Interbang User Guide

Interbang brings the capture, cleanup, and share loop of moderated usability testing (UT) into one workflow. This guide covers how each feature works and why we built it that way. As you build your product, we hope it helps you run UT more often and in more depth.

1. Getting Started

When you sign up, you get 5 free credits and sample UTs ready to explore. The samples are a UT for a fictional SaaS called "Beam" — fully populated with scenarios, sessions, notes, transcripts, and reports. Before you build anything yourself, you can see what a finished UT actually looks like.

The getting-started checklist in the bottom-right corner walks you through 5 milestones, from creating your first UT to comparing rounds. Finish all five to earn 3 more credits.

When you no longer need the samples, Workspace settings > Sample data clears them in one click.

2. Creating a UT

Start from the New UT button on the usability tests list. The title is the only required field, but you can expand Details to fill in three meta fields:

• Background — why this UT, why now
• Hypotheses — what you expect to learn
• Recruitment criteria — who you're inviting

Each UT contains one scenario, organized in two layers: use cases, then tasks. For example, a "Team invite flow" use case might hold three tasks: "Copy invite link," "Send by email," "Connect Slack."

Meta-field tips Multi-round UTs tend to drift from their original intent. Pinning the hypothesis down lets you check, in the round comparison view, whether the initial guess actually held up. All fields are optional, so the first run stays light.

Splitting into use cases Real UTs often validate several use cases at once. A flat task list makes group-level questions ("what was the average success rate for the invite flow?") awkward to answer. Two layers give you per-use-case KPIs for free.

After creating a UT, the Scenario tab lets you add and reorder use cases and tasks. Drag to reorder. Once any session has started, the scenario gets snapshotted at that moment — in-flight sessions keep their original structure even if you edit the scenario later.

3. Running a Session

In the UT's Sessions tab, the Add session button creates a new session. Participant name is required; age, gender, occupation, and background are optional and can be filled in later.

A session has two steps: prepare, then start

The gap between them is pre-session mode: a buffer for the moderator to set up the mic, recording, and introductions. That time doesn't count toward task measurements. Note input is locked during pre-session; pressing Start session charges the credit and unlocks it.

Preparing is free Credits are spent only at the moment a session actually runs. Re-reading the scenario or filling in a participant profile on the prep screen costs nothing. If you change your mind during prep, just delete the session.

The session screen has three panes:

• Left — scenario outline: opening / per-task guide / closing, in order
• Middle — note input: fast capture with four tags
• Right — transcript panel: drop a recording after the session ends; we transcribe it automatically

The four note tags

(⌘ is Mac, Ctrl is Windows.)

Problems and bugs also accept a severity (minor · major · critical), and problems can be filled with three structured fields: observed behavior → why it might have happened → expected behavior.

Why four tags Observations (qualitative findings), problems and bugs (action needed), and insights (improvement ideas) all go to different places afterwards. Splitting them as you capture means they're already sorted by the time you reach review and hand-off.

You can take notes the same way on the opening and closing steps, not just on tasks — handy for capturing answers to the closing wrap-up questions ("How was it overall?") on the spot. Opening/closing notes carry the same tags, severity, and transcript linking as task notes, and they appear under their own Opening notes / Closing notes sections in Copy/share. (Since they don't belong to a task, they're left out of the per-task stats and report highlights.)

Task measurement

Each task captures a 5-level success score and a 5-point participant rating.

• Success score (observer's call): independent · minor difficulty · struggled but resolved · observer helped · failed
• Participant rating (the participant's call): very hard (1) – very easy (5), based on the SEQ (Single Ease Question) standard

Why five levels Binary success/fail throws away too much information. Following Nielsen Norman Group's "self-resolved" criterion, the 5-level scale lets you distinguish a minor difficulty from a near-failure — same "success" label, very different follow-up priority.

Measuring both views The observer sees what really happened; the participant reports how it felt. The most interesting moments are when those two diverge — a fast completion the participant found hard, or a struggle they rated easy.

When you're done, hit End session at the top of the screen.

4. Transcripts

After the session ends, if you recorded it separately, upload the file to the right-side transcript panel. It's transcribed automatically within a few minutes. Keep the view open until the upload finishes; once transcription starts, closing it won't stop the job.

Files can be mp3, m4a, wav, webm, or mp4, up to 25MB. A 60-minute session recording fits under the limit as m4a (AAC) at 48 kbps or lower. wav and high-bitrate mp3 files get much larger at the same length, so check the export format for long sessions.

