The presentation is ready, the arguments hold up and the meeting could begin. Then someone asks: “Did you do this with AI?” The question may be harmless. It can also abruptly shift the debate: the focus is no longer on the work, but on the legitimacy of the person putting their name to it. By September 2026, knowing how to explain your use of artificial intelligence could become a professional skill as useful as knowing how to present a budget. Not to defend every click, but to make your work understandable, verifiable and trustworthy.
Transparency that exposes as much as it protects
Since ChatGPT arrived in late 2022, followed by the integration of generative assistants into office software suites, intellectual work has acquired a new behind-the-scenes dimension. Summarising a document, getting an outline moving, rewording a message: these tasks can now be partly delegated. But their growing technical ubiquity does not erase the social questions. If a tool contributes to the result, where does the employee’s value begin? And why do some uses seem acceptable while others immediately arouse suspicion?
The unease is not imaginary. A series of experiments published in 2025 in the Proceedings of the National Academy of Sciences highlighted a potential social cost of using AI professionally: users could be perceived as less motivated or less competent. These findings do not describe every company. They do, however, shed light on a tangible tension: asking for transparency is not enough if, in practice, that transparency becomes grounds for devaluing someone’s work.
The outlook presented here for September 2026 is therefore an analysis, not a statement of a universal norm: the more assistants become embedded in everyday tools, the more teams will need a shared language to distinguish assistance, delegation and decision-making. Without it, two bad habits could coexist: hiding AI use or disclosing it with a lengthy pre-emptive defence.
1. Describe a task rather than confess to using a tool
“I used AI” ultimately tells people very little. The phrase lumps a spelling correction together with the drafting of an entire strategic recommendation. It forces the listener to imagine what happened, sometimes assuming the worst. It is better to name the operation performed: organising notes, suggesting alternatives, generating a first draft or identifying objections.
Take an account manager preparing a sales proposal. She can explain that the assistant suggested a structure and shortened certain passages, while she selected the offer, checked the commitments and drafted the terms. This description does not seek to downplay the tool. It simply draws the line between its contribution and hers. Precision replaces justification.
2. Match transparency to the actual risk
Not every use requires the same level of explanation. Flagging every reworded internal email can create more noise than trust. Conversely, failing to disclose that an analysis used to recruit, invest or advise a client was generated by AI can prevent others from assessing its limitations. The right question becomes: does this information change how the recipient should evaluate or use the result?
Three factors help guide the decision: the consequences of an error, the sensitivity of the data and the tool’s role in the reasoning. The more significant these factors, the more explicit the explanation must be. Internal rules, contractual commitments and applicable obligations still take precedence. The aim is not to invent a personal right to remain silent, but to avoid ritualistic transparency that treats a syntax correction like a sensitive decision.
3. Show the checks, not just the intention
The phrase “I read everything through” offers little reassurance if no one knows what that review covers. Fluent writing can conceal a non-existent source, a mix-up of dates or an overly categorical conclusion. For important elements, explaining your use of AI therefore means specifying the checks performed: consulting the original documents, recalculating amounts, comparing against a reference or seeking validation from a suitably qualified colleague.
A short framework is often enough:
- Delegation: what the tool actually produced or transformed.
- Checks: the verification carried out on the decisive points.
- Responsibility: the choices you stand behind and the limitations that remain.
This method does not require sharing an entire conversation history. It makes the work open to scrutiny at the right level. Above all, it allows uncertainty to be acknowledged honestly: a check that has not been performed must remain visible, rather than being buried under a blanket assurance. Taking responsibility does not mean claiming to be infallible.
4. Respond to suspicion without becoming defensive
When faced with a disparaging remark, the temptation is to recount all the hours spent on the project. That is understandable, but rarely effective: it means accepting that the value of the work is measured by the amount of visible effort. A stronger response refocuses the discussion on the choices made and the requirements of the deliverable. What needed to be resolved? Which options were ruled out? What supports the recommendation?
In a hypothetical meeting, one possible response would be: “The tool suggested several outlines. I chose this one to address the client’s two constraints and checked the data against our documents. Which point would you like to examine?” The final question matters. It turns a sweeping judgement into a focused discussion. However, if a rule has been broken or an error identified, the problem must be addressed, not sidestepped with this communication technique.
5. Do not confuse transparency with exposing data
Explaining your process does not give you carte blanche to share prompts, source documents or raw responses. A conversation history may contain personal information, confidential business material or data about colleagues. Traceability must therefore itself be proportionate and secure. A brief record of operations and checks may be more relevant than an eye-catching screenshot.
The European framework reinforces this need for judgement. The AI Act, which entered into force in 2024, includes provisions for AI literacy measures for relevant staff, applicable since February 2025. This should not be interpreted as a general obligation to label every AI-assisted email. For teams, the practical challenge is to learn about the tools’ capabilities, risks and limitations, then translate that learning into rules suited to different situations.
6. Make explanation a shared rule
This skill cannot depend solely on individual confidence. A manager who demands productivity gains while making sarcastic remarks about AI-assisted output encourages concealment. Instead, a team can agree on permitted uses, cases that require disclosure, expected checks and spaces where people can ask questions without being ridiculed. Leaders must also explain their own uses: one-way transparency soon starts to look like surveillance.
What now? Over the coming years, the distinction between work done “with” and “without” AI could become less relevant in many professions. The distinction between work that has been checked and work that has merely been delivered should become more important. The reasonable bet, for September 2026 and beyond, is to shift the conversation: focus less on who typed each sentence and more on understanding who selected, checked and decided. That is where explanation becomes a skill rather than an excuse.
