← The Clair Blog

    ChatGPT for executive assistants: what to trust it with, and what to never

    EAs are professional delegators nobody showed what to delegate. Where AI carries the load, where it drops it, and the inbox rule that protects everyone.

    A note before the split. Tools change constantly and every executive's setup differs, so treat specifics as a snapshot, your organisation's policy as the rule, and the judgement underneath as the part that lasts.

    There is a small irony in the executive assistant's position with AI. The job is professional delegation, taking an ambiguous instruction and turning it into finished reality. Yet almost nobody has shown the profession's best delegators what they might now delegate in turn.

    The advice that does exist tends to come from people who have never protected a calendar in their lives. It recommends things like "brainstorming", while the actual day is eleven scheduling threads, a briefing pack due before the 9am, and a voice note from the executive containing, somewhere, four action items and a flight preference.

    The useful answer starts from that actual day. EA work divides more cleanly than most jobs into the parts a language tool carries well and the parts where a single wrong detail is a career moment.

    Where it earns its place

    The job What that looks like
    The volume correspondence The scheduling reply, the polite chase, the decline that keeps the relationship warm, drafted in seconds and taking your thirty-second edit
    Minutes into actions Transcripts turned into decisions, owners and a recap ready for attendees, converting the 6pm typing-up into a two-minute review
    Briefing packs The meeting's attachments summarised into the three things the executive needs and the question likely to come up, so the 9am arrives prepared
    Untangling voice notes Three rambling minutes structured into a clean list of actions and preferences, possibly the single most EA-shaped job these tools do

    Across a week of scheduling threads alone, the reclaimed time is real. The pattern is the same in every row: the tool does the typing and the shaping, and your judgement stays on what actually goes out.

    What to never trust it with

    Keep it away from Because
    Logistics facts it produced itself A flight time, an address or a timezone conversion stated confidently and wrongly is the profession's worst afternoon. Every bookable fact gets verified at the source
    The executive's confidential material on consumer versions Compensation, deals, board matters and personal affairs pass through that inbox, and consumer tiers can train on what is pasted
    Messages in the executive's name, unedited The tool drafts a competent generic voice, and the executive has a specific one that recipients know. Their phrasing goes back in before anything sends
    Board-level and tone-critical communications These carry stakes a draft's confidence cannot, and they get the full human pass every time
    The two-line replies Explaining a tiny task takes longer than doing it, so the quick ones stay with your own hands, a judgement the best delegators already have

    The inbox rule

    The confidentiality stakes in EA work are unusual because they are borrowed. The sensitive material belongs to the executive, and the exposure lands on both of you. Compensation discussions, unannounced moves, personal matters, and a calendar that reveals a company's next quarter to anyone who reads it carefully all sit in the same file.

    Treat the executive's inbox as the most sensitive in the building, because it usually is. None of it goes into a consumer AI account, in the same way none of it would be read aloud in a café. There is a longer piece on the material that never belongs in a chat window, and the same logic applies with the volume dial turned up.

    The workable version is the same split as everywhere else. Anonymise anything sensitive into a genuine hypothetical if you only need help with the wording, since "declining a meeting request from a senior external contact without causing offence" carries everything needed and nothing identifying. Where the company provides an enterprise tool with proper data protections, the sensitive work lives there, and when neither is possible, the task simply stays manual.

    The line that holds it together

    The arrangement that works mirrors the one an EA already runs with the executive, because a good delegate is trusted with the volume and never with the final say. The tool gets the correspondence drafts, the minutes, the briefing packs and the untangling.

    You keep the verification of every fact someone will act on, and the executive's voice. There is a separate piece on the tells that give a draft away, worth a read before anything sends in someone else's name. You also keep the judgement about what was too sensitive or too small to hand over at all.

    Run it that way and the role's actual craft, the anticipation, the diplomacy, the knowing which meeting matters, gets more of your week rather than less. Delegating well has always meant knowing what not to hand over, and no profession arrives better trained for that.

    Not sure where to start?

    Delegation is your craft. This is the toolkit.

    The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.

    Try the free finder →

    One question in, one answer out.


    Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.

    Common questions

    How can executive assistants use ChatGPT?

    For the volume language work: drafting scheduling, chasing and declining correspondence, turning minutes and transcripts into actions and recaps, building briefing packs from meeting attachments, and structuring rambling voice notes into ordered lists. Every fact someone will act on still gets verified, and the executive's voice goes back into anything sent in their name.

    Is it safe to put my executive's emails into ChatGPT?

    Not on a consumer account, where inputs can be used for training. The executive's inbox carries compensation, deals and personal matters, so sensitive content stays out entirely, gets anonymised into a genuine hypothetical if you only need wording help, or goes through a company-sanctioned enterprise tool.

    Can ChatGPT handle scheduling and travel?

    It drafts the correspondence around scheduling and travel very well, and it cannot be trusted for the facts inside them. Flight times, addresses and timezone conversions it produces are predictions rather than lookups, so every bookable detail gets verified at the actual source before anyone acts on it.

    Will AI-drafted emails sound like my executive?

    Not without your help. The tool drafts a competent generic voice, and recipients know your executive's specific one. Use the draft for structure and speed, then restore their phrasing, their sign-off habits and their level of warmth before anything sends, which is a two-minute pass that protects both of you.

    What EA tasks should stay manual?

    The two-line replies where explaining the task takes longer than doing it, anything too sensitive to anonymise, and the judgement work itself: anticipating needs, reading the politics, knowing which meeting matters. Delegating well has always meant knowing what not to hand over, and that skill transfers directly.

    The AI Starter Kit

    Set up ChatGPT and Claude properly. Once.

    The workflows, rules and habits that hand the boring recurring work to your AI and give you your evenings back.

    Get the Kit →