Question 1

How does ish work?

Accepted Answer

Point ish at the whatever you are working on: a local build, a live URL, a Figma prototype, a video, a slide deck, a draft of copy for your next campaign, or a chatbot. Tell ish who you are building for and what you want to test. A simulated panel opens it, works through it, and reports back what they noticed, if they got stuck, what they would do next, and whether they reach the end.

You get a reported journey for each participant, with the reasoning behind every reaction, so you can understand not just what, but why.

Question 2

What can I run ish on?

Accepted Answer

Anything you would put in front of a person really. That means a local build, a Figma prototype, a draft of copy, a video, audio, an image, a PDF or a deck, a chatbot endpoint. You can upload images from your ad campaign in the metro, a video navigating through an airport or that report you spent four months preparing. Most of what people run through ish is work in progress: drafts, flows, copy, decks.

Question 3

What do I get back, and how long does a run take?

Accepted Answer

Each participants experience step by step. For each simulated person, you see what caught their attention, where the work made sense and where it lost them, the moments that landed, the points where they stalled, and whether they finished. You see their reasoning the whole way.

You also get the signals that summarize a run: sentiment, friction, and engagement, each one tied back to the moments that drove it. What you do not get is a single grade or a conversion forecast that buries the why. For interactive products, a run takes a few minutes, otherwise you will likely have your report back in under a minute.

Question 4

Can I ask the simulated audience my own questions?

Accepted Answer

Yes. Past the reported journey, you can put your own question to the simulated audience: did the headline make sense, would they trust this, what would they do next. Read their answers, then add a follow-up round to probe what they said, the way you would in a real interview. Each round is the simulated audience reacting again, so you see the reasoning behind every answer, not a real person on the other end.

Question 5

Can I compare two versions of the same thing?

Accepted Answer

Yes. Run the same study against any number of versions and read their experiences side by side: where one flow loses people that the other keeps, which headline gets noticed, where the shorter cut drops a beat. The audience stays the same across both, so you are comparing how each version is experienced, not two different crowds.

Question 6

Can I share a run with my team or a client, or come back to it later?

Accepted Answer

Yes. Your runs stay in your workspace, so you can reopen any one and read all journeys again later. To bring in a teammate or a client, create a share link. It shows the same reported journey, read-only, with no account on their end, and you can set it to expire or revoke it later.

Question 7

Who is ish for?

Accepted Answer

Designers, developers, marketers, founders, writers, creatives, knowledge workers in general - anyone making something for other people. If you have a draft and a hunch about how it will land, ish gives you a reaction before you commit. Agents can autonomously drive the same runs from the CLI or over MCP, giving agents access to realistic human feedback.

Question 8

Is any of this grounded, or is it guessing?

Accepted Answer

Grounded. ish opens the actual work you point it at and each simulated participant reacts to what is really there, not to the idea of it. You describe who you are building for, and ish grounds each one in that. The reasoning comes attached to every reaction, so you can check it against what you already know.

Question 9

How accurate is it, and what can ish not see?

Accepted Answer

ish is an approximation, built to catch confusion, friction, and dead ends before you ship. We do not publish an accuracy number, because one figure would oversell it. What it surfaces is what to fix, not a forecast of how a launch performs.

The boundary is worth naming. ish cannot see real money on the line, a specific device in someone's hand, or how people behave once the thing is live. For those you still run it in front of real people. ish is the cheap, fast read you do first.

Question 10

Does ish actually click through my whole flow, or just react to the first screen?

Accepted Answer

It drives the actual flow. A simulated visitor opens the work, moves from step to step, fills the form, follows the link, and reports what it saw at each one, not a reaction to a single screen.

Logins are handled too. Add your test credentials and ish signs in with them to run the gated flow, without ever exposing the values to the simulated visitor. The whole path comes back with the reasoning, including anywhere it got stuck.

Question 11

Can ish invent friction that isn't really there?

Accepted Answer

It can. A simulated visitor is a model reacting to your work, and now and then one will describe a button or a step that is not there. We tune against it, but no simulation is immune. So ish never hands you a verdict to take on faith. Every reaction comes with the reasoning behind it, tied to the specific work in front of the visitor, so you can read it against what is actually on the page and set aside anything that does not match. You see the account and the reasoning behind it, so the parts that drifted are yours to catch.

