Premature convergence
The room locks onto the first plausible solution and stops searching. The most obvious door gets walked through before anyone checks the others.
How might we build a gamified scoring app to keep restless kids busy at the gate?
Six design-thinking lens families
Start ideation from the right question, not the first one.
HowMightWe.ai turns a messy problem statement into a set of sharp, well-framed How Might We questions, organized by design-thinking lenses, so a team starts from diverse, unbiased framings instead of the obvious one.
No sign-up. Paste a problem statement, press Generate, read the reframes in seconds.
Leaky
How might we build a gamified scoring app to keep restless kids busy at the gate?
Valid
How might we make the gate feel less like waiting and more like arriving?
One opening, asked ten ways
How might we
Example reframes the system produces, such as: How might we make waiting feel like the start of the trip, not a pause before it?
Same problem · six lens families · never the obvious framing
Why it matters
Group ideation reliably under-performs, and the framing is usually the reason.
Before a single idea is written, the way the problem is framed has already narrowed what the room can see. Three well-documented forces do the damage.
Frame anchoring
Everyone stays trapped inside the problem's original framing, what Kahneman called frame blindness. You can't out-ideate a question that was already pointed the wrong way.
Production blocking
In a live brainstorm only one person can talk at a time, so half-formed ideas get lost waiting for the floor. Volume goes down exactly when diversity should go up.
Underneath all three
Solution leakage. The questions a team starts from often already contain the answer, like "How might we build an app that…", quietly foreclosing the search before it begins.
HOW MIGHT WE EXPLAINED
Watch one messy problem become six well-framed questions.
Not a claim about the method, but the method itself, running on one real problem statement. Step through the same pipeline the product runs: diagnose, select lenses, generate, then validate and repair.
Interactive walkthrough. HMW.ai turns one messy problem statement into well-framed How Might We questions: diagnose, select lenses, generate, validate and repair.
the how might we process
The problem
problem statement
Four of us used to eat out together every week. Then our diets diverged. One's vegan, one's celiac, one's low-FODMAP, one keeps kosher. For about two years now we just… don't. It's easier to skip it than to find somewhere that works for everyone.
It starts from a messy problem in the user's own words, not keywords, not a form to fill in.
That's the whole pipeline, running on one example.Try it with your own problem
The core idea
A question that resists leakage names a force and a target state. Nothing else.
How might we [ a force ]
(a friction, energy, emotion, or hidden assumption)
become [ a target state ]?
(a friction, energy, emotion, or hidden assumption)
become [ a target state ]?
It never names the artifact, the mechanism, the role, or the format. That single line is the difference between a question that opens the space and one that closes it. It's also the most teachable thing about the method.
Leaky
Valid
How might we make the gate feel less like waiting and more like arriving?
How it works
Diagnose, reframe in parallel, validate.
Diagnose
Read what kind of problem it is
The system reads the problem statement and works out what kind of problem it is and where its framing is stuck: the forces and tensions hiding in plain language.
Reframe
Six independent lenses at once
It generates questions across six independent lens families in parallel, so the framings stay genuinely different instead of collapsing toward one another.
Validate
No leakage, on scope, earns its reframe
Every question is checked: no solution leakage, stays on-scope, and genuinely depends on the reframe rather than restating the problem in new words.
WHY HOW MIGHT WE
This happens before the team meets. Computing many diverse reframings in parallel is something a group can't do unaided. It raises cognitive diversity at the door and short-circuits the fixation cycle before it starts.
The six lenses · design-thinking families
Six genuinely different doors into the same problem.
A
Reframe the emotion
Change how the experience feels without changing what physically happens.
B
Challenge assumptions
Surface a presupposition the problem rests on, and question it.
C
Shift perspective
Move agency or attention to a different person in the situation.
D
Change the system
Alter the structural conditions: assets, environment, timing, incentives, norms.
E
Use creative provocations
Import a frame from outside the domain: a metaphor, an analogy, an inversion.
F
Decompose and recombine
Break a tangled problem into the smaller problems hiding inside it.
What your team gets
De-risked ideation, before the meeting starts.
Diverse framings, precomputed
Six independent reframings arrive at once: the cognitive diversity a room struggles to produce live, ready before anyone sits down.
→ teams open from breadth, not the first ideaAnti-leakage by construction
Every question is validated to name a force and a target state, never a solution. The output can't quietly answer itself.
→ the search stays open where it mattersOrganized by lens, not by luck
Results land sorted across the six families, so a facilitator can see coverage at a glance and pick the doors worth opening.
→ a structured agenda, not a blank wallFrom statement to questions in seconds
No account, no setup, no facilitator training. Paste a problem statement, press Generate, and start from a stronger brief immediately.
→ usable the first time, by anyoneWhy trust it
Kickstart your brainstorm with the right questions.
Try it on a real problem you have at work — no account needed. Paste your situation in and see six well-framed How Might We questions in seconds.
01 The grammar
A valid How Might We names a force and a target state, never an artifact, mechanism, role, or format. Validation enforces that line on every question.
force + target state · nothing else
02 The taxonomy
Six distinct reframing lenses, each a different cognitive move. One coherent taxonomy, applied at machine breadth.
EmotionAssumptionsPerspectiveSystemProvocationRecombine
03 The grounding
It targets three documented failures of group ideation directly: premature convergence, frame anchoring, and production blocking. Each is countered by parallel precomputation done before the team meets.
diversity raised at the door
Questions
Before you try it.
Do I need an account to try it?
No. You can paste a problem statement and generate reframes without signing up. Your work won't be saved unless you choose to create a free account to keep it.
What exactly is a "How Might We" question?
It's a question that frames a design opportunity. A strong one names a force at play and the target state you want. It deliberately leaves the solution out, so a team can explore widely instead of building the first idea that comes to mind.
Is this just a prompt wrapper around a chatbot?
No. The method diagnoses the problem, reframes across six independent lens families in parallel, and validates every question against solution leakage and scope. The structure is the point. A single open-ended chat reliably collapses toward one framing.
Will it just hand my team the answer?
The opposite. Validation rejects any question that names a solution. You get better questions, not answers. The divergent thinking stays with your team, now starting from a stronger, less-biased brief.
Does it replace a facilitator or a workshop?
No. It sets them up. It does the parallel precomputation a room can't do live, so the session opens from diverse framings. The facilitation, judgment, and ideation remain human.
Bring a problem. Leave with better questions.
It takes one statement and a few seconds. No account, nothing to install.