The PMM AI Hacks Guide
The principles, tools, and workflow hacks that product marketers actually use, pulled from two live AI exploration sessions and pressure-tested on real client work.
Most AI advice for product marketers is either too abstract to act on or too tool-specific to be useful beyond a single use case. This guide is neither. It is the synthesized output of two peer-led AI exploration sessions with practicing PMMs, what they were building, what failed, what worked, and the principles that held across every tool and workflow discussed.
The prompting framework referenced throughout was built by Lance Spence, a fellow PMM, and is available as a free PDF at the bottom of this page. Everything else is drawn from live group experimentation and practitioner reflection, not best-practice lists from vendors trying to sell you software.
Three principles before any tool.
These came up in every discussion, across every tool. They are the reason good prompts produce good output and bad prompts produce noise.
Give AI one or two sentences and it will take the shortest path to a technically correct but generically useless result. It is not trying to fail you, it is optimizing for the fastest plausible answer with the information it has. More context produces better output, every time. The prompting framework in this guide exists specifically to force that context in a structured, repeatable way.
Asking AI to reason through a problem before producing a final answer, to show its work, to argue with itself, catches hallucinations, flags weak citations, and produces output you can actually stand behind in front of an executive. In Claude, the extended thinking mode does this. In other models, you prompt it explicitly. It is not foolproof, but it is the single highest-leverage change you can make to your prompting practice.
Platform access can disappear overnight. Export your frameworks, messaging documents, and strategic work to somewhere you control, Notion, local storage, your own drive. The IP you build with AI is yours. Do not leave it inside a tool you do not own.
Eight tools and what each one is actually for.
No single tool does everything well. Route work to the right tool by default rather than defaulting to the same one for everything.
"The best people using AI are not just prompting and using the spit-out. They are layering it, running through multiple tools and adding their own judgment at every step."
Eve Horne · Plankowner MarketingTwelve things to change about how you use AI today.
Goal, Who, What, Why, Structure, Examples, Think. Seven elements that turn a generic output into a usable one. The framework PDF at the bottom of this page walks through each one with examples. Copy it, paste it, fill it in before every important prompt.
Instead of "write in a professional tone," say "clear and technical, like Notion's documentation" or "direct and human, like Stripe's developer content." Named style references give AI a specific target. The output changes noticeably.
Ask AI to show its reasoning before producing the final answer. In Claude, use extended thinking. In other models, prompt it to work through the problem step by step before concluding. This surfaces hallucinated stats, weak citations, and logical gaps before they end up in your deliverable.
After getting a result, ask: "Where might this be inaccurate? What assumptions did you make that I should verify?" AI will flag its own weak spots more reliably than it will just produce solid output unprompted. Treat the critique as part of the workflow, not an afterthought.
Give AI examples of content that has already worked: a one-pager that opened sales conversations, a campaign that outperformed, a customer story that resonated. It learns the standard from what you show it, not from what you describe to it.
The first output is a draft conversation, not a final answer. Push deeper: "This is too generic, go narrower on the enterprise use case." "The CTA sounds like every SaaS product, differentiate it." The second and third prompts are where the useful work happens.
Run identical prompts through Gemini, ChatGPT, and Claude and compare outputs. Each model has a distinct personality and draws from different data sources. Once you know which handles which task better, you stop fighting the wrong tool and start routing work intentionally by default.
A NotebookLM audio summary can become Canva animation input. A Claude messaging document can become a Lovable landing page brief. The most effective AI-assisted work is a series of deliberate handoffs, with your judgment connecting them at each step.
Upload technical documentation or research materials and have NotebookLM generate a custom audio explanation. Listen on the go, interrupt to ask questions, and arrive at your first stakeholder conversation already knowing the vocabulary. Dramatically reduces hallucination risk compared to web-scraping LLMs.
In Canva AI, Lovable, or any tool that generates from combined inputs: feed one element at a time. Logo first, then color palette, then reference image. Feeding everything at once produces hard-to-control output. Incremental inputs give you more leverage over the result.
AI scrapes broadly. If you are writing positioning for a market leader, there is a real chance a competitor's talking points or even exact phrases showed up in the training data. Always read AI-generated positioning for language that feels borrowed rather than distinctly yours.
Platform access can and does disappear overnight. Export frameworks, messaging documents, and strategic work to somewhere you control. The IP you build with AI is yours. Build like it might not be there tomorrow.
You are still the talent.
AI does not know that the same word means something completely different to your sales team and your engineering team. It does not know that a competitor's talking point slipped into your positioning draft. It does not know what a specific executive actually cares about after a rough quarter. It cannot replace the cross-functional judgment, the audience instinct, or the institutional vocabulary you have already built.
Use AI to multiply what you are already good at. That is the only version of this that works long-term.
Take the prompting framework with you.
The AI Onboarding Prompting Framework was built by Lance Spence and shared with the PMM Jam community. It walks through all seven elements of a high-context prompt in a format you can copy and use immediately.
Need help applying this to your GTM?
Plankowner Marketing helps B2B companies in regulated and complex industries build the product marketing foundation they are missing. Fixed-scope engagements with defined deliverables, no open-ended retainers.