Here is a test worth running today. Open ChatGPT, Claude, Gemini, or Perplexity and ask it to explain your product, who it is for, and how it compares to your top competitor.
Read the answer closely, because a growing share of your buyers are reading some version of it too. They are asking an AI to summarize the market, shortlist the options, and tell them who fits their situation, all before they ever land on your website or talk to a human on your team.
Picture a real one. Say you sell Acme, a project management platform built for agencies. A buyer asks the model what Acme is and who it is for. If your positioning is sharp, the answer comes back with your category, your buyer, and your edge. If it is fuzzy, you get something like "Acme is a software tool that helps teams stay organized," a description that fits a thousand products and sells none of them.
That changes the job of product marketing. Not the fundamentals. The stakes.
The AI is now the first analyst your buyer talks to
For years, positioning had a forgiving reader. A human visitor would land on a messy homepage, do some charitable interpretation, and piece together what you do. Buyers gave you the benefit of the doubt.
A model does not. It repeats what is clearly stated and fills every gap with whatever is most common across its sources, which is often a generic category description or a competitor's framing. Ask about a tool with fuzzy positioning, and you get "Acme is a platform that helps teams work more efficiently." Technically true, instantly forgettable, and impossible to choose. The model did not fail. It just returned the average of everything you left unsaid. If your positioning is fuzzy, the machine does not guess kindly. It defaults to average.
So the belief you were trying to build in your buyer's mind is now being drafted, first, by an intermediary you do not control. The job of product marketing is to make that intermediary repeat you accurately.
Why this is a positioning problem, not an SEO problem
It is tempting to hand this to the SEO team and call it a technical fix. It is not. What an AI says about you is your narrative, compressed. And compression is brutal to weak positioning.
Vague superlatives get stripped. "Acme is the leading, innovative, best-in-class platform" collapses in a summary into just "a project management tool."
Specific, verifiable claims survive. "Acme is built for agencies that bill by the hour, with time tracking and profitability reporting in one place" is something a model can hold onto and repeat.
Contradictions get hedged. If your site, your LinkedIn, and your review profiles describe you three different ways, the model picks the loudest source or refuses to commit. Neither helps you.
None of that is fixed with keywords. It is fixed with sharper positioning, which has always been product marketing's job. AI just made the cost of doing it badly impossible to hide.
How I approach it
I have been leading the shift from classic search optimization to generative engine optimization, and the work is far more product marketing than it is technical. A few principles I keep coming back to:
Write positioning a machine can repeat. If a smart reader cannot restate what you do in one clear sentence, neither can a model. Clarity beats cleverness. Say the category, the buyer, and the difference plainly. "Acme is a project management platform for agencies that bill by the hour" hands a model the category, the buyer, and the hook in one line. "Acme helps modern teams do their best work" hands it nothing.
Treat the model as a persona you enable. You would never send a sales rep into a deal without a battlecard. The AI is in the deal too now, earlier than the rep. Give it the same clear, honest comparison you would hand a seller.
Own the comparison narrative. Buyers ask "you versus them." When someone asks "Acme vs Asana," the model will answer whether or not you gave it anything to work with. Publish the honest comparison yourself, or let a competitor's page become the source it quotes. Silence is not neutral here.
Feed proof, not adjectives. Models reward evidence that survives summarization: specifics, numbers a buyer can check, named use cases. Adjectives evaporate.
Be consistent everywhere the model reads. Your site, your profiles, third-party reviews, and credible mentions all get triangulated. Say the same true thing in all of them, so the model has nothing to hedge against.
The good news
If this sounds like the discipline you should have had all along, that is the point. Generative engine optimization does not reward tricks. It rewards exactly what strong product marketing always rewarded: a sharp category, a clearly named buyer, a real difference, and the evidence to back it.
What is new is that a machine now sits between you and your buyer, and machines do not do charity. Fuzzy positioning used to cost you a little conversion. Now it costs you the entire first conversation, because that conversation happens without you in the room.
So run the test.
Ask the AI what you do.
If the answer is not the story you would tell, you do not have a search problem. You have a positioning problem, and that is the most solvable problem in marketing.