Work, Clients and Jobs: Using AI Responsibly Where You Are Paid to Decide

At work, you are paid to decide. A manager decides whether a report goes out. An analyst decides which numbers to trust. A freelancer decides what to deliver to a client. AI can speed up each of these decisions, but it cannot take over the responsibility that comes with them. This page looks at the workplace side of responsible AI use: employer rules, what to tell clients, who answers when something goes wrong, and how AI is changing the kinds of work people can find.

It continues from Learning, Integrity and Disclosure: Using AI in Study and Assessment, where you practised reading rules and disclosing briefly, and from Creative Rights, Imitation and Freelance Practice: Fair Work with AI, where you saw how a freelancer protects clients and creators. Here the cast is employees, managers and freelancers, and the tone stays respectful of all of them. Employers have real duties to customers and staff. Workers have real needs for time and skill. Both deserve a fair hearing.

What you will learn

• How to review an employer's AI policy, and what to do when there is none or when it is unclear.
• How to decide what to tell a client about AI-assisted work, and why a person stays accountable even when a tool helps.
• How AI assistance differs from replacing a human service, and how AI may change entry-level work.
• The main arguments about job displacement and job creation, and why access to tools can be unequal.
• How AI use can strain trust between colleagues, how to write an honest AI-use statement, and how to decide what to disclose in three workplace cases.

One running scenario

Invented scenario:

Kojo is a junior analyst at Ridgeway Logistics, a company in Accra. His manager, Ms Adeyemi,
asks him to prepare a monthly delivery-performance report for a client, a retail chain. Kojo
uses a chat assistant to summarise a long supplier document and to draft the commentary. His
colleague Kavya writes the charts. The company has an "AI guidelines" page that Kojo has never
read. The client's contract is long and nobody has checked whether it mentions AI.

Review your employer's AI policy first

A policy is the employer's own rule. It is not the same as an ethical concern or a law (see Frameworks, Standards and Competing Principles: Ethics, Law and Policy), and it can be stricter or looser than either. This tutorial cannot tell you what employers usually allow, because policies differ widely. It can teach you to read yours. Here is an invented example with the usual gaps.

Example policy:

AI GUIDELINES (Ridgeway Logistics, invented example, version 1, 2026-03-01)
1. Staff may use approved AI tools for drafting and summarising.
2. Do not enter confidential client information into AI tools.
3. Staff are responsible for the accuracy of anything they send.
4. AI-generated content in client deliverables must be labelled where the client requires it.

When you read a policy, ask these questions and note the answers.

If there is no policy, say so in writing to your manager, and propose a temporary approach, such as: use AI only for non-confidential tasks, check everything, and tell the manager what you used. Never use a personal account for confidential work data unless someone with authority says you may.

Disclosing AI-assisted client deliverables

Telling a client about AI help is not an admission. It is information that may affect their decision, their risk, or what they are paying for. Three questions guide you: Does the contract or a rule require it? Would the client reasonably want to know? Could the AI use change quality, risk, ownership or confidentiality? If the answer to any is yes, tell them, early and simply.

A note to a client can be short.

Template:

Subject: Delivery report - note on how it was prepared

Dear Ms Okafor,
In preparing this report we used an AI writing assistant for a first draft of the commentary
and a summary of the supplier document. Kojo Boateng reviewed every statement against the data,
and Kavya Nair produced the charts. We did not enter your customer names into the tool.
Please tell us if your contract or policy asks for anything different.
Ms Adeyemi, Ridgeway Logistics, 2026-10-12.

Reasonable people disagree about how much to say. Some argue for a short note on every deliverable, so that the client never has to wonder. Others argue that clients pay for results, not tools, so mentioning a spell-checker-level aid is noise. Both sides agree that a contract term, a material effect on quality or risk, or a client's direct question settles the matter: disclose.

Human accountability and tool assistance

A tool can assist. It cannot answer for the result. When a report goes out under your name, you are accountable for it, whoever or whatever drafted it. The same holds for organisations: a company that puts an AI system in front of customers is usually treated by customers as having made its statements. The following documented case shows one tribunal's reasoning.

Documented case:

Moffatt v. Air Canada (2024)
Source: a law firm's published analysis of the decision (decision: 2024 BCCRT 149, February 2024).
Established: a customer was told by the airline's website chatbot that he could apply
retroactively for a bereavement fare. A linked policy page on the same site said otherwise.
The British Columbia Civil Resolution Tribunal found negligent misrepresentation, rejected the
argument that the chatbot was a separate entity, and said the airline is responsible for all the
information on its website.
Limits: a small-claims tribunal; limited precedent. Other courts may differ.

