Introducing Raconte

This article was not written by an AI left alone in front of a blank page. It was written from an interview. Someone asked me questions, I answered out loud for a few minutes, and the transcript of that conversation became the raw material. The tool that ran the interview is Raconte. It is also the subject of this article.

It all started with a LinkedIn post

At first, I simply wanted to write my LinkedIn posts faster. But I did not want the AI to write everything for me. The most important part had to come from my own brain: my experiences, my feelings, my opinions. An AI cannot make that up. And one of the best ways to avoid AI slop is precisely to go and get the information from human brains.

So I built a first version of Raconte. It interviewed the user, then drafted a post. It let me work on the whole voice side, and above all it helped me see where the real problem was: information retrieval. I realized it would be far more interesting to do it not just for myself, but to interview other people and collect their knowledge.

Generation, in the end, is something you may want to keep in your own tools. Claude Code, ChatGPT, everyone has their own preferences of tool and prompt. So I steered a second version toward something less focused on generation and more on collecting and structuring information, as a list of typed, prompted properties. But it turned out a bit too complex to use, and once again it pulled me away from the original mission: getting the information out of people’s heads.

Hence this third version, now online, fully centered on the interview.

One prompt to write it, one link to share it

The principle is deliberately simple. You describe your interview in a prompt, a single exchange is enough. You optionally adjust a few parameters: the title, the language, the introduction message. Then you share a link.

You can send that link by email, by SMS or as a QR code. The person opens the conversation in their browser, with nothing to install, and answers whenever they have a moment. That is what makes the tool so versatile: the same link can question a customer, a colleague, a candidate or a user. If you want to see what a first end-to-end interview looks like, the getting started guide walks through it step by step.

The real problem is AI slop

AI slop is the kind of content AI produces when everyone leans on the same models. Fairly classic turns of phrase, and above all generic information that anyone can get with the same prompt. The output all looks alike, and tells you nothing you did not already know.

To fix that, you need two things: good prompting, and above all authentic or original information. Something to chew on, that gives the AI the raw material to produce genuinely interesting content. That raw material exists nowhere in the training data. It lives in the heads of your experts, your customers, your teams. You still have to go and get it.

This is also what makes the difference for SEO. Search engines, like AI-generated answers, reward original, first-hand content, the kind you cannot find anywhere else. Publishing what anyone can generate with the same prompt does not set you apart. Publishing the unique expertise of your people does.

Talk, don’t type

A form is not enough. It freezes the questions in advance and collects short answers, dashed off in a hurry. An interview, on the other hand, follows the thread of a thought.

	Form	Raconte interview
Questions	Fixed, identical for everyone	Adapted live to each person
Follow-ups	None	The AI digs when an answer deserves more
Answer format	Short text, often rushed	Voice, natural and detailed
Material collected	Generic	Authentic and original
Effort for the respondent	High (they have to write)	Low (they just talk)

Non-deterministic interviews

This is where Raconte parts ways with a plain questionnaire. You prompt an interview: you describe questions, or you let the agent ask its own. You give objectives, instructions, limits. The agent then asks its questions and reacts to what the person says, following up where it matters, until it gets the most useful information possible.

These are non-deterministic interviews. Nothing like a classic form with well-defined fields. The questions can vary enormously from one person to the next. The agent simply follows its instructions and its objectives, the way a good journalist would: it listens, it bounces back, it goes after the nuance instead of the checkbox.

From voice to insight

Once the interview is done and the answers are in, you retrieve the transcripts. In the interface, or through the API, the MCP server, the CLI or a webhook. From there it is yours to reprocess: sentiment analysis, extraction of concrete actions, writing authentic and interesting articles.

But Raconte does not leave you in front of a wall of raw text. Every interview comes back with a first layer of insight, generated automatically at the end: a summary of what the person actually said, and a per-message sentiment analysis. Nothing to configure, it is automatic. To wire all of this into your agents, the MCP server gives them direct access to interviews and transcripts.

Built to plug into your tools

Around the interview, I built a whole set of tooling: an SDK, webhooks, a skill and a CLI. The idea is simple. You should be able to configure agents or systems that go out, interview other people by voice, collect the information, then reprocess it inside your own stack.

In practice, an agent can trigger an interview, wait for the completion webhook, pull the transcript and its summary, then move on to whatever task you care about. The @raconte/cli makes all of this scriptable from a terminal, and the SDK from a TypeScript application. You keep control over generation, Raconte handles the collection.

What people do with it

The original goal was to write LinkedIn posts. But once you have a tool that can interview anyone by voice, the use cases multiply.

• Ghostwriting: write on behalf of a founder or an expert, with their real material.
• Team blog: interview your in-house