AI and the reporting tax

No one becomes a project manager to write documents about the project. The status deck, the meeting notes, the risk write-up and the sponsor update still arrive every week. Together they are a quiet tax on the hours you meant to spend on the work itself. AI drafting collects most of that time back. What it cannot do is decide whether the draft is right, and that decision is what the Certified AI Project Manager credential assesses.

Where the hours go

Look at a status report you wrote last month. Most of the time went into assembly: which tickets closed, which date moved, which decision was made in a meeting nobody minuted, all arranged in the format your sponsor expects. A smaller share went into judgment. What to lead with. How to frame a date that has slipped twice. Which risk to name and which to leave off. Assembly is the part a model does well. Paste a week of the project log and one past update into whichever AI assistant your organization allows, ask for this week’s report in the same format, and the draft arrives in minutes. The judgment is untouched, and it is the part that matters.

PMI’s Pulse of the Profession finds 82 percent of senior leaders driving AI into project workflows, so the reports will be AI-drafted either way. The open question is whether the person signing them can catch what a draft gets wrong.

Rebuild one status report

The Certified AI Project Manager brief starts here. You take one real status report of your own and rebuild it so a stakeholder reads it to the end. That sounds modest. It is harder than it looks, because a report a stakeholder finishes leads with the decision they need to make and states the date honestly. The AI draft will usually bury the decision in paragraph four and soften the date. Your job is to see that, fix it, and write down why in the change log.

That change log is one of six sections in the submission pack, alongside a cover block, the prompt set, the raw AI output, the final version and a boundary note. The raw output sits next to the final version on purpose. The assessor reads both and looks for the difference. If there is none, the pack lands in Not yet.

Sweep the risk log for what is missing

Risk logs fail in a predictable way. A list gets brainstormed at kickoff, owners stay vague, and nobody touches it again until something on it happens. The second half of the brief is a sweep. You give the model your own log and the week’s meeting notes and ask what is missing, then you list the gaps with a reason for each. The reasoning is the assessed part. “The model suggested a vendor delay risk” will not pass. “We have one supplier for the integration and their last two deliveries were late, which the log does not mention” will.

The first module of the class (45 minutes) has you write your own boundary list before anything else: three tasks AI may draft, three it may assist with a human check, and three it must never touch. Pasting a confidential risk register into a public tool may be one of the things you rule out, and saying so in the boundary note is scored under the Boundary awareness criterion. If your employer forbids using real material at all, a supplied fallback dataset is admissible, marked the same way, and declared on the cover block.

Judging the output is what gets assessed

The class is one live online day, and 300 of its 390 minutes are hands-on on your own material. The third module (75 minutes) works through six matched pairs of AI output, one weak and one strong, and asks you to name what separates them. The fourth (90 minutes) has you write a three-band rubric and judge ten outputs against it. The fifth (75 minutes) starts the brief. Those middle modules are where the reporting tax gets paid down, because reading a draft like an editor is a faster skill than writing from a blank page, and it carries over to every document on the list.

The submission is due 14 days after the last day of class. An assessor who did not teach it, and who never sees your name or employer, marks it against four criteria published before you buy: Judgment, Specificity, Boundary awareness and Reusability. The result comes back inside 10 working days as Not yet, Pass or Strong. A Not yet comes with written feedback and one free resubmission. Pass or Strong on all four criteria issues the credential, for life, with a printed Certificate ID and a digital badge that anyone can check at /verify/. The full standard is at /standards/. Classes are run by our authorized training providers, with dates at /schedule/.

Start with this Friday’s report. Take the log and one past update, and draft the report you were going to write anyway. If the reviewed draft ships faster and still reads like you, you have found the edge of the tax. The brief asks you to prove you can find it on purpose.

Sources: PMI Pulse of the Profession; American School of AI catalogue and certification standards

Frequently asked questions