Question 12

Won't a simulated read just flatter me?

Accepted Answer

That is the exact failure mode we built against. ish is tuned to surface friction and dead ends, not flattery, and it samples where people struggle rather than averaging the room into something glowing.

The useful run is the one that stalls: where most people lost the thread at the same step, where a group never found the button, where a claim did not convince them. That is what ish is looking for.

Question 13

If I run it twice, do I get the same answer?

Accepted Answer

Not word for word, the same way two real people never give you the identical session. The wording shifts, but the pattern holds: the step where people stall, the part that confuses, the moment that lands.

That variation is the signal working as intended, not noise. Read the consensus across a run rather than any single line, and treat a result that shows up again and again as the one to act on.

Question 14

How is this different from real user testing?

Accepted Answer

Real sessions are slow and costly, so they tend to happen once, late, after the work is mostly set. ish is available when you need it, runs in just a few minutes and costs just a fraction. It does not replace real testing. It clears the way, gives you early signal, so the sessions you do run start past the obvious problems.

Question 15

How is this different from a survey or an analytics dashboard?

Accepted Answer

A survey collects opinions about your work. Analytics tells you what already happened after you shipped. ish does neither. It opens the actual link or prototype with a simulated panel and reports their experience step by step, so you learn where people stall before there is anything to measure.

Not a survey. Not a panel. Not another analytics tab. Realistic feedback, grounded in who you are building for.

Question 16

Why not just ask ChatGPT to role-play my user?

Accepted Answer

Ask a model to role-play and it hands you a plausible story about a user. It cannot open your live flow, click through your prototype, or watch your video. ish works through the actual thing, step by step, grounded in the audience you describe. You read the reasoning at each step and check it, instead of trusting a confident summary.

Question 17

Where do the simulated people come from?

Accepted Answer

You describe who you are building for, their goals and their situation, and ish grounds each simulated visitor in that. They are close approximations of your audience, not real people and not a panel you are renting. Your audience, ish.

Under the hood, a simulation can lean on a real consented person, on real data aggregated across many, on demographic parameters with no one specific behind them, or on a short role you sketch on the fly. Today you steer by describing the audience. The origins are how the approach is built, not a menu you pick from yet.

Question 18

Can I pick and segment the audience, or model a specific person I have in mind?

Accepted Answer

Yes. You shape who runs the study: the demographics, the goals, the situation, the segments you care about. Keep it broad, or narrow it to one kind of person, like a recruiter or a first-time buyer.

When you have a real person in mind, you describe them as a role and context rather than by name, and ish approximates from that. You are steering who experiences the work, not picking from a fixed list.

Question 19

How many simulated people run in a study, and can I set the number?

Accepted Answer

A study runs a cohort of simulated visitors, not a single persona, and you set how many. That is why the findings read as a pattern: where most of them lost the thread, where a group never found the button. Each simulated visitor draws credits, so the count is also a cost lever you control. Keep it small for a quick read, widen it when you want more coverage of who you are building for.

Question 20

Can ish simulate non-English audiences and localized copy?

Accepted Answer

You steer language and region the way you steer any other audience trait, by describing who you are building for. Set a Spanish-speaking audience in Mexico, or a German one in Berlin, and the simulated visitor reads your work in that language and reacts to the localized copy in front of them. It works best in widely used languages, and English is where we have exercised it most. We do not publish a per-language coverage list, so treat a smaller language as a rougher read, not a precise one.

Question 21

How do real people fit in?

Accepted Answer

Today ish runs a simulated audience. Live human runs are a lane still on the way: real people working through the same study alongside the simulated ones, for when you want human reactions in the loop. The two are complements. The simulated run is your fast, cheap pass on every draft. Real people are how you confirm the calls that matter most.

Question 22

What happens to the work I upload, and do you train models on it?

Accepted Answer

Whatever you upload (a link, a Figma file, a PDF) is used only to run your simulations, and we do not train models on your work. The simulation itself runs on Google Vertex AI inside Google Cloud. Running a study also uses a few specialist services, like browser automation and transcription, each listed with its location on /security. Full privacy detail is on /privacy.

Question 23

Do you have SOC 2 or ISO 27001, and where is my data processed?