Legal note (as of 2026-10-12; not legal advice): According to a law firm's analysis of a February 2024 decision of the British Columbia Civil Resolution Tribunal, the tribunal held that an airline was responsible for what its website chatbot told a customer and rejected the idea that the chatbot was a separate entity. This is one tribunal's decision on its own facts. It is not binding on other courts and says nothing about your country. Check the current official text for your country and sector.

The practical lesson is a habit, not a legal rule. Before you send AI-assisted work, ask: can I explain every claim in it, and can I fix it if it is wrong? If the answer is no, you are not yet ready to be accountable for it. The editors' recommendation you met on the last page makes the same point from another field: a chatbot cannot be an author because it cannot be accountable.

AI assistance or replacement of a human service

When AI helps a person do a service better or faster, a person still provides it. When AI replaces the person, the receiver may get a different service from the one they thought they were getting. Neither is automatically wrong. The ethical question is whether people receive what they reasonably expect, and whether they can reach a human when it matters.

Invented scenario: a small clinic in Cairo uses AI to draft appointment reminders. A person reviews them: assistance. The same clinic uses an AI chat to answer patients' symptom questions with no staff review: closer to replacement, and a much higher stake. Karim, the clinic manager, may still choose a limited AI role, with a visible route to a nurse. The question is not "AI or people?" but "what is the right mix for this service and its risk?"

How AI may change entry-level work

Many people learn a job by doing its simpler tasks: drafting first versions, summarising documents, cleaning data, writing basic code, answering simple queries. These are the tasks AI is often good at. That raises a real concern: if those tasks shrink, how do beginners build skill? It also raises possibilities: beginners might do more interesting work earlier, or new kinds of entry roles might appear. Nobody knows yet how this will turn out, and it will differ by sector and country.

For a manager, a useful question is: if our juniors never write a first draft, how will they ever learn to judge a good one?

Displacement and creation: arguments from several sides

Source note (as of 2026-10-12): According to summaries of an International Labour Organization working paper (2025 update of an index published in 2023), the main effect of generative AI is more likely to be transformation of jobs than their elimination; exposure to the technology rises with national income; and clerical work is among the most exposed. These are modelled estimates of exposure to tasks that AI can do, which is not the same as jobs lost. Uncertainty is high, and this tutorial gives no numbers.

The table sets out strong arguments in the open debate. The voices are invented to show perspective; the arguments are not facts about any named person or organisation.

Each side has points the other should hear. History cannot settle it, because this technology differs from earlier ones in speed and in the kinds of tasks affected. For your own decisions, the useful stance is practical: look at your own tasks, learn what tools do well and badly, and keep building the skills they cannot replace, such as judgement, trust and domain knowledge.

Unequal access to AI tools at work and school

Not everyone can use these tools equally. A better paid version may cost money that some people cannot spend. Some workers need a fast connection or a recent device. A tool may work better in some languages than others: one study found that the same content takes different amounts of text processing, and therefore different costs, across 22 languages, so speakers of some languages can be charged more for poorer results (Ahia and colleagues, 2023). Employers may give tools to some teams and not others, and schools in different places may have different rules or resources.

Pay attention when evaluating performance: if some workers use stronger tools, a comparison of outputs may measure the tool, not the person.

When AI use could undermine trust between colleagues

Teams run on trust. AI can quietly weaken it when people use it in ways their colleagues would not expect. Trust does not need a ban on AI; it needs honesty and consent about how it is used.

The same questions come from earlier pages: who else is affected, and would they be surprised? If a colleague would be surprised and upset to learn how a tool was used, that is a strong sign to change what you do or to tell them first.

An honest AI-use statement for a report

Whether a report is for a client, a manager, a school or the public, an AI-use statement should be specific, short and checkable. Replace "AI was used" with the actual facts.

Template:

AI-USE STATEMENT
Tool: a chat assistant (name and version: ______), used 2026-10-05 to 2026-10-08.
Used for: a summary of the supplier document and a first draft of section 3.
Not used for: the data, the charts, the conclusions or the recommendations.
Checked by: Kojo Boateng, who compared every figure and claim with the source data on 2026-10-09.
Confidential information: no customer names or contract terms were entered.
Accountable author: Kojo Boateng.

A weak statement says: "This report was created with AI assistance." It does not say what was done, who checked it, or who is accountable. A statement can also be too long. Aim for what a careful reader needs to judge the work, in five to seven lines.

Where this shows up at work

Teams that treat these questions as normal tend to run smoothly. A short policy, a habit of asking before you paste, a standard sentence for client deliverables and a manager who invites questions all help. Freelancers who state their approach in the proposal, and employees who flag unclear rules early, are usually seen as careful and not as difficult.

