← The Clair Blog

    ChatGPT for product managers: what to trust it with, and what to never

    Working in tech makes the AI pressure worse, not better. Where ChatGPT fits a PM's actual week, and the four ways it burns product people specifically.

    A note before the split. This is general guidance, your company's AI and data policies override it, and the tools change constantly, so treat specifics as a snapshot and keep the judgement. It holds whether your team runs on ChatGPT, Claude or an internal deployment.

    Product managers occupy the most awkward seat in the AI moment. You work at a company that probably builds or ships this technology. The engineers around you use it to write half their code, and leadership mentions it in every all-hands.

    Somewhere in that noise you are quietly unsure what it actually does for your job. The pressure to already be fluent is nowhere stronger than in tech, which is exactly what makes admitting the question feel impossible. The listicles do not help either, because they were written for someone with a vaguer job than yours.

    Here is the specific answer, and it starts from what a PM's week is really made of.

    The short answer: a product manager's job is translation, the same idea carried between engineers, executives and customers in three different languages, and translation is the one thing these tools do natively well. Trust ChatGPT with the writing between people: PRDs and user-story first drafts, the same update reshaped for three audiences, synthesis across a pile of discovery notes, and fast orientation on a competitor or an unfamiliar domain. Never let it near customer-identifiable data or the unreleased roadmap on consumer versions. Never let an invented user quote or statistic into a strategy document. And never hand it the prioritisation call, because deciding what to build is the entire job and it will happily fake the reasoning.

    The translation job

    It is worth sitting with why the fit is unusually good here before listing tasks. Most professions use these tools to produce documents. A PM mostly produces alignment.

    The raw material of alignment is the same message rewritten for different rooms. The technical detail for engineering, the business case for the exec review, the plain-language version for the customer email.

    That rewriting-for-audiences work is precisely what a language model is, which is why PMs who click with the tool tend to click hard. You supply the substance once, and it carries the substance across registers, which used to be your Thursday.

    What to hand over

    The job What that looks like
    Spec and story first drafts A rough braindump turned into a structured PRD or ticket set for your edit, with the acceptance criteria you forgot to consider flagged
    The three-audience rewrite One launch update reshaped for engineering standup, the leadership deck and the customer changelog, in minutes
    Discovery synthesis Fifteen call transcripts or a quarter of feedback clustered into themes, anonymised first, so patterns surface before the roadmap review
    Fast orientation Conversationally fluent on a competitor, an adjacent market or an unfamiliar technical concept before the meeting where it comes up
    The pre-mortem partner A roadmap or a spec argued against by something with no politics, before engineering finds the holes for you

    Where it burns product people specifically

    The invented evidence. Ask it to support a direction and it will volunteer user quotes nobody said and adoption statistics from nowhere, formatted exactly like the real thing. In a strategy document that is not a typo, it is fabricated evidence with your name above it. Every quote traces to a real call and every number to a real dashboard, or it goes. (ChatGPT made up a citation covers why this happens and what to do when it already has.)

    The customer data paste. Discovery transcripts, feedback threads and usage exports are full of identifiable customers, and the consumer versions of these tools can learn from what you paste. Strip names and identifying details before synthesis, or run it inside whatever sanctioned deployment your company has, which in tech it very likely does. The tasks you should never hand to ChatGPT has the full confidentiality treatment.

    The roadmap leak. Unreleased plans are the most leak-sensitive material a PM touches, and tech companies leak. The same rule as customer data applies, and it applies double the week before an announcement.

    The outsourced judgement. It will produce a prioritisation framework, score your backlog and defend the ranking with fluent confidence, and the reasoning underneath is pattern-matched plausibility rather than knowledge of your users, your strategy or your constraints. Use it to stress-test a call you have already made. The moment it makes the call, you have automated the one thing you are paid for.

    The line that holds

    The clean split for a PM is substance against carriage. Everything that moves information between people goes to the tool and comes back faster than you could type it:

    • The specs
    • The updates
    • The synthesis
    • The translations

    Everything that decides stays with you, checked and owned. What to build, what the evidence really says, what ships and what waits.

    Run the week that way and the fluency question answers itself, quietly, which in a tech office is the best available way to answer it. (ChatGPT for marketing managers is the sibling post for the commercial side of the building.)

    Not sure where to start?

    The right tool for the ticket in front of you.

    The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.

    Try the free finder →

    One question in, one answer out.


    Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.

    Common questions

    How should product managers use ChatGPT?

    For the translation layer of the job: first drafts of PRDs and user stories, the same update reshaped for engineering, leadership and customers, synthesis across anonymised discovery notes, fast orientation on competitors and unfamiliar domains, and pressure-testing plans before review. Prioritisation calls and evidence verification stay with you.

    Can ChatGPT write a PRD?

    It writes a strong first draft from a rough braindump, structured with sections and acceptance criteria for your edit, which converts blank-page time into review time. The substance, the trade-offs and the final call on scope remain yours, since the draft's confidence is formatting rather than knowledge of your product.

    Is it safe to put customer feedback into ChatGPT?

    Not identifiably, on consumer versions, which can learn from what you paste. Strip names, companies and identifying details before synthesising transcripts or feedback, or use your company's sanctioned AI deployment, which most tech organisations now have. Unreleased roadmap material follows the same rule.

    Can AI prioritise my product backlog?

    It can produce a scored ranking with fluent reasoning, and the reasoning is pattern-matched plausibility rather than knowledge of your users, strategy or constraints. Use it to stress-test a prioritisation you have already made, never to make one, since deciding what to build is the core of the role.

    Why does ChatGPT make up user quotes and statistics?

    Because it generates text that looks like evidence rather than retrieving evidence, and quotes and adoption numbers are trivially easy to fake convincingly. In product work that becomes fabricated support for a direction, so every quote gets traced to a real call and every figure to a real dashboard before it enters a document.

    The AI Starter Kit

    Set up ChatGPT and Claude properly. Once.

    The workflows, rules and habits that hand the boring recurring work to your AI and give you your evenings back.

    Get the Kit →