The AI Shortcut Every Executive Assistant Needs for Stress-Free Project Tracking

The reason project tracking feels chaotic is rarely that there are too many projects. It’s that the data is scattered across email threads, Slack DMs, Teams messages, three spreadsheets, and a Notion page that hasn’t been updated in two weeks. AI doesn’t fix scattering. It compresses what you’ve already collected. So the trick is upstream: get every project update into one place, in one format, and then let the model produce the digest your exec actually reads.

Salesforce’s State of Sales 2025 found that high-performing teams are 1.7x more likely than underperformers to use AI agents for routine tracking. The pattern isn’t fancy tools. It’s that the high performers built a single source of truth first, then automated the reporting on top of it. The same applies to EA project work.

The project log

Pick one document. Notion, a Google Doc, OneNote, doesn’t matter. The format matters more than the platform. For each project your exec touches, set up these fields:

• Project name and owner.
• Current status (one line: “On track”, “At risk”, “Blocked”, “Done”). Date this.
• The next visible milestone with date.
• What’s blocking us, if anything, in one sentence with the person who can unblock.
• Open decisions waiting on the exec, with the deadline by which a decision is needed.
• Last update (free text, two or three sentences, your most recent intel).

The discipline is updating this log whenever you get an update, not at status-report time. Five seconds in the moment beats forty-five minutes of reconstruction on Friday afternoon.

The weekly digest prompt

Friday morning, paste the project log into ChatGPT, Claude, or Notion AI and run:

“You’re producing the Friday project digest for [EXEC]. Here’s the project log:

[PASTE LOG]

Output four sections, in this exact order:

1. Decisions waiting on [EXEC]. List each, the project, the deadline, and a one-line summary of what they’re deciding. Sort by urgency.

2. Projects at risk or blocked. For each, name the project, what’s blocking, who can unblock, and what I’m doing about it.

3. Quiet wins. Projects that hit a milestone this week with no drama. One line each, no need for detail.

4. New since last week. Anything that wasn’t on the log last Friday.

Constraints: total length under 250 words. No filler (‘all teams continue to make progress’). If a section is empty, write ‘Nothing this week’ rather than padding.

Tone: brisk, EA voice, written for an exec who’ll read it standing up before their first meeting.”

The “nothing this week” line is what keeps the digest honest. A digest that pretends every section has content is the one your exec stops reading.

A worked example, before and after

Suppose a busy week. Six projects active, two milestones met, one major escalation. A standard “summarise these notes” prompt produces a paragraph that reads like every status email ever:

“This week saw continued progress across multiple workstreams. The marketing campaign is on track, with assets in final review. The vendor onboarding has hit some delays but the team is working to resolve them. The product launch remains on schedule for Q3…”

Worth nothing. Run the digest prompt instead and you get:

“Decisions waiting (act this week): 1) Approve the revised marketing brief, blocking creative review since Tuesday. 2) Sign the vendor MSA, deadline Friday or we lose Q3 onboarding slot.

Projects at risk: Vendor onboarding (legal redline back-and-forth, I’ve put a 30-min call on Monday with [LEGAL] to break the impasse).

Quiet wins: Product launch milestone hit Wed. Internal town hall ran 11 mins under and got positive feedback.

New since last week: Talent review request from [HR] now on the log, owner TBD.”

Same data, completely different artefact. The exec reads it in 45 seconds, knows exactly what they need to do, and moves on.

The “status of the status” problem

The single biggest project-tracking failure mode in EA work isn’t bad updates, it’s stale updates. A project marked “on track” three weeks ago is not necessarily on track now, but the log will quietly carry the old status until someone questions it. The model can flag this if you tell it to. Add to the digest prompt:

“Also flag: any project where the ‘last update’ field is more than 10 days old. List the project name, the date of the last update, and a one-line draft I can send to the project owner asking for a fresh status.”

This is the prompt that turns project tracking from reactive to proactive. You catch the silently stale ones before your exec asks “how’s project X?” and you have to admit you don’t know.

The follow-up generator for chasing owners

Once you know which projects are stale, drafting six “just checking in” emails one at a time is slow. Batch:

“For each of the following projects, draft a short follow-up email to the owner asking for a status update. Voice: collegial, low-pressure, EA voice, no implication that they’re behind. Each under 60 words. Mention the project name, that it’s for the Friday digest, and ask for: current status (on track / at risk / blocked), next milestone, anything blocking.

Projects: [LIST] Owners and emails: [LIST]”

Six emails, 90 seconds, one prompt. The owners reply, you update the log, the next digest is current.

The failure mode to avoid

The model summarises whatever you give it, which means a sloppy log produces a sloppy digest, no matter how good the prompt is. Garbage in, polished garbage out. Discipline on the log is non-negotiable: updates go in within 24 hours of receiving them, statuses get re-classified honestly even when nothing has changed (a project that’s been “at risk” for four weeks is no longer at risk, it’s failing, and the digest should say so).

The other one: don’t let the model invent status. If a project’s last update field is empty, the digest should say “no update on file” rather than producing a confident-sounding sentence the model made up. The instruction “If a project has no recent intel, write ‘no update on file’ instead of inferring status” added to the prompt prevents this.

A counterintuitive observation

The most valuable section of the digest, by far, is “decisions waiting on [EXEC]”. Most EAs write project digests that lead with status (what happened) and bury the asks (what the exec needs to do) at the bottom. Reverse it. The exec reads to do their job, not for entertainment. Lead with decisions, the rest is context.

What you stop doing

You stop reconstructing status from a half-dozen email threads every Friday afternoon. The hours that come back aren’t going to a free afternoon, they’re going to the project work AI can’t do: sensing which “on track” status is real and which is wishful, and giving your exec the read they need before they ever have to ask.