PMOpackFree check
Article

Is it safe to use ChatGPT or Claude for project reports?

It can be, with two conditions. First, know what your plan does with your data: the vendors say their business plans do not train on it by default, and their personal plans may. Second, treat the output as a draft. You check every figure, and you own the report.

Most project reports are written by hand, late, by someone tired. AI chat tools are a real help with that. The danger is not that they exist. It is that they are confident, and a tired person wants to believe them.

Step one: check what your plan does with your data

This is the part people skip. The answer depends on which product you are using, not the brand.

  • ChatGPT on a business plan (Enterprise, Business, Edu) or through the API: OpenAI says it does not use your inputs or outputs to train its models by default.
  • ChatGPT on a personal account: OpenAI says it may use your content to train, unless you turn that off in Settings, under Data Controls.
  • Claude for Work or the API: Anthropic’s Commercial Terms say it may not train models on customer content.
  • Claude on a personal plan (Free, Pro, Max): Anthropic says training is a setting you choose.

Those are the vendors’ own statements, as they read on the date in the Sources below. Terms change. Read the current page for your plan, not this article.

If you are pasting a client’s budget into a personal account, stop and check the setting. If your company has a policy, read it. If it has none, ask before you paste.

Step two: decide what you will and will not paste

A simple rule:

  1. Fine to paste: your own words, structure, headings, a plain-English rewrite of a paragraph.
  2. Think first: figures, names, dates and client names. Strip what you do not need.
  3. Do not paste: anything you would not email to a supplier. Commercial terms, HR matters, security details.

Step three: treat the output as a draft, always

AI chat tools make things up. OpenAI’s own research paper from September 2025 says plainly that hallucinations “remain a fundamental challenge for all large language models.” That is not a knock on them. It is how they are trained: the paper says the training and scoring reward a confident guess over admitting uncertainty. We go into this in Can AI make up numbers in a project report?

So the checklist before any AI-drafted paragraph goes into a report:

  • Every number in the paragraph is in your source file. Not close. The same.
  • Every date matches the plan.
  • Every name is a real person who owns the thing.
  • Every RAG colour matches the one in the RAID log and the Milestone Report.
  • Nothing has been “smoothed”. If the plan says red, the paragraph says red.

A real example

A programme lead I worked with used an AI chat tool to draft the summary slide for a monthly SteerCo. The draft read well. It said the go-live date was the end of the quarter. The Milestone Report, built from the plan, said six weeks later. The chat tool had seen an older date in a pasted paragraph and picked it. Nobody noticed until a board member asked which date was right.

The fix was not to stop using the tool. The fix was to check the pack against itself before issue, every time, and to write down who decided what when two documents disagreed.

Where PMOpack fits

PMOpack uses AI to draft the pack from your data too. It is honest to say so. What it adds is the step after the draft: fixed checks that run across the whole pack at generation time, a block on publish while a critical finding is open, and a stored record of what a person accepted and why. The checks are not AI. They run the same way every time. A person still decides. It does not prove a figure is true against your finance system, and it does not catch every error. It makes the next mismatch visible before the board sees it. For a side-by-side view, see ChatGPT compared with PMOpack and Claude compared with PMOpack.

Questions people ask

Does ChatGPT train on what I paste?
On ChatGPT Enterprise, Business, Edu and the API, OpenAI says not by default. On a personal account it may, unless you turn the setting off under Data Controls. Check your plan.
Does Claude train on what I paste?
Under Anthropic's Commercial Terms, which cover Claude for Work and the API, Anthropic says it may not. On a personal Free, Pro or Max plan, training is a setting you choose.
Can I trust the numbers an AI tool writes?
No. Treat every figure as a draft and check it against your source file. AI models can state a wrong number with confidence.
What should I never paste into an AI chat tool?
Anything you would not email to a supplier: commercial terms, HR matters, security details, and client data on a personal account.

Sources

Last checked 28 October 2026.

  1. OpenAI, “Business data privacy, security, and compliance”: “By default, we do not use data from ChatGPT Enterprise, ChatGPT Business, ChatGPT Edu, ChatGPT for Healthcare, ChatGPT for Teachers, or our API platform — including inputs or outputs — for training or improving our models.”openai.com/business-data
  2. OpenAI Help, “How your data is used to improve model performance”: for individual (consumer) accounts OpenAI may use content to train its models unless the user opts out in Settings → Data Controls or the privacy portal.help.openai.com/en/articles/5722486
  3. Anthropic, Commercial Terms of Service (effective 17 June 2025): “Anthropic may not train models on Customer Content from Services.”anthropic.com/legal/commercial-terms
  4. Anthropic, “Updates to Consumer Terms and Privacy Policy” (28 August 2025): training on Free, Pro and Max is a user setting; the change does not apply to services under the Commercial Terms, including Claude for Work and the API.anthropic.com/news/updates-to-our-consumer-terms
  5. OpenAI, “Why language models hallucinate” (September 2025): training and evaluation reward guessing over admitting uncertainty; hallucinations “remain a fundamental challenge for all large language models.”openai.com/index/why-language-models-hallucinate
  6. PMOpack behaviour statements are drawn from PMOpack’s own claims register.

See the checks on your own pack.

We are working with a small group of delivery professionals to prove the reporting journey on real work. One active workspace per person.

Or email [email protected]

For people who run project reporting for a living. Not for students, personal projects or coursework. We read every request and reply to those who fit.

Want to know where your data goes and who can see it? Read the Trust page.