Asking Questions About Your Research Data in Koji

You have run the interviews. Now you want answers. What did people say about pricing, who pushed back, how strong is the signal, what should we tell the leadership team on Thursday.

Koji does not have a chat box over your study data. It answers those questions in two places instead.

Inside Koji, the work is already done before you ask. Every interview is analysed on its own, the report opens with the answer rather than the raw material, and every claim in it links to the quote behind it. Outside Koji, the MCP connector lets Claude, Cursor or any MCP assistant read your study data and your transcripts, so you can type a free-form question in the assistant you already use and get an answer grounded in the real conversations.

This guide covers both, and where the line between them falls.

Every interview is analysed on its own

The moment an interview finishes, Koji analyses that single conversation. It produces:

• A quality score from 1 to 5. Only interviews scoring 3 or above are billed and carried into the report. See Understanding Quality Scores.
• Structured answers, per question. A scale answer is stored as a number, a choice answer as the chosen option, a ranking as an ordered list, a yes/no as a boolean.
• Themes for that conversation, each with the supporting quotes from that transcript.
• A summary and an overall sentiment for the interview.

Nothing waits on a nightly job and nothing needs tagging by hand. By the time you open the study, every completed interview already carries its own analysis, and the report is built from those analyses rather than from raw text.

The report opens with the answer

Your research report is not a pile of transcripts with a summary stapled on top. It leads with what was learned.

• Key takeaways come first: the handful of findings that answer the question you set out to ask.
• Themes follow, each with a description and an evidence count, so you can tell a pattern that showed up in 14 of 15 interviews from one that showed up in 3.
• A section per question shows how people answered that specific question: a synthesis and quotes for open-ended questions, a distribution with mean and median for scale questions, frequencies for choice questions, average position for rankings.

That covers most of what people used to want a chat box for. "What were the top pain points" and "what was the average satisfaction score" are not questions you need to ask. They are already on the page.

Every claim links to the quote behind it

This is the part that does the work a chat box usually promises.

Claims in the report carry citations. Click one and an evidence panel opens beside the text with the participant quotes that support that claim, plus a link straight into the full transcript where it was said. The charts are traceable the same way, so a bar in a distribution leads back to the people inside it.

So when the report says six participants raised concerns about pricing, you do not have to take it on trust and you do not have to go looking. You click the claim, read the six quotes, and open any of the conversations they came from. Disconfirming evidence is in the same place: a theme that only two people expressed still carries its quotes, and the quotes say what those two people actually said.

Personas and question coverage

Two more report sections answer questions people used to type out.

Personas. Koji groups participants into personas based on how they answered, and shows which themes and behaviours cluster with which group. This is where "did enterprise users think differently from small teams" gets answered, as long as you captured something that distinguishes them. Attributes from imported personalised links and from your lead form or screener are what make personas sharp, so decide what you want to compare before the interviews run, not after.

Question coverage. The report shows how well each question in your brief was actually covered across the interviews, so you can see where the evidence is thin before you build an argument on it.

The responses grid

The study's Responses view is a grid: one row per respondent, one column per question. It is the fastest way to read across everyone's answer to the same question, and rows sort by date or by quality score. You can download it as CSV from the same view.

Use it when the question is comparative and simple. "What did everyone say to question four" is a scroll, not a query.

Reading one conversation end to end

Every interview has a readable transcript, with the structured answers alongside it and the follow-ups the AI interviewer asked in place. When a claim or a quote makes you curious about the rest of the conversation, this is where you land from the evidence panel. Read three transcripts in full at the start of any study. The report gets much easier to interpret once you have heard how the conversations actually go.

Free-form questions: the Koji MCP connector

For anything the report does not already answer, and for questions that span more than one study, connect Koji to the assistant you already use.

The MCP connector is at https://www.koji.so/api/mcp/mcp and authenticates over OAuth. Add it to Claude, Claude Code, Cursor, or any client that supports the Model Context Protocol, sign in once, and the assistant can read your research. The full setup is in the MCP overview.

Four tools do the reading:

Then you just ask. Questions that work well:

• "Read my churn study and tell me which participants disagreed with the rest."
• "Compare what people said about pricing in these two studies."
• "Draft a two paragraph summary of the main finding for our leadership update."
• "Pull the three quotes that best illustrate the onboarding friction theme."
• "Across my last four studies, has anyone asked for a shared workspace?"

The assistant answers from the transcripts and study data it actually retrieved. You can check any quote it gives you by opening the same interview in Koji, and the report's own citations are the fastest way to do that.

Structured questions make the numbers exact

When your study includes structured questions, both the report and the assistant have typed values to work from instead of impressions.

Combining a typed number with the probing follow-up that explains it is what separates this from a survey tool. You get the distribution and the reason for the distribution in the same interview.

Taking the answers somewhere else

There is no chat transcript to save, so the ways out of Koji are the ordinary ones:

• The public report link. Publish the report and share the link with anyone. It is the thing to paste into Slack, Notion or an email.
• CSV and JSON export of interviews and responses, for a spreadsheet or your own analysis. See Exporting Research Data.
• The MCP connector, when you would rather have an assistant read the data and write the summary for you.

What this does not do

Being clear about the edges:

• There is no chat box inside Koji, and no cross study search over your transcripts inside the app. Cross study questions go through an assistant over MCP.
• Nothing here reaches outside your research. An assistant reading your Koji data answers from your interviews, and if a topic was never discussed there is nothing to retrieve.
• Small samples are still small samples. Three interviews cannot carry a claim about a population, whichever surface you read them in.
• None of it replaces judgement about whether a study was designed to answer the question you are now asking it.

Availability and plans

The report, the evidence panel, the responses grid, export and the MCP connector are available on every plan, including free. All features are granted on every plan, so nothing here is locked behind a tier.

New accounts start free with 10 credits and no card. After that it is pay as you go, as low as €1 per qualified interview.

Related Resources

• Koji MCP Integration Overview - Connect Claude, Cursor or any MCP assistant to your research
• Reading Your Research Report - Every section of the report and how to read it
• Exporting Research Data - CSV and JSON export
• Structured Questions Guide - The question types that give you exact numbers
• Understanding Quality Scores - Why only interviews scoring 3 and above reach the report
• Understanding Themes and Patterns - What the analysis extracts from each interview

Further reading on the blog

• Best AI Market Research Tools in 2026: The Complete Buyer's Guide: AI has fundamentally changed market research. This guide compares the leading AI market research platforms, from AI-native interview tools li
• How AI Is Disrupting Market Research Agencies in 2026 (And What to Do About It): The $150B market research industry is being rebuilt around AI. Three of the top five firms today are technology companies, 83% of practition
• B2B Customer Research: The Complete Guide for Product Teams (2026): B2B customer research is harder than B2C. You are navigating buying groups of 10+ stakeholders, gatekeepers, and enterprise procurement cyc

<!-- further-reading:blog -->