Writing a Project Charter with AI (Then Stress-Testing It)

I used to treat the charter like a form people filled in after the real decisions were already made in hallway conversations. Sometimes that is still true. The difference now is that you can get a coherent first draft out of discovery notes in one sitting, then spend your energy where it belongs: poking holes in it before anyone pretends the project is “approved and clear.”

AI is useful for assembly. It is less useful for courage. A charter that reads well and avoids the hard tradeoffs is still a weak charter.

Start from discovery notes, not from a blank template

Feed the model what you actually have:

• Workshop notes and sticky-note clusters
• Sponsor interview summaries
• Current process pain points
• Known systems, vendors, and constraints
• Any early “must / should / won’t” language from stakeholders

Ask for a structured charter draft: problem statement, objectives, in-scope and out-of-scope, stakeholders, high-level timeline assumptions, success metrics, risks, and open questions. Keep the prompt boring and specific. Fancy language in the prompt often produces fancy nonsense in the charter.

Then read it like a skeptic, not like a proud author.

Stress-test checklist: what I look for

1. Scope that can survive a hallway challenge

If everything important is “in scope,” nothing is. I look for crisp exclusions. What are we not doing in this phase? What adjacent work will people try to smuggle in later?

• Are deliverables named in business terms, not only system modules?
• Is phase 1 thin enough to finish, or is it a wish list with a date stuck on top?
• Do out-of-scope items match the political reality, or only the happy path?

2. Constraints that are real, not decorative

Budget, regulatory windows, freeze periods, vendor lead times, data access, and key-person availability are constraints. “We value quality” is not a constraint. AI may invent tidy constraint bullets. You need to confirm which ones would actually stop work.

3. Success metrics people might measure

Charters love phrases like “improve efficiency” and “enhance experience.” Those might be directionally fine, but they are not enough. Prefer metrics you can observe:

• Cycle time for a named process
• Error or rework rate on a named handoff
• Adoption by a named user group
• Cost avoided or revenue protected, with assumptions stated

If the metric depends on data nobody owns, fix ownership before you celebrate the KPI.

4. Political risk, said out loud

This is the part templates often skip. Who loses budget, status, or control if this project succeeds? Whose process gets standardized against their preference? Which committee can stall you without ever saying no?

I ask AI to list possible political risks from the notes, then I rewrite that section in plain language with human judgment. Soft-pedaling politics in the charter does not make politics go away. It just means you discover it mid-delivery, when options are worse.

A simple generation-then-test loop

1. Generate charter v0 from discovery notes.
2. Mark every claim you cannot defend with a source.
3. Run a 30-minute stress session with sponsor and one delivery lead: what would make this fail?
4. Rewrite scope, metrics, and risks. Leave open questions visible.
5. Only then socialize more widely.

AI can help again on the rewrite: shorter version for executives, fuller version for the delivery team. Same facts. Different altitude. If the two versions disagree on scope, you still have a problem.

What AI gets wrong in charters

It may invent stakeholders who sound plausible. It may turn a vague wish into a confident objective. It may understate integration effort because the notes never mentioned the ugly system. It may write success criteria that sound measurable until you ask who reports the number and how often.

Your job is not to polish those mistakes into prettier mistakes. Your job is to catch them early.

Where this fits in delivery

A charter is a shared bet, not a novel. Keep it short enough that people read it. Strong enough that later scope arguments have a reference point. After approval, the same clarity should show up in your plan and RAID log.

I built a free Project Charter Generator for that first cut. Guided steps, live preview, Word and PDF export. Data stays in your browser. The tool page has the walkthrough and a button to open it. After the charter lands, the free project tool under Free Tools is one option for phases, milestones, and risks on smaller efforts.

I still believe good initiation saves pain later. AI just moves the bottleneck from typing the document to testing whether the document tells the truth. That is a better place for a PM or BA to spend their skill.