The participant experience

This is what someone sees when they open your study link. Participants never sign in and never see Graham or your dashboard.

1. Welcome

The participant lands on a welcome screen with your study title, your instructions, and a consent notice (a sensible default unless you've customized it). If you've added a privacy policy URL, a "Privacy details" link appears. Clicking Begin is the consent action and starts the study.

2. Screener (if any)

If you've built a screener, those questions come next under "A few quick questions." Based on their answers they either qualify and continue, get screened out, or hit study full. (See The screener.)

3. The card sort

Participants get one of two sorting experiences, chosen automatically by device. A study's UI mode defaults to auto (phones get Mobile Sort, desktop/laptop get the traditional board); you can also force mobile or traditional for everyone, e.g. to A/B test one against the other. Both record the same data and enforce the same quality rules below, only the interaction differs.

Traditional board (desktop/laptop): cards start in a shuffled unsorted pile. They place cards by dragging a card onto a group, or tapping a card to pick it up and then tapping a group to file it. A progress counter and a "How this works" helper are always visible.
Mobile Sort (phones): first they flip through every card one at a time and can't continue until they've seen them all. Then they sort: six small stacks feed one large "active" card in the middle, tap a stack to bring a card up, tap a colored folder to file it, tap New folder to make a group, and tap any folder to peek inside or pull cards back out.

Sort types apply to both:

Open sort: they create their own groups with Add a group / New folder (groups get temporary color names until named) and must make at least 3 groups to finish.
Closed sort: they sort into your categories only, with no group creation or naming.
Hybrid sort: your categories plus, if enabled, their own.
They can move cards between groups freely. Every card must be placed before they can finish. A progress counter and a "How this works" helper are always visible.

Naming groups (open/hybrid)

When they finish, the platform suggests names for the groups they created (Graham-assisted), shown in a quick review step where they can edit before submitting. If a participant already named all their groups, they're offered the choice to keep their own names. If one group looks like a grab-bag of unrelated cards, they get a gentle nudge offering to label it "Doesn't apply to me."

4. Finishing

They see a "Thank you for taking part" screen with your thank-you message. If you set a redirect URL, they're sent there after 5 seconds (with a "click here" link to go immediately), use this to return panel participants for completion credit. If you don't set one, they see a "you can close this window" message.

Pairwise comparisons with pictures

On a pairwise study, an item can carry a picture, and the participant sees it above the item's words on each card. An animated GIF plays. The pictures for the next comparison are fetched while the participant is answering the current one, so the study does not pause between pairs, and the response time we record starts when the pictures are actually on screen rather than when we started fetching them, so somebody on a slow connection is not mistaken for somebody thinking hard. A picture that fails to load leaves the card showing its words and the study carries on. See pairwise.

Quality safeguards (automatic)

Junk group names are blocked as participants type them: names need at least 2 letters, can't be all symbols/numbers or keyboard mashing, and max 60 characters.
On submit, each response gets a quality read: an effort flag (based on timing and sorting behavior, including too-few-groups, one-giant-pile and under-minimum-time checks) and a category-quality flag, each rated ok or set aside. These never block the participant; they're signals you can filter on in analysis, and a set-aside response is left out of the scores. The middle "suspect" rating was retired: a response is either fine or it is set aside, and softer signals now appear on the participant list as reasons to take a second look rather than as a verdict. The thresholds behind all of this are set platform-wide by SortedResearch, not per study.
Finish nag: if someone tries to finish with fewer than the study's minimum groups, or with one group holding more than half the cards, they're asked to split their cards up first. They can still proceed, and the response is flagged if they do. See the study's Response quality settings.
Speeding guard: on every study type, a run of answers each given faster than the question could be read pops up a "please slow down" message the participant has to acknowledge. The third warning says plainly that it is the last. After that, a fourth run removes them from the study and discards their unsubmitted answers. This is always on, and the warning count survives a page refresh, so reloading is not a way to clear it.

A removed respondent you bought is rejected with the recruiting panel, which means they are not counted, not paid for, and their slot is freed so a replacement is recruited. They are sent back to the panel rather than to a dead end. Somebody on your own link is simply stopped, since there is nobody to tell. Reloading the page does not get them back in: a removed respondent who returns is sent straight back to the panel.

On a Findability study, giving up on a task never counts as rushing. "I can't find it" is a real answer and is always paid for, however quickly it comes.

The warning is a proper dialog: it is announced to a screen reader, keyboard focus moves into it and stays there until it is acknowledged, and Escape dismisses it. That matters because the study behind it is otherwise still answerable by keyboard, and somebody who could not see the message would have carried on answering and been removed without ever knowing they had been warned.

The bar, how many fast answers in a row trigger a warning, and how many warnings there are before removal are all set per study type in the platform's quality settings, and the guard can be switched off there for any study type. On a card sort the bar is worked out from each card's own label, so a wordy card is given longer than a one-word one.

One response per device

After someone completes a study, that browser is marked done (for ~180 days), so refreshing or returning shows a "You've already taken part" message. This is device-level rather than person-level, and it deliberately avoids blocking different people on a shared office or campus network. It can be bypassed with a private window or cleared cookies, so it's a strong deterrent rather than a hard lock. For strict one-per-person control, use unique invite links (when available).

If they leave mid-way

Answers are held in the browser only and aren't saved until each step is submitted, so closing the tab mid-screener or mid-sort loses progress on a card sort (the browser warns them first).

Findability is different. A Findability answer is saved the moment it is given, so somebody who closes the tab, takes a phone call or runs out of battery keeps everything they had already answered. Reopening the link puts them back on the task they had reached rather than at task one. On a bought audience that matters, because an abandonment part-way through is one you have already paid for.

On a Findability study the browser's Back button also means "up a level" rather than "leave the study". At the top level it still leaves, because trapping somebody on a page is worse than the problem it solves.

Rate limits

To prevent abuse, participant actions are rate-limited per network (e.g. ~30 study starts or sort submissions per minute from one IP). A real participant never hits these; bursts from one network show a brief "Too many requests, please slow down" message that clears within a minute.