The transcript shows up in chronological order, and clicking a note highlights the matching time range in the transcript. If the session clock and recording clock drift apart, use the offset control at the top of the transcript to align them by ±seconds. The session start dialog reminds you to start your recording; if you started it together with the session, the offset stays at 0.

If transcription fails, a retry button appears. You can try the same file again or upload a new one.

If a completed transcript turns out to be the wrong file (or you'd rather use a different recording), use the trash icon next to the file card to delete the transcript and upload a new one. Notes stay where they are; once the new transcript finishes, they re-link by time automatically.

Why link notes to the transcript Moderators often jot just keywords during a session. In review, you want to hear exactly what the participant said — but if the notes and the transcript live in separate places, re-checking and matching them every time gets tedious. Linking them by time means one click restores the full context.

Transcription is optional. Sessions complete and reports render fine without it.

5. Reports

The UT's Report tab shows quantitative analysis and findings in one place.

Four summary cards at the top:

• Total sessions (completed / total)
• Average session duration
• Average success rate
• Average participant rating

Below them, highlights split three ways:

• 💡 Insights
• ⚠️ Problems (sorted by severity × frequency)
• 🐛 Bugs (sorted by severity × frequency)

Why split the three They each lead to a different follow-up and a different owner. Insights feed the PM's next hypothesis; problems feed the designer's improvement queue; bugs go to the engineer's tracker. Lumping them together lets the loudest item drown out the rest.

Next come the per-task stats — each task's success distribution (stacked bar), duration, rating, and note chips. Tasks are grouped by use case, with rollup KPIs in each group header.

If there are notes from the opening/closing steps, an Opening & closing notes reference section appears. These notes don't belong to any task, so they're kept out of the per-task stats and highlights — collected as reference only.

Why keep these separate Session-level notes like closing wrap-up answers are overall feedback that doesn't reduce to one task's success or failure. Folding them into task metrics would skew the numbers, so we split them out as their own qualitative reference.

At the bottom, the participants table lists each session's alias (P1, P2, …), demographics, and background. In shared views, real names are hidden automatically; only the alias remains.

6. Comparing Rounds

When you repeat a UT for a second or third round, UT detail > ⋯ menu > New round clones the scenario for the next round. The scenario, use cases, and tasks stay linked by lineage as you move from round 1 to round 2.

Once you have two or more rounds, the report tab shows a Compare rounds button. The comparison page has three sections:

• Change summary: how success rate, duration, and rating shifted, in one line
• Issue movement: problems and bugs grouped as baseline-only / comparison-only / both
• Per-use-case change: KPI deltas grouped by use case

Non-adjacent rounds compare too. If you went 1 → 2 → 3, you can still compare round 1 directly against round 3 — Interbang follows the lineage chain to match tasks across intermediate rounds, so changes don't break the link.

UT is iterative by nature A single round rarely tells the full story. Interbang is designed to be iteration-native, with round comparison built in.

7. Sharing Results

Three ways to share a UT or session result.

Export

Picking Copy Markdown from the Export menu in the report tab bundles quantitative data, notes, and transcript excerpts into one Markdown blob you can paste into ChatGPT, Claude, or any LLM. There's a 40K-character safety cap; if truncation is needed, the priority order is ISSUE → INSIGHT → general notes — general notes are trimmed first. The same menu's Download JSON gives you the full, untruncated set as a file.

Analyze with AI

The Analyze with AI button on the session review screen and the report tab goes one step further. It copies an analysis prompt plus the record to your clipboard and opens Claude or ChatGPT in a new tab — one paste kicks off the analysis. Opened from session review, it works over that single session: top problems by severity, per-task patterns, and improvement hypotheses. Opened from the report tab, it pools every session in the test, so the prompt asks for recurring problems across sessions (with frequency), patterns by scenario, and a suggested priority. The same menu's Set up MCP connection leads to the deeper integration (see AI integration below) where the AI reads your data directly, no copy-paste needed.

Why Interbang doesn't build in its own AI analysis Qualitative synthesis differs by team — the AI you use and the questions you ask are yours. Instead of bundling its own AI, Interbang structures the record so it flows straight into whichever AI you already work with. The preset prompt is a starting point; steer the follow-up conversation wherever you need.

Share link

The Create share link button first shows what the link will expose, then creates an anonymized public link. What's included depends on where you create it.