Common mistakes

• Not reading the policy until after something goes wrong. Fix: read it when you start, and ask about gaps.
• Assuming the client does not care. Contracts and expectations differ. Fix: check the contract, and when unsure, tell them.
• Blaming the tool. "The AI wrote it" is not an answer to the client or to your manager. Fix: check before you send, so you can explain every claim.
• Replacing a human service without saying so. People get a different service than they expect. Fix: tell them, and keep a route to a person.
• Treating tool differences as skill differences. Fix: ask who has which tools before comparing performance.
• Using secret AI in sensitive human tasks. It harms trust. Fix: be open about it, and do the personal parts yourself.

Try it yourself

1. Easy. For each case write A (assistance: a person reviews and owns the result) or R (closer to replacement): (a) a translator uses AI for a first draft, then edits it and signs it; (b) a shop replaces all support staff with a chat tool and gives no way to reach a person; (c) a lawyer's assistant uses AI to summarise a long file for the lawyer, who reads the original key pages.
2. Medium. Use the example policy in this page. List four questions you would ask your manager before using a chat assistant for a client report. For each, say what could go wrong if you do not ask.
3. Hard. Exercise: decide what to disclose in the three workplace scenarios below. For each, say whether you would disclose, to whom, what you would say in one or two sentences, and what rule or fact would change your mind.

Invented scenario:

A. Kojo summarises a 40-page supplier document with a chat tool to understand it quickly. He then
   writes his own two-page internal memo for Ms Adeyemi. He copies no tool text into the memo.
B. Greta is a freelance translator in Berlin. Her client's contract says "translation by a human
   translator". Greta produces a first draft with an AI tool and edits it heavily. The client
   has not asked how she works. Her invoice is for human translation.
C. Kavya is a junior analyst in Pune. Her company policy says AI content in external reports
   must be labelled. She asks a tool to write the commentary for a chart in a client report,
   rewrites two sentences, and checks the numbers. The report goes out under the company's name.

Answers

Easy. (a) A: a person edits and signs it. (b) R: the service is replaced and there is no human route. (c) A: the lawyer reads key parts of the original and owns the result. Pass: you apply the questions "who owns the result?" and "can the receiver reach a human?" Weak: labels with no reason.

Medium. A strong answer picks four from: (1) Which AI tools are approved? Risk: using an unapproved tool. (2) What counts as confidential, and can a supplier document with client names go in? Risk: a breach of the client's trust. (3) Does this client's contract require AI content to be labelled? Risk: breaking the contract. (4) Who approves exceptions, and where do I record them? Risk: no cover if something goes wrong. (5) Is the policy still current? Risk: following old rules. Pass: four clear questions, each tied to a specific consequence, and a plan to ask in writing. Weak: general questions such as "Is AI okay?" with no consequence.

Hard. A strong model answer. A: No disclosure to the client is needed, because the tool only helped Kojo understand, and nothing it wrote is in the memo. A short note to Ms Adeyemi is good practice if the employer's policy asks about internal tool use or if the document was confidential. Check whether the policy allows putting this supplier document in a tool. What would change my mind: the document turns out to be confidential, or the memo reuses tool text. B: Disclose to the client, before delivering. The contract asks for a human translator, and the invoice says so; an AI first draft is a material difference, and silence would mislead. Say: "I used an AI tool for a first draft and I edited every sentence. The contract says human translation, so I want to check you accept this approach; if not, I will translate it myself." Do not deliver until the client agrees, or redo it by hand. What would change my mind: a written clause that allows tool-assisted translation. C: Label it, as the policy says. Add: "Commentary drafted with an AI tool and reviewed by the analyst." Tell the manager that Kavya checked the numbers and who is accountable for the report. What would change my mind: a policy exception for charts, in writing. Pass: each scenario gets a clear choice, an audience, a sentence to say, and a condition that would change the choice; A is not over-disclosed and B is not under-disclosed; accountability is named. Weak: "disclose everything" or "disclose nothing" without using the rule and the stakes.

Notice that all three used the same questions: Does a rule or contract require it? Would the other person want to know? Does it change quality, risk or what they are paying for?

Summary

• Read your employer's policy, note its gaps, and ask in writing. Policy, ethics and law are three different things.
• Disclose AI help to clients when a rule requires it, when they would reasonably want to know, or when it changes quality, risk or what they pay for.
• A tool assists, but a person stays accountable. Check before you send, so you can explain every claim.
• Assistance keeps a human in charge; replacement changes the service. Ask who owns the result and whether people can reach a human.
• Arguments about jobs, entry-level work and unequal access are real and unsettled. Keep building judgement, check who has which tools, and keep trust with colleagues by being open about how you use AI.

Next: Safety, Misuse and Sensitive Topics: Health, Money, Self-Harm and Harassment turns to the situations where a wrong or careless AI answer can hurt someone, and how to respond safely.