AI can save time, but it can also turn a small mistake into confident-looking work. A simple review habit keeps the technology useful while ensuring that responsibility stays with the person using it.
In this article
Start with the task, not the tool
The most useful question is not “How can we use AI?” It is “Which part of this task is repetitive, slow or difficult to begin?”
AI is well suited to producing a first draft, summarising material you have already checked, reorganising notes, suggesting alternative wording and identifying questions you may have missed. It is less suited to making decisions that require professional accountability. An AI assistant does not understand your client, recognise the consequences of a mistake or know when an unusual situation deserves extra care.
| AI can help with | A person remains responsible for |
|---|---|
| Producing a rough first draft | Deciding whether the content is correct |
| Summarising supplied material | Checking that nothing important was omitted |
| Reformatting or simplifying text | Protecting confidential information |
| Generating possible approaches | Choosing the appropriate approach |
The UK Government’s AI Playbook recommends meaningful human control where mistakes could cause harm. It also advises against allowing generative AI to trigger destructive or irreversible actions without review. The same principle works for a small company: speed is useful, but responsibility cannot be automated.
Treat the prompt as information you are handing to another company
Before entering anything into an AI tool, ask where that information will go. Customer records, health information, contracts, passwords, internal financial figures and identifiable employee details should not be pasted into a public AI service simply because doing so is convenient.
Even when a business account offers stronger controls, someone should understand whether prompts are retained, whether they can be used for model training, where the information is processed, who can access previous conversations and how information can be deleted.
Anonymising information can help, but removing a name is not always enough. A combination of job title, location, age and circumstances may still identify someone. The ICO’s guidance makes clear that data protection duties still apply when AI systems process personal data.
Five checks before sharing AI-assisted work
A polished answer can feel trustworthy before it has earned that trust. Use these checks before AI-assisted work leaves your desk.
- Accuracy. Check names, dates, figures, quotations and references against a reliable source. Asking the same AI system whether its earlier answer was correct is not verification.
- Completeness. Look for what is missing. A plausible general answer may overlook an exception that matters to your client or sector.
- Confidentiality. Confirm that the prompt and output contain no information that should remain private.
- Fairness. Ask whether the output makes assumptions about people or groups, particularly in recruitment, healthcare, finance and access to services.
- Ownership. Read the final version closely enough that you are willing to put your name to it. If you cannot explain the reasoning, it is not ready.
What this looks like in a working day
Imagine a manager preparing a proposal after a client meeting. They have handwritten notes, a previous proposal and a list of services. AI could organise the notes and suggest a first structure. That may save half an hour and remove the blank-page problem.
The unsafe approach would be to upload the client’s full contract, confidential budget and personal contact details, then send the generated proposal without checking it.
A better workflow is to remove confidential information, provide a short factual brief, request a rough draft, check every service and commitment against the original records, rewrite the recommendation in the company’s own voice and obtain approval from the person responsible for the client. The AI accelerates preparation. It does not become the account manager.
A small-team AI policy can fit on one page
Responsible use does not require a hundred-page manual. A small organisation can begin with five decisions:
- Approved tools: which services may staff use?
- Permitted information: what may and may not be entered?
- Human review: who checks work before it is published or sent?
- High-risk uses: which decisions must never be delegated?
- Reporting: what should someone do if an output causes concern?
Name one person to maintain the policy and review it regularly. Responsible AI depends on processes that operate throughout the working lifecycle, not abstract principles stored in a document.
AI is most valuable when it removes friction without removing thought. Use it to begin faster, organise better and explore alternatives. Keep people responsible for facts, context, privacy and decisions.
English (UK) 
Français 