• Report (quantitative) share: quantitative metrics (success rate, duration, rating), per-use-case KPIs, and participant profiles (age, gender, occupation, background). Note bodies and transcripts are left out; notes appear only as per-tag counts.
• Session evidence share: everything above, plus that session's note bodies and transcript excerpts, verbatim. This is the most revealing share, so review the content before you send it out.
• Round comparison share: only the quantitative comparison of two rounds (note bodies stay private).

In every case, real participant names are auto-anonymized to aliases (P1, P2…), and turning sharing off expires the link immediately.

Sharing is anonymized by default UT data carries participants' words, reactions, and mistakes — it's sensitive. Without an automatic PII screen on share, a single mis-shared link could turn into a trust incident. That's why anonymization is the default.

AI integration (MCP)

Connect an AI app like Claude directly to Interbang and it reads your UT data live — no copying Markdown back and forth. How to connect and what it can do is covered in 8. AI integration (MCP).

8. AI integration (MCP)

You can connect AI apps like Claude directly to Interbang. Once connected, the AI reads your UTs, sessions, notes, and transcripts live for analysis, and can even draft the scenario for a UT you're preparing. The connection uses MCP (Model Context Protocol), an open standard, so Cursor and other OAuth-capable clients connect the same way.

Why connect to your AI Analysis gets better with context. The AI you already use holds your product context — PRDs, design discussions — while Interbang holds clean, in-depth UT data. Connecting the two lets you analyze them together, and the data stays yours.

Connecting

Start from Settings > AI integration. No tokens to generate, no config files to edit.

1. Copy the connection URL (https://interbang.app/api/mcp).
2. In Claude, open Settings → Connectors → "Add custom connector" and paste the URL.
3. Sign in with Interbang, then allow read (default) and write (optional).
4. Ask the AI in plain language.

The connection is account-wide: connect once and the AI can read UTs in every workspace you belong to — no per-workspace setup.

Why sign-in, not tokens Tokens leak easily — copied, pasted, stored in config files — and once issued, their permissions can't be narrowed. With sign-in (OAuth) you choose read or read & write at connection time, and disconnecting cuts access on the spot.

One security note — only approve connections you just started. If Interbang's consent screen appears and you didn't just initiate a connection, deny it.

What the AI can use

9 read tools — granted by default when you connect.

9 write tools — these only appear if you allowed write when connecting, and they only work on PREPARING drafts.

The data boundary is strict, too. Real participant names never reach the AI — sessions are anonymized to the same aliases (P1, P2…) as share links. Workspaces you don't belong to stay invisible.

Writing stops at PREPARING drafts

Three layers limit what the AI can write.

• Only if you allowed write when connecting
• Only in workspaces where you're a Member or above (Viewers stay read-only)
• Only on PREPARING drafts — UTs that are running, completed, or archived are read-only to the AI

So the AI can't start sessions, change a UT's status, touch the measurements or notes of sessions already run, or delete a UT itself.

Why writing stops at drafts A session's measurements, notes, and transcripts are a record of real observation. If an AI could rewrite the record after the fact, the data couldn't be trusted. Starting a session — the moment a credit is deducted — also stays a human decision. The AI helps you prepare; running the test and owning the results stay with you.

Things to ask

• Drafting — "Draft a scenario from this PRD"
• Session review — "Give me the top 3 problems from this UT"
• Numbers — "Break down per-task success rates for this UT"
• Round comparison — "Summarize what improved since the last round"
• Note search — "Find every note mentioning 'payment failed' or 'error'"

Managing connections

Settings > AI integration > Connected AI lists the apps connected right now, each with its scope badge (read-only / read & write) and when it was last used.

Disconnect cuts that app's access immediately. To use it again, you connect from scratch.

If something's off

• Tools are missing, or only some show up — your AI client may be holding on to an old tool list. Disconnect the connector and reconnect; the full set (9 read + 9 write) should appear.
• No write tools — write wasn't allowed when connecting. Check the connection's badge under Connected AI: if it says "Read-only", disconnect and reconnect, allowing write this time.
• The AI says it can't edit — check that the UT is still in PREPARING, and that you're not a Viewer in that workspace. Ask the AI to "check my access" and it reports what the connection is allowed to do.

9. Credits & Billing

Pricing model

1 session = 1 credit. Usage-based, not subscription.

