard.

What changed: The client requested a dashboard today.

What I know: It adds design, data mapping, and review work.

What I decide: I will name the change now and price it after checking my hours.

This separates evidence from emotion. It also stops the model from deciding whether the client is unreasonable.

Do not ask, “Who is right?” Ask, “Using only this record, show the difference between the agreed work and the new request.”

Failure mode: A confident draft turns frustration into an accusation.

Then do this: Remove every adjective about the client. Keep the agreement, change, and next step.

Everyday AI · Client Work 5 / 20

HOW TO USE THIS

One job. One record. One checked draft.

Inez has a contract, a notebook, and three client channels. She opens the page matching today’s decision.

  1. Collect the smallest useful record.
  2. Replace every bracket. Write [unknown] when needed.
  3. Run one cream box.
  4. Check names, dates, deliverables, hours, totals, approvals, and quotes.
  5. Send in your own system.

Privacy: Follow your agreement and workplace rules. Remove private IDs, passwords, access keys, sensitive personal data, and material you cannot share. Use relevant excerpts when permitted.

Hard nos: No invented fees, rates, hours, discounts, metrics, dates, approvals, client quotes, legal terms, or social proof. These are communication drafts, not legal or tax advice.

Then do this: If the first reply is wrong, use the correction on that page. Do not send faster because the prose sounds polished.

Everyday AI · Client Work 6 / 20

SCOPE

Draw the boundary from the record

Adele’s kickoff notes mix promises, ideas, and “maybe later” items. She needs an internal scope map.

COPY THIS

Role: You are a scope editor for an active paid engagement.

Task: Turn my record into included, excluded, unclear, and change path.

Details: Client/project: [Adele / brand refresh]. Signed scope: [paste]. Kickoff notes: [paste]. My current view: [paste].

Format: Four sections. Tie each item to my record. Label it “Internal draft—not a contract.”

Hard nos: Do not add deliverables, fees, hours, deadlines, legal terms, or client statements. An idea is not an approval. Mark unsupported items unclear.

Failure mode: Brainstormed ideas become promised deliverables.

Then do this: Compare it with the signed scope. Move every unsupported item to Unclear.

Everyday AI · Client Work 7 / 20

KICKOFF

Ask without reopening the deal

Farid starts a site rebuild Monday. Price and page count are settled. Access and approvals are not.

COPY THIS

Role: You are a kickoff-question editor for a project with agreed commercial terms.

Task: Draft questions that expose missing inputs and protect scope.

Details: Agreed outline: [paste]. Settled items: [paste]. Known gaps: [CMS access, approver, content owner]. Attendees: [roles].

Format: Eight questions under Access, Decisions, Inputs, and Review. Add one-line reasons. End with two questions to avoid.

Hard nos: Do not suggest new deliverables, workshops, fees, timelines, policies, or contract language. Do not assume authority.

Failure mode: A generous question creates unpaid discovery.

Then do this: Circle the three questions that must be answered live.

Everyday AI · Client Work 8 / 20

STATUS

Name the slip without spin

Kira’s Friday update is due. Two pages slipped while legal review sat with the client.

COPY THIS

Role: You are a plain-language status editor.

Task: Draft a short update from my facts.

Details: Completed: [paste]. Slipped: [paste]. Known reason: [paste]. My next action: [paste]. Honest date: [date or unknown]. Needed decision: [paste or none].

Format: Under 120 words. Headings: Done; Changed; Next; Needed. Missing dates become [confirm].

Hard nos: No invented percentage, date, blame, motive, or claim that a delay was unavoidable. Do not hide the slip.

Failure mode: “Waiting on the client” becomes an accusation.

Then do this: Replace motive with chronology: what was due, what arrived, and what happens next.

Everyday AI · Client Work 9 / 20

CHANGE REQUEST

Name the delta before the price

Ben’s site was scoped for two languages. The client requested a third. He needs the extra work made visible.

COPY THIS

Role: You are a change-request editor. I decide price and terms.

Task: Draft a change summary from the original scope and new request.

Details: Original scope: [paste]. New request: [paste]. Verified extra work: [paste]. Schedule effect: [known or unknown]. Next step: [paste].

Format: Original agreement; New request; Work added; Open questions; Price [I will add]; Timing [confirm]; neutral approval line.