Accepted Answer

We do not hold SOC 2, and ISO 27001 is in progress. Your account and study data are stored and processed in the EU, on Google Cloud in Belgium. Model inference runs on Google Vertex AI inside Google Cloud, and is not yet pinned to the EU, so some of that processing can happen in other Google Cloud regions. Nothing is sent to a third party outside Google Cloud, and we do not train on your work. Our intention is to be fully EU, and we are working toward it. A Data Processing Agreement is available on request, and the subprocessor list is on /security.

Question 24

How long do you keep my data, and can I delete it?

Accepted Answer

Account data is kept while your account is active, plus 30 days after deletion. Simulation data is deleted on request. Two things outlive that: billing records up to seven years to meet Swedish law, and analytics up to 26 months. Ask us and we will remove what we can. The full retention table is on /privacy.

Question 25

How does ish fit where I already work, in Figma, the CLI, and with agents over MCP?

Accepted Answer

ish meets you where you already work. A Figma plugin runs a prototype straight from the canvas, and the ish CLI runs studies, asks, and iterations from your terminal (it can expose a local URL for a session). For agents, ish is exposed as native tools over MCP. An agent can create a study, run it, and read the journey back inside its own loop. No editor plugin to install.

Question 26

What does it cost, and is there a free tier?

Accepted Answer

ish is free to start. You get 200 credits on signup, no card needed, enough to run your first simulations. A credit is a usage allowance you draw down, not money charged per call. The signup grant is one-time. Paid plans add credits that refill each month. Interactive and chat work on every tier. Plans are on /pricing.

Question 27

How many runs do my free credits cover, and what happens when I run out?

Accepted Answer

Your signup credits cover your first several runs. A run draws credits, not a fixed fee, and the draw scales with what you point ish at. A draft of copy is cheap, a full interactive flow costs more, and more simulated visitors draw more. See /pricing for live rates and a calculator that sizes a typical month. The signup credits are one-time and do not refill, while paid plans add credits that reset every month. If you run out, runs pause. There is no surprise charge and no overage, and everything you have already run stays put. Upgrade or add a credit pack to keep going.

Question 28

Can my whole team use ish together, and how does seat pricing work?

Accepted Answer

Yes. The Teams plan puts your team in one shared workspace, billed per editor seat at $25 per seat per month. The owner is a paid seat. You can also invite viewers with a read-only role at no cost, and viewers use no seat. Credits belong to the organization, so everyone draws from the same balance, and studies, iterations, and results sit in one place. You pick a starting seat count at checkout, and after that seats adjust automatically as editors are added or removed, prorated each time. The plans are on /pricing.

Question 29

How do I get started?

Accepted Answer

Sign up and your starting credits land right away. No call, no waitlist. Point ish at a link or drop in a file, tell ish who you are building for, and a simulated audience starts reacting in under five minutes. Prefer the terminal? The ish CLI runs the same studies, and agents reach ish over MCP.

Question 30

Can I create my own ish?

Accepted Answer

Yes. Build a profile from your own media (a voice note, some photos, your writing) and let it take part in studies other people run. Every run is logged in a ledger, so you can see where it ran and earn credits each time.

Two guardrails hold. Nobody can ever chat with your ish. It reacts to things and reports back, it is not a profile anyone talks to. And agent access only runs on terms you approve, never open access.

Question 31

If I build my own ish, how does it earn, and how do I stay in control?

Accepted Answer

Two things are real today. Every use of your ish earns credits, and every use is recorded in an attribution ledger, so you see exactly where it ran. Agent access is the part you control: nobody runs your ish until you approve the terms first.

What we do not claim yet: credits are a usage allowance you draw down on your own runs, not money paid out, and we have not built a fixed per-use rate, a cash-out, or a one-click pause. Until those are real, we will not pretend they are.

Understand anyone's experience before going live

Creation accelerated. Feedback did not.Products, ideas and content are created faster than traditional feedback loops can keep up with. ish makes human feedback available at the speed and convenience of creation.

See through someone else's eyes

No recruiting, no waiting

See which version actually lands

Run user studies from your terminal

Bring human feedback to agents

Questions & Answers

How ish works

Accuracy and trust

ish vs other tools

Your audience, ish

Your data

Workflow and agents

Pricing and plans

Create your ish

Experience your work through the people it's for