• 5 free credits on signup
• 3 bonus credits for completing the 5-step getting-started checklist
• Packs when you need more: Small (5 credits) / Medium (10 credits) / Large (20 credits)

Why usage-based, not subscription UT doesn't happen on a regular schedule. One month it's ten sessions; the next, zero. A subscription makes you pay for the quiet months and turns signing up into a gamble. Pay-as-you-go means you only pay for what you actually use.

Who buys, and where credits land

Buying credits (topping up) is Owner-only. Purchased credits land in the pool of whichever workspace is active when you buy — buy while viewing a team workspace and they go to that team's pool; buy from your personal workspace and they go to your personal pool.

Deduction

• Deducted when you press Start session at the top of the session screen to start the timer — 1 credit (free credits drain first)

Expiry

Paid credits expire 1 year after purchase; free credits never expire. The credits page shows the purchase history with expiry dates so you can spot anything close to expiring. When a credit is within 30 days of expiring, the dashboard also shows an alert banner up top.

Receipts

The credits page's purchase history → Receipt column links to your official receipt. Open it whenever you need it.

Referrals

Recommend Interbang to a friend and you both earn credits. Copy your referral link from the credits page, where you can also track every reward as friends sign up and pay.

• The +5 first-purchase reward fires once, the moment a friend buys credits for the first time.
• There's no cap on referral earnings, and referral credits never expire.

Got a referral link? Sign up through the link and it applies automatically. Already signed up? Paste the link or code on the credits page within 7 days of signing up, before your first purchase.

10. Workspaces & collaboration

Everyone starts with one personal workspace, and you can create separate shared workspaces to work with a team.

Switching workspaces

The workspace switcher at the top of the sidebar moves you between the workspaces you belong to. After you switch, the UT list, credit balance, and any UT you create all reflect that workspace.

Creating a workspace

Pick New workspace in the switcher, give it a name, and you become its Owner. A new workspace starts with zero credits.

Roles

Spending a credit isn't gated by role: whoever starts a session draws 1 credit from the workspace's shared pool, so Members can run sessions too. Only refilling the pool — buying — is Owner-only.

Inviting members

In Workspace settings (sidebar), pick a role (Member or Viewer), create an invite link, and send it to your teammate. When they sign in with Google, they join automatically. Revoke a link to turn it off instantly.

Managing members

From the member list in Workspace settings, an owner can change anyone's role or remove them. Anyone can also leave a workspace themselves.

A workspace must always keep at least one owner, though. The last owner can't leave, can't be demoted, and can't delete their account while still holding sole ownership — transfer ownership to another member or delete the workspace first.

Why block the last owner A workspace with no owner has no one who can invite, manage, or delete it. So when the last owner tries to leave, we ask them to hand off ownership or close the workspace first.

Who moderated each session

In a team workspace, a UT's session list shows who ran each session — its moderator. Whoever starts a session is recorded as its moderator. (Personal workspaces don't show this, since it's always you.)

Running a session as a pair

In a team workspace, two people can share the same live session. Try running it as a pair, with one person moderating and the other taking notes.

• Conductor: the person who started the session. They run the test through the scenario and handle the timer, task measurements (success level and participant rating), and ending the session.
• Note-taker: a teammate who joins an in-progress session later. From the session list, they can open a session the conductor started with the Take notes button and take part by adding notes only.

Why we recommend working in pairs When one person moderates and takes notes at the same time, things can slip. Nielsen Norman Group recommends keeping the facilitator and note-taker as dedicated roles, and Rubin and Chisnell's 'Handbook of Usability Testing' assigns the two roles separately too.

Heads-up Two people can split the roles and join the same session, but notes don't update in real time yet. Refresh when you need the latest state.

11. Account

Open Settings from the account menu at the bottom of the sidebar (your avatar, name, and email). The settings window covers four things in one place: Appearance · Language · AI integration · Account. Signing out is in the same menu.

Deleting your account

Settings > Account > Delete account removes your account. Share links expire immediately, and connected AI apps lose access on the spot.

After deletion, your data is preserved for 30 days. After that, it's permanently removed and can't be recovered, so please be sure before you go ahead. Remaining paid credits aren't refunded.

If you're the sole owner of a shared workspace, you can't delete your account right away. So that workspace isn't left ownerless, we ask you to transfer ownership to another member or delete the workspace first.

We hope Interbang helps you run UT more often and in more depth — not just once. We'll keep refining it.