Hard nos: No invented fee, rate, hours, deadline, penalty, legal clause, or consequence. Do not judge the request.

Failure mode: The draft invents leverage, such as an unverified launch delay.

Then do this: Verify the delta with the person doing the work. Add money yourself.

Everyday AI · Client Work 10 / 20

ESTIMATE

Use real hours and show the math

Mei has remaining tasks in a notebook. One task has no estimate. She does not want a guessed total.

COPY THIS

Role: You are an estimate formatter and arithmetic checker.

Task: Organize only my numbers and show the calculation.

Details: Tasks and hour estimates: [paste]. Rate: [paste or not provided]. Already billed: [paste]. Currency/tax treatment confirmed: [paste or unknown].

Format: Table with Task, Low hours, High hours, supplied rate, low subtotal, high subtotal, Notes. Show formulas. Mark missing inputs.

Hard nos: Never invent hours, rates, fees, discounts, tax, contingency, currency, or a midpoint.

Failure mode: Correct-looking arithmetic rests on an invented input.

Then do this: Recalculate outside the chat. Delete any total touched by an invented number.

Everyday AI · Client Work 11 / 20

FEEDBACK ROUND

Sort comments without agreeing to them

Tasha receives 40 comments. Some fix errors. Some request new concepts. Two are too vague to classify.

COPY THIS

Role: You are a feedback sorter, not an approver.

Task: Compare each comment with the scope.

Details: Scope: [paste]. Numbered comments: [paste]. My known interpretations: [paste]. Allowed review round: [paste].

Format: Table with Comment, Plain meaning, In scope/Possible change/Unclear, Evidence, Question needed. Add three receipt sentences that accept no extra work.

Hard nos: Do not invent or strengthen comments, decide a fee, approve work, or attribute motives. Weak evidence means Unclear.

Failure mode: “Will do” accidentally accepts a change.

Then do this: Replace commitment verbs with “noted,” “reviewing,” or “needs confirmation.”

Everyday AI · Client Work 12 / 20

DECISION LOG

Rescue the decisions from the thread

Noor has call notes and two Slack messages. Everyone remembers the color approval differently.

COPY THIS

Role: You are a decision-log editor.

Task: Extract decisions, owners, open questions, and contradictions.

Details: Notes or permitted transcript: [paste]. Follow-ups: [paste]. Existing log: [paste or none]. Meeting date: [date].

Format: Table with Item, Confirmed/Open/Conflicting, exact support, explicit owner, explicit date, confirmation needed. End with a five-line correction request.

Hard nos: Discussion is not approval. Do not infer owners, dates, votes, or consensus. Do not invent quotes.

Failure mode: Repeated suggestions are mistaken for decisions.

Then do this: Send the log to the real decision-makers. Extraction is not approval.

Everyday AI · Client Work 13 / 20

CLIENT DECK

Build the story from work that exists

Owen presents Thursday. The design is ready, but results data will not exist until after launch.

COPY THIS

Role: You are a client-deck outline editor, not the presenter.

Task: Build a presentation sequence from my work and evidence.

Details: Audience: [paste]. Time: [20 minutes]. Completed work: [paste]. Three points: [paste]. Evidence available: [paste]. Evidence unavailable: [results]. Decision needed: [paste].

Format: Eight slide titles. For each: point, evidence, and what I say. Mark two optional slides.

Hard nos: No invented results, metrics, testimonials, quotes, praise, reactions, or certainty. An artifact is not an outcome.

Failure mode: “What we made” becomes “what succeeded.”

Then do this: Label each slide Work, Evidence, Interpretation, or Decision.

Everyday AI · Client Work 14 / 20

PAUSE NOTE

Stop the work without inventing the terms

Luis’s client writes, “We need to hit pause.” Two files are delivered. A third is half built.

COPY THIS

Role: You are a pause-note editor. Do not interpret the contract.

Task: Draft a calm operational note stating what stops and what needs confirmation.

Details: Exact request: [paste]. Delivered work: [paste]. Work in progress: [paste]. Actions stopping: [paste]. Handoff: [paste]. Restart needs: [paste]. Confirmed money/notice terms: [paste or undecided].

Format: Under 150 words: Pause acknowledged; Work status; Immediate actions; Questions. Bracket unconfirmed terms.

Hard nos: No invented fee, refund, credit, notice, storage promise, restart date, right, or threat. Pause is not termination.

