Skip to main content
ish has two run verbs. Both put simulated people in front of something you made and report back what they noticed. They differ in what the people do and in what comes back.
  • A study run sends people through a real surface end to end: a link, an app, a video, a PDF, an audio clip, or a chatbot. They experience it, get stuck, push on, and finish (or do not). What comes back is a per-person journey.
  • An ask puts one prompt and one or more variants of creative (text, image, video, audio, or a document) in front of a fixed panel and collects their reactions. What comes back is a per-person reaction per round, with an optional pick and ratings rolled up.
Same engine, same kind of simulated people. Different question.

​
The mental model

A study is a persistent research artifact. It carries a modality, an iteration (the URL or content people experience), and interview questions. Running it is the act of dispatching simulations against the latest iteration. People navigate, scroll, click, read, watch, or chat, and the run records the trip: where attention went, where friction showed up, what blocked completion, and the answers to your questions. Study output is a journey. An ask is lighter. It pairs a fixed panel with one prompt and a set of variants, and it asks people to react. Nobody navigates a surface; they look at the creative and tell you which one lands and why. An ask can hold several rounds against the same panel, so you can probe a winner (“now compare just A and C”) without rebuilding the audience. Ask output is a comparison: reactions, an optional winner with a confidence label, and ratings. The honest line: a study is “how do people experience this thing?” and an ask is “which of these lands, and why?”

​
When each fits

Reach for a study

The person has to do something on a real surface. Walk an onboarding flow, read a landing page, watch a 30-second spot, skim a PDF, or talk to a chatbot. You want the journey, the friction, and the completion, not just a verdict.

Reach for an ask

The person has to react to creative. Compare two headlines, three hero images, or a pair of ad cuts. You want a fast read on which one lands, the reasoning behind it, and (optionally) a pick or ratings.
The decision rule, in one sentence: if there is a surface to move through, run a study; if there are variants to choose between, run an ask.

​
Side by side

study runask run
What people doexperience a real surface end to endreact to one or more variants of creative
What it targetsthe latest (or a named) iteration of a studya new ask, or a follow-up round on one
Modalitiesinteractive, text, video, audio, image, document, chattext, image, video, audio, document variants
Audiencereuse the iteration’s panel, or resolve a fresh one with filtersfixed at creation; later rounds inherit it
Outputper-person interactions plus questionnaire answersper-person reactions per round, with optional pick and ratings
Repeat workadd an iteration, then run againappend a round to the same ask, same panel

​
How they map to the surfaces

Both verbs exist on the CLI and the MCP server with the same shape. 
# study: send people through the latest iteration of a study
ish study run -y --wait

# ask: create an ask and dispatch round one in a single command
ish ask run --new --name "tagline AB" \
  --prompt "Which sounds better?" \
  --variant text:"A" --variant text:"B" \
  --sample 30 --wants-pick -y --wait
Full flags and parameters live in the reference: study run and ask on the CLI, study tools and ask tools on the MCP.

​
Two things people mix up

A study run with no audience filters reuses whatever participants already exist on the iteration, so re-running the same iteration runs the same people again. Pass --person or demographic filters with --sample / --all (CLI) or an audience (MCP) to resolve a fresh group. An ask, by contrast, locks its panel at creation: every later round runs the same people. To compare a different audience, start a new ask.
To put a changed surface in front of people, add an iteration to the study and run it. To put another prompt in front of the same panel, append a round to the ask. Iterations carry the thing being experienced; rounds carry the next question for a panel you already have.

​
Both draw credits

A study run draws credits per participant who completes, scaled by how far they went (the draw grows with the interaction or turn count, floored at one). An ask draws one credit per participant per round, regardless of how many follow-up questions a round carries. Pre-flight failures (an unreachable URL, a refused load, a chatbot smoke-test failure) cost zero and are safe to retry; the pool ticks down only for participants that produced a result. Credits are funded by the workspace plan, so no per-dispatch approval is needed on the MCP. The CLI guards the spend differently: in a non-interactive context (the default for agents and CI), study run and ask run refuse without -y / --yes and exit 2, so spending is never silent. The full model is in credits.
Need a checkpoint before drawing credits? On the MCP, stage participants on a study with study_run(dispatch=False), then dispatch them later with audience=None. The MCP ask_run always dispatches round one, so review the panel first with person_get or person_generate. On the CLI, ish ask create --no-dispatch stages a draft ask that bills nothing; start it later with ish ask dispatch <id>.