How can AI be applied to project management?
AI is useful wherever a project manager turns messy input into a structured document that someone will then check. That covers a large part of the job: notes, logs, reports, plans and messages. It is far less useful for the parts that depend on knowing the people, the politics and the real constraints of the organisation.
These are the tasks a general assistant such as ChatGPT, Claude, Gemini or Copilot can help with today, using text you paste or files your company allows you to share:
Meeting notes to action log
Extract decisions, actions and open questions from notes or a transcript, with an owner and date only where the notes actually state one.
Status report first draft
Turn the action log and milestone list into a report for a named audience, with the RAG status you have already set.
Risk and dependency review
Ask the assistant to challenge a plan: which tasks depend on something outside the team, what happens if one supplier is late, which assumptions are untested.
Work breakdown first pass
Draft a breakdown for an unfamiliar type of project, then correct it with the people who will do the work.
Change request summary
Summarise what a change request asks for and list the questions it raises for scope, cost and schedule, without deciding the answer.
Stakeholder messages
Rewrite the same update for a sponsor, a delivery team and a client, keeping every fact identical.
Lessons learned synthesis
Group retrospective comments into themes, keeping the original comments linked so nobody's point is lost or reworded.
What goes wrong when AI turns meeting notes into actions?
The most common failure is that the assistant fills gaps. Notes are vague about who will do something and by when, and a model asked for a tidy action log will often supply a plausible owner and date. That makes the log look finished when the meeting was not.
A fictional example. Harbour Lane Studio, a small design agency, is moving office. The project manager pastes these notes: 'Lease signed, keys 3 March. Someone needs to chase the removals firm about the quote. Priya thinks the server cabinet needs an electrician. IT cut-over date not agreed. Budget for furniture still with the directors.'
The prompt: 'Turn these notes into an action log with columns for action, owner, due date and status. Leave owner or date as TBC unless the notes state them. List anything undecided separately as an open question.' Without the second sentence, the first draft below is the kind of result you get.
| From the notes | Loose AI draft | Checked log |
|---|---|---|
| Someone needs to chase the removals firm | Priya to confirm removals quote by Friday | Chase removals quote. Owner TBC. Date TBC |
| Priya thinks the cabinet needs an electrician | Electrician booked for server cabinet | Confirm whether an electrician is needed. Owner: Priya to advise |
| IT cut-over date not agreed | IT cut-over 3 March | Open question: agree IT cut-over date |
| Furniture budget still with directors | Furniture budget approved | Open question: directors to decide furniture budget |
How do you check an AI action log before sending it?
Read the log against the notes line by line and ask one question of each entry: does the source say this? In the loose draft, three of four lines say something the meeting never agreed. Priya was given a task she never accepted, an electrician was 'booked' when someone had only raised the possibility, and a date was copied from the keys line to the IT line because it was the only date in the notes.
A useful habit is to ask the assistant to show its working: 'For each action, quote the sentence in the notes it came from.' Any action without a quoted sentence is an invention. This does not make the model reliable, but it makes the check faster, and the project manager is still the one who reads the quotes.
The NIST Generative AI Profile describes two risks that meet in this example: confabulation, where content is confidently stated but false, and automation bias, an excessive deference to automated output. A neat table invites exactly that deference.
How do you use AI for a status report without softening bad news?
Set the status yourself, give the assistant the facts, and tell it not to change the rating or the wording of the risks. Models tend towards reassuring language, so an amber project can read as green by the third paragraph.
A prompt that works: 'Draft a one-page status update for the project sponsor. The overall status is AMBER because the removals quote is late and the IT date is undecided. Do not change the status or describe the risks more mildly than I have. Put the two decisions I need from the sponsor at the top.' Then check that the decisions are at the top, the status is unchanged and nothing has become 'on track' that is not.
Want to see how this connects to a role?
Explore the planned team trial ↗Which project decisions should AI not make?
Anything that commits the organisation or allocates people. That includes agreeing dates with a client, accepting a risk, approving a change request, setting a budget and deciding who works on what. AI can lay out the options and their consequences; the decision and the accountability belong to a person.
Allocating work to individuals deserves particular care. The EU AI Act lists, among high-risk uses in Annex III, AI systems intended to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate the performance and behaviour of people at work. A project manager asking a chatbot to draft a plan is not running such a system, but feeding team members' performance data into a tool and asking it who should do what moves towards that territory. This is general information, not legal advice.
Estimates sit in between. An assistant can suggest a range for an unfamiliar task, but it does not know your team's speed, your client's approval habits or your supplier's track record. Treat its estimate as a question to put to the people who do know.
Which AI tool is best for project management?
There is no single answer, and we have not tested or ranked project management tools. It helps to separate two categories. General assistants such as ChatGPT, Claude, Gemini and Copilot work on whatever text you give them. AI features built into project management software work on the tasks, comments and dates already stored there.
Choose based on where the project information lives and what your company allows. If the plan is in your project tool and that tool has an approved AI feature, it can save copying and pasting. If the work starts in notes, emails and documents, a general assistant is often enough. In both cases the skills on this page are the same: give clear input, ask for traceable output, and check before anything is sent.
Can AI replace a project manager, or make PMP certification irrelevant?
AI changes how much of a project manager's time goes on drafting, but the parts of the job that hold a project together, such as negotiation, trade-offs, escalation and keeping people committed, are not text tasks. We do not make predictions about the profession.
On certification: a project management credential records knowledge of a body of practice, and only the awarding body can say how it is changing. Check the current exam content and guidance from the body itself. A Tosheo.ai certificate is a different thing: the planned recognised certificate records completion of a practical AI-at-work path and assessment, and it is not a project management qualification.
How should project managers practise using AI?
Use your own past material, with anything sensitive removed, so you already know the right answer.
Pick a finished meeting
Take notes from a meeting whose outcomes you remember, remove client and personal details, and ask for an action log.
Mark every invention
Highlight each owner, date or decision the notes did not contain. This shows you how your assistant fills gaps.
Rewrite the prompt
Add the TBC instruction and the 'quote the source line' request, and run it again. Compare the two logs.
Draft one status report
Set the status yourself and check whether the draft keeps it. Note any phrase that softened a risk.
Agree team rules
Decide which project documents may go into which tool, and that every AI-drafted log is checked against its source before it is shared.
When is this guide not the right answer?
If you need to choose project management software, build schedules with resource levelling, run quantitative risk analysis or prepare for a formal project management exam, look elsewhere: vendor documentation, a scheduling specialist or the awarding body's own preparation materials.
Tosheo.ai is being built to teach people to use general AI tools in their everyday work through short lessons and graded practice tasks. Project managers would sit in the planned Operations or General paths, or a custom role on the planned Growth plan. It does not plan projects for you, integrate with project management software, or award a project management credential.
Sources and scope
NIST AI 600-1, Generative Artificial Intelligence Profile, July 2024 ↗