AI products have a measurement problem unique to the moment: the demo is so good it lies to you. A slick first impression gets thousands of people to sign up and try it once. Early enthusiasm spikes. And then a lot of them never come back — because "wow, neat" is not the same as "I depend on this."
The benchmark is the ordinary 40%. The discipline AI demands is refusing to let novelty inflate it.
Is the PMF benchmark different for AI startups? No — it's still 40%. The hard part is making sure the score reflects retention, not novelty.
Why "no market need" is the AI-era risk
The biggest reason startups fail is building something nobody needs — and AI has made that easier, not harder. When you can ship a product in a weekend, the temptation is to skip validation entirely and let the technology be the pitch. But cheaper building doesn't create demand; it just means more products chasing the same unmet (or non-existent) needs. A PMF score is how you find out whether you're solving a real problem or just demoing a capability. More on why discovery matters more than ever.
Who counts as an engaged user for an AI product
An engaged AI user is someone who has come back and used the product for real work, repeatedly, and trusts the output enough to act on it. First-session users are reacting to novelty. Repeat users who've folded the tool into a workflow are the only ones whose "very disappointed" means anything.
What "very disappointed" looks like for AI
For an AI product with real fit, the "very disappointed" answer sounds like "this does something I genuinely couldn't do before, or does it 10x faster, and I rely on it now." If the open-ended answers are mostly "it's cool" or "fun to play with," you have a novelty hit, not fit. The language of dependency — not delight — is what you're listening for.
See where you land against 40%
The free PMF score calculator runs the Sean Ellis survey on your users and shows your score against the benchmark — no signup.
Calculate your PMF score → Built on the Sean Ellis 40% method.How to measure PMF for an AI startup
- Survey repeat users only. Exclude first-session signups; novelty will spike their enthusiasm and lie to you.
- Watch retention alongside the score. If the survey says fit but week-2 retention is collapsing, trust the retention.
- Listen for dependency, not delight. "I rely on this" is fit; "this is amazing" is a demo reaction.
- Validate before you scale. Cheap to build means easy to build the wrong thing. A 40%+ score is your permission to pour on fuel.
AI lowered the cost of building to almost nothing. It didn't lower the cost of building something nobody needs. The PMF score is how you tell the difference before the runway's gone.
Measure your fit, find your ICP, track the trend
PMFtracker runs the Sean Ellis survey on your engaged users, scores you against the 40% benchmark, surfaces your ICP from the open-ended answers, and tracks the trend over time.
Start measuring free → Set up in 5 minutes · No credit card required
