AI for project managers: 5 prompts and 4 PM tools' AI limits

What can AI do for a project manager?

Microsoft's Planner Agent writes a whole status report from plan progress, but only for people with a Microsoft 365 Copilot license. Here's where each job runs as of October 2026, inside the PM tool or in a general assistant, and the check you run before anything goes out:

Project managers took to this fast. PMI surveyed 500 project professionals who already used generative AI, first in December 2023 and January 2024, then again in March and April 2024. The share using it on more than half their projects, the group PMI calls Trailblazers, rose from 20% to 37% in those three months (PMI, Pushing the Limits, September 2024). PMI's own summary: "of the project professionals we surveyed who utilize GenAI, two of every five use it in more than half of their project work."

The same report says light users start with "communication and reporting", and heavy users push on into planning, budget and cost, resources and decision support. That order matches the table. A status report is the safe first job, since every line in it can be checked against the board.

Heavy users report more gains. In PMI's March 2025 follow-up, 48% of Trailblazers reported advanced productivity gains, against 3% of the lightest users (PMI, The Project Professional's GenAI Journey). There's a cost too. PMI's November 2025 study of 2,000 project professionals found advanced users report collaboration problems at nearly twice the rate of basic users.

How much AI do Asana, monday, ClickUp and Jira include?

Each vendor meters the AI on its cheapest paid plan in its own unit. We read each pricing page on 12 October 2026 and counted what a team of 10 gets a month in each vendor's cheapest metered AI action. The counts differ 40 times over. A Rovo Chat answer does more work than a ClickUp AI Field fill, so read it as what a quiet month buys, not as a quality score:

Jira's number is the one to remember. Atlassian's Rovo credits page says each user message in chat is a billable event at 10 credits, so 25 credits a user is two and a half answers per person per month. Credits pool across the organization, and Atlassian says billing for extra usage applies from 3 December 2026. Asana's cap works per account. A team of 30 on Asana Starter still shares 50 requests, under two each.

Meetings get priced too. monday's AI feature catalog lists its Notetaker at "120 credits per meeting hour", so the 2,000 credits on monday's Standard plan ($12 a seat) cover about 16 hours of recorded meetings a month for the whole account. That's before anyone runs any other AI.

Some built-in AI costs nothing extra. monday lists its Updates summaries as "Does not consume credits", and Risk Insights is credit-free too, but only on the Enterprise plan (monday Support). Asana includes AI Assists, its in-project drafting and summaries, on every paid plan. Free plans are thinner: Jira Free has no Rovo, monday's AI starts on Basic, and ClickUp Free Forever gets 500 credits for the whole workspace.

Should a project manager use the PM tool's AI or a general assistant?

Use the PM tool's AI for jobs that need live board data, and a general assistant for jobs that combine sources or need a long, careful draft. Microsoft's Planner Agent builds its report from progress, milestones and upcoming tasks it can already see (Microsoft Support). A chat model can't do that unless you paste the export. It can, though, take the Jira export, the budget sheet and the sponsor's last email in one go, and you pick the model.

Connectors blur the line. Claude's connector directory lists Asana, Atlassian (Jira and Confluence), monday.com and ClickUp as of October 2026. Asana's own AI page says its connectors and MCP link ChatGPT and Claude to its work graph. A connector reads live tasks, so the paste step goes away, but it also spends the assistant's usage limits on reading your board.

Where the paste goes matters more than which model reads it. Asana's product terms say it "does not use or permit third-party AI partners to use Customer Data to train generative artificial intelligence." OpenAI says it may train on content from its services for individuals unless you turn off "Improve the model for everyone", and it doesn't train on ChatGPT Business, Enterprise or Edu (OpenAI Help Center). Client names, salaries and contract values belong in neither until your company approves the tool. What never to paste into an AI chat covers the redaction step.

Whizi sits on the general assistant side. A Whizi project holds up to 10 pinned files (PDF, CSV, .xlsx or .docx, 10 MB each) and 32,000 characters of instructions, so your status template and RAID log ride along on every chat in it. On Pro ($29.99 a month), side by side model chats run one risk prompt through Claude and GPT at once. Where it loses: Whizi has no Jira, Asana or monday connector and no meeting notetaker, so it reads only what you paste or upload, and it has no team accounts.

Which AI prompts should project managers start with?

Start with the five prompts below, one per job in the first table. Each one works only from what you paste and marks what it couldn't confirm. None of them sets a date, a budget or a risk score for you. Those are yours to commit to, and a model writes a wrong date as calmly as a right one.

The status report. Paste the export and ask for the sponsor's format, not a summary:

```

Write this week's status report for [project] from the task export and notes below. Sections: Overall status (Green, Amber or Red, with one sentence on why), Done this week, Next week, Risks and issues, Decisions needed. Use only facts from what I paste. If a date, percent or owner is not in it, write [DATE NEEDED] or [OWNER NEEDED] instead of guessing. Under 250 words.

Task export: [...]

My notes: [...]

```

The risk log. Ask the model to find risks and leave the scoring to you:

```

From the project brief below, list 10 risks for a [type] project due on [date]. For each give: a one-line risk statement in "If X, then Y" form, the cause, an early warning trigger, and one mitigation. Leave likelihood, impact and owner blank for me to fill. Flag any risk that depends on an assumption I did not state.

Brief: [...]

```

Meeting notes. It's the job a model does best, if the transcript is in front of it:

```

Below is a meeting transcript with external names replaced. Give me: decisions made (one line each), actions as a table with action, owner and due date, and open questions. Only list an owner if that person agreed in the transcript; otherwise write UNASSIGNED. Quote the line that supports each decision.

Transcript: [...]

```

The work breakdown. Ask for dependencies, because that's where a plan breaks:

```

Break this goal into phases and tasks for a team of [roles]: [goal], due [date]. For each task give a short name, the role that owns it, what it depends on, and a rough effort in days marked ESTIMATE. List the three tasks on the critical path. Do not add scope I did not mention.

```

The sponsor email. One decision per email gets answered:

```

Write a 120-word email to [sponsor] about [project]. Status is [Amber] because [reason]. I need one decision: [decision], by [date]. Plain English, no exclamation marks, no new facts beyond what I wrote here.

```

The AI workflow templates page turns the meeting notes prompt into a full notes to action plan workflow. For product specs and feedback theming, the AI for product managers guide covers the other PM role.

How do you check an AI status report before it reaches a sponsor?

Check every date, percent and owner against the board, then rewrite the overall status line yourself. The model's RAG color is a guess built from whatever you pasted. Yours is a call you'll defend in the meeting.

1. Search the draft for [DATE NEEDED] and [OWNER NEEDED] and fill each from the board.
2. Check every date and percent against the live tool, not the export you pasted on Monday.
3. Read the risks section against your risk log. A risk the model added is a question for the team, not a fact.
4. Write the Green, Amber or Red line yourself.
5. Keep the board as the record. The chat is a drafting pad.

Models do invent details, and AI hallucination statistics has the measured rates by model. Asana's and monday's built-in drafts read the same task data you do, so they get dates right more often. They still can't tell why a task slipped. You can.