Guides

How account managers can use AI (without mixing up their clients)

By Rory Wilmer, InkByte · 6 min read · 22 October 2026

Account managers sit in the middle of everything. You take the client's call, turn it into a brief, chase the studio, write the status report, and then explain to the client why the thing they asked for on Monday looks different on Friday. A lot of that is writing, and a lot of the writing is repetitive. That makes it a good fit for ChatGPT or Claude, as long as you set it up so it helps rather than embarrasses you.

This is the first in a series of guides on how each agency role can use AI well. It covers where AI genuinely saves an account manager time, where it doesn't, and how to keep five clients' voices and confidential details from getting mixed up.

Where AI actually helps an account manager

The best uses are the ones where you already have the raw material and need it reshaped quickly:

• Call notes into a recap email. Paste your rough notes and ask for a short recap with decisions, actions, owners and dates. Then check every action against what was actually agreed.
• Client asks into a brief. Turn a rambling email thread into a first-draft brief with objective, audience, deliverables, deadline and open questions. The open questions are often the most useful part.
• Weekly status reports. Give it last week's report and this week's notes, and ask it to update what changed and flag what's late.
• Tricky emails. Ask for three versions of a difficult message (a delay, a scope change, a "no") at different levels of firmness, then write your own from the best parts.
• Meeting prep. Ask it to list the questions a client is likely to raise about a piece of work, so you're not caught out in the room.

In each case the model is reshaping your material, not inventing the content. That's where it's most reliable.

Where it doesn't help

Be honest about the limits. The model doesn't know the relationship history: that the marketing director hates being chased on a Friday, or that "quick tweak" means a full rewrite with this client. It can't tell what was agreed verbally, and it will write a confident recap of a meeting it didn't attend. Treat every output as a draft from a capable junior who has never met the client. Read it all before it goes anywhere, and never send an AI-written email you haven't checked line by line.

Step 1: Check what you're allowed to paste in

Before you paste a client's brief, figures or launch plans into any AI tool, check the contract, any NDA and your agency's policy. Some clients are fine with it; some forbid it outright; some only allow approved tools. If you're not sure, ask. It's a much easier conversation before something goes wrong than after.

Then check your tool's training settings. OpenAI's help centre explains that if you turn off "Improve the model for everyone" (Settings, then Data controls), your new conversations won't be used to train its models. It also says that by default OpenAI does not use content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces for training. Anthropic's privacy centre says that on consumer Claude plans (Free, Pro and Max) your chats are used to improve its models if you choose to allow it, and that by default it doesn't train on inputs or outputs from its commercial products, such as Claude for Work and the API. Both companies note that if you give thumbs up or down feedback, the related conversation may be used, so be careful what you rate.

None of that replaces the client's own rules. A setting that stops training doesn't make it fine to paste something the contract says stays inside the agency.

Step 2: One workspace per client

If you look after several accounts, don't run them all from one chat. Set up a separate ChatGPT Project, Claude Project or Gemini Gem for each client. Put three things in it:

• Account basics. Who the client contacts are and what they care about, the current projects, key dates, and any words or topics to steer clear of.
• The client's voice. Short rules and three to five real examples of their copy, so anything you draft for them to approve already sounds like them.
• Your voice. A few of your own sent emails, so the recaps and updates sound like you, not like a press release.

Keeping clients separate stops one account's details or tone turning up in another account's email, which is the mistake that does real damage.

Step 3: Save the prompts you use every week

Most account work repeats. Keep a short set of prompts you reuse, and paste them into each client's project instructions or a notes file. For example:

RECAP: Turn these call notes into a recap email from me to [client].
Sections: Decisions, Actions (owner + date), Open questions.
Use only what's in the notes. If an owner or date is missing,
write [TO CONFIRM]. Under 200 words. My usual sign-off.

STATUS: Here is last week's status report and this week's notes.
Update it. List anything late first, with the reason and the new date.
Don't change anything the notes don't mention.

The "use only what's in the notes" and "[TO CONFIRM]" lines matter. They stop the model filling gaps with plausible guesses, which is exactly what you don't want in a document the client will hold you to.

Step 4: Keep a 30-second check

Before anything goes to a client, run a quick check: Is every action and date one that was really agreed? Is anything confidential from another client in here? Does it sound like me? Would I be happy if the client forwarded it to their boss? If any answer is no, fix it by hand, and then fix the prompt so it doesn't happen next time.

The short version

• Use AI to reshape material you already have: notes, threads, last week's report.
• Check the contract and your tool's training settings before pasting anything in.
• Give each client their own project, with their voice and yours.
• Save your weekly prompts, and tell the model to mark gaps instead of guessing.
• Read everything before it's sent. You're the one who knows the client.

If you want ready-made prompts and checklists for this, the AI Workflow Kit for Account Managers collects them in one place.