I am seeing few PMs use AI mindlessly to create product documentation – PRD, strategy docs, research notes etc.
The recurring problems I see (when AI not applied right) are –
1. Too long docs whose content adds no value.
A product lead produced a 50-page document for product and team plan. The actual useful content was 2-3 pages max. The doc had even things like the history of the team, very long term possibilities that don’t matter now. Many sentences said the same thing.
2. Unclear Priorities: AI throws a bunch of possibilities and may not always take a clear stand. PMs should realise that AI is asking them to take a stand 🙂 When that doesn’t happen, readers hardly get a sense of what the top 3 priorities are.
3. Adding irrelevant details – A seed stage team planning document contained details of an APM program to grow talent in-house. How is it even relevant for a seed stage company? AI added it because it took generic best practices from the internet.
4. Abstracting details from readers in the name of summary – This is by far the most irritating part of AI-generated docs. A PM created a research note detailing what he learnt from multiple customer conversations. However, there were zero customer names in the whole document. Many key details told by customers were abstracted out without any details. The final document sounded like a 19th century philosophy book instead of customer insights doc.
5. Lost evidence – Most founders and leaders want to know details on insights. AI polishes conversations and creates a summary without evidence. When left unchecked, this frustrates the audience.
6. Reinforces org biases – In one company, the internal docs were driven by stories and opinions. When people started using AI, AI naturally produced more stories and opinion pieces. Someone trying to introduce a more data-driven way of working now had a subtle push back: “But this document was generated by AI, so it must be good.” That made changing the culture even harder.
I am not proposing to get rid of AI in your document workflow. It is better to use AI-assisted thinking rather than AI-generated thinking. Use AI for structure, clarity, grammar etc. But the judgement about what matters, what to leave out, what to prioritize and what evidence is convincing still needs to come from the person writing the document (i.e. the PM). Some of the above issues would be solved through better prompting. But before that the PM needs to have clarity on what he/she wants.
Otherwise, we may save the PM an hour and waste 20 hours of everyone who reads it. Importantly end up with bad decisions because “we didn’t think through enough”
What other challenges have you noticed with incorrect application of AI in generating documentation? Please share them in the comments.