Failure mode: “Pause” is rewritten as termination.

Then do this: Check the agreement or ask the appropriate professional. Fill terms yourself.

Everyday AI · Client Work 15 / 20

CASE STUDY

Build the proof file before the story

Hanna has screenshots and deliverables, but no permission to name the client and no measured result.

COPY THIS

Role: You are a case-study evidence editor.

Task: Separate usable facts from claims needing proof or permission.

Details: Work: [paste]. Verified facts: [paste]. Measured results: [paste or none]. Supplied quotes: [paste or none]. Permission: [status and wording]. Names/images: [paste].

Format: Brief; Work; Verified evidence; Permission/proof needed. Add a ledger: Claim, Source, Permission, Safe to publish. Label Internal Only unless confirmed.

Hard nos: No invented metrics, quotes, outcomes, praise, names, permission, or anonymity promise. Delivery is not impact.

Failure mode: A plausible result becomes fake social proof.

Then do this: Ask for permission and evidence separately. Publish only what passes both.

Everyday AI · Client Work 16 / 20

CLOSE-OUT

Finish the handoff, not the legend

Petra’s project is done, but one login transfer and two source files remain.

COPY THIS

Role: You are a close-out editor.

Task: Draft a factual handoff and internal lessons list.

Details: Delivered and accepted: [paste]. Still owed: [paste]. Client actions: [paste]. Files/access: [paste]. Unresolved items: [paste]. Slips and known reasons: [paste].

Format: Client section: Delivered; Remaining handoff; Client next actions; Open items. Internal section: three fact-linked lessons and three possible process changes.

Hard nos: Do not declare acceptance, completion, satisfaction, or results without evidence. No testimonial, blame, fee, waiver, or invented lesson.

Failure mode: A polished close-out hides unfinished obligations.

Then do this: Complete or date every remaining handoff item before the goodbye.

Everyday AI · Client Work 17 / 20

STUCK

Correct the draft you have

A first reply may sound official while being slightly untrue. Use one precise correction.

Expanded scope: “Use only the pasted scope. Move unsupported items to Unclear.”

Accepted extra work: “Classify the request. Do not agree to it.”

Invented money: “Delete every number I did not provide.”

Hidden slip: “State the slip, chronology, and real next action.”

Blame: “Remove motives. Keep dates, dependencies, and actions.”

Fake decision: “Show exact evidence or mark it open.”

Legal tone: “Use plain operational language. Invent no rights or remedies.”

Fake proof: “Delete unsupported outcomes, quotes, and testimonials.”

Failure mode: Repeated polishing makes a false draft more persuasive.

Then do this: If the structure is wrong, start again from the smallest verified record.

Everyday AI · Client Work 18 / 20

BEFORE YOU SEND

The seven-line engagement check

Rosa has a clean change note ready. Before sending, she checks the source—not the tone.

  1. Scope: Can I point to supporting words?
  2. Numbers: Did every number come from me?
  3. Authority: Is this approval or discussion?
  4. Quotes: Are they exact, permitted, and needed?
  5. Commitments: Did the draft promise anything I did not approve?
  6. Next action: Is the real owner and deadline clear?
  7. Channel: Does this belong in the engagement record?

Read it as the client. Is the boundary clear without your private frustration?

Read it as the person doing the work. Did you promise something they cannot deliver?

Failure mode: Smooth language makes an unsupported claim feel settled.

Then do this: Keep the record. Check the draft. Make the decision.

Everyday AI · Client Work 19 / 20

SOURCES

Research and guidance used on page 4

Chatterji et al. “How People Use ChatGPT.” NBER Working Paper 34255, September 2025.

https://openai.com/index/how-people-are-using-chatgpt/
https://www.nber.org/papers/w34255

Anthropic. “Be clear and direct.” Prompt-engineering documentation.

https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/be-clear-and-direct

Anthropic. “The AI Fluency Index.” Education report/tutorial, 2026. The reported relationships are correlational.

https://academy.claude.com/tutorials/the-ai-fluency-index

Google Cloud. “Prompt design strategies.” Current prompt-design documentation.

https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/prompts/prompt-design-strategies

Research can guide a process. It cannot verify your engagement. Your scope, hours, approvals, and messages remain the source.

Everyday AI · Client Work 20 / 20