How PMs Are Using ChatGPT (and Where It Falls Short)

ChatGPT entered product teams the way most productivity tools do: someone tried it on a whim, it saved them two hours on a spec, and within a month half the team was using it. By now, most PMs have an opinion — either it's changed how they work or they tried it a few times and went back to their original process.

This is an honest look at where ChatGPT genuinely helps product managers, where it misleads them, and what the gaps are that more specialised tools fill.

What PMs Actually Use ChatGPT For

Surveys of PM communities consistently surface the same use cases:

First drafts of PRDs and specs — The most common and most cited use case. Give ChatGPT a problem description and it generates a structured document with sections, user stories, and acceptance criteria. The draft needs editing but it eliminates the blank-page problem.

Rewriting and clarifying — "Make this user story clearer" or "rewrite this acceptance criterion to be less ambiguous" are prompts PMs use constantly. ChatGPT is excellent at this kind of copy editing for technical documents.

Research synthesis — Pasting in a set of user interview notes and asking "what are the top three themes?" works surprisingly well. It's not as structured as a dedicated research tool, but for ad-hoc synthesis it's fast.

Competitive analysis summaries — Asking about a competitor's product, positioning, or known weaknesses. Results are decent but need verification — ChatGPT's knowledge has a cutoff and may be outdated on fast-moving competitors.

Preparing for stakeholder meetings — "I need to present this prioritisation decision to the exec team. What objections should I prepare for?" is a prompt PMs use to stress-test their thinking before high-stakes conversations.

Writing job descriptions, interview questions, team updates — The writing-adjacent work that PMs do constantly but that isn't core to the job. ChatGPT handles this quickly and well.

Where It Genuinely Helps

Breaking through writing paralysis

The hardest part of writing a PRD is often starting. ChatGPT's instant first draft removes that friction. Even a mediocre draft is easier to edit than a blank document.

For PMs who are good at doing product thinking but slow at translating it into docs, this is the highest-leverage use case.

Consistent structure across the team

When every PM on a team prompts ChatGPT with the same template, the output has more structural consistency than individual human writing styles produce. Teams that have developed a "PRD system prompt" used by everyone get more uniform documents with less effort.

Pressure-testing ideas

"Here's my proposed solution to this problem. What am I missing? What are the edge cases? What assumptions am I making?" — this type of prompt is where ChatGPT can be a useful thinking partner. It doesn't know your product, but it knows common failure modes and will surface them.

Where It Falls Short

It doesn't know your customers

This is the core limitation. ChatGPT has no access to your user research, your support tickets, your NPS data, or your customer interviews. When you ask it to write user stories, it writes stories for a hypothetical user based on what users of products like yours typically need.

Sometimes that's close enough. Often it isn't. The requirement that looked plausible in the doc turns out to be solving for a user behavior that your customers don't actually exhibit.

Every session starts cold

ChatGPT has no memory of your product between sessions. You start from scratch every time. Every context rebuild ("this is a B2B SaaS tool for product managers, we have three user types, our main constraint is…") takes time and never fully captures what someone who works on your product knows.

It confidently generates inaccurate specifics

Ask ChatGPT to estimate the conversion rate impact of a feature or the industry standard for a specific metric and it will give you a precise-sounding answer with no basis. It's trained to be helpful, which means it fills uncertainty with confident-sounding filler. Product managers who don't notice this end up with PRDs containing invented statistics.

It doesn't surface what it doesn't know

A good product manager asks "what am I missing?" ChatGPT is poor at identifying its own blind spots. It won't flag "I don't have information about your specific user segment, so this requirement may not apply." It just writes the requirement and moves on.

Using Claude vs ChatGPT for PM Work

Many PMs who've tried both find Claude (Anthropic's model) handles longer-context tasks better — specifically relevant when you're pasting in long interview transcripts or multi-file research notes. Claude also tends to be more willing to say "I'm not sure" on factual questions, which reduces the confident-misinformation problem.

For pure first-draft speed on structured documents like PRDs, the two are close. For reasoning tasks ("here's my situation, what should I do?") and for longer context inputs, Claude is often preferred.

The practical implication for teams: if you're processing long research documents, Claude has an edge. If you're writing shorter specs or copy edits, the difference is small.

The Better Pattern: Specialised Tools Over General AI

The PMs getting the most value from AI aren't using ChatGPT as a generic PM assistant. They're using specialised tools that are purpose-built for specific PM workflows.

For PRD generation from customer evidence: Tools like PMRead take your actual research and generate requirements grounded in what your customers said — not what generic customers typically want.

For feedback analysis at scale: Purpose-built analysis tools that understand the structure of qualitative data and can tag, theme, and quantify across large sets of responses.

For meeting notes and interviews: Transcription tools with PM-specific features like research tagging and insight extraction.

ChatGPT is best as a fallback for the work that doesn't fit a specialised tool — the one-off writing task, the quick synthesis, the stakeholder prep. It's a versatile generalist, not a deep specialist.

Getting More Out of ChatGPT (If You're Using It)

The difference between mediocre and useful output is almost entirely in the prompt. Key habits:

Give it context before the task. Don't start with "write a user story." Start with "I'm a PM at a B2B SaaS company building an export feature for data analysts. The main pain point from user research is X. Now write a user story."

Tell it what evidence you have. Paste in the relevant quotes, data, or summaries before asking it to synthesise. It will use what you give it.

Ask for reasoning, not just output. "Write these acceptance criteria and explain why you included each one" produces better output than "write these acceptance criteria."

Be specific about what you don't want. "Do not include generic metrics like DAU or NPS. Our success metric should relate to export completion rate."

Verify everything factual. Every statistic, every user count, every competitive claim should be treated as a placeholder until verified from a primary source.

ChatGPT is useful for product managers in the same way a fast, articulate assistant who's never used your product is useful. Great for structure, drafting, and thinking aloud. Poor at accuracy and judgment without context. Use it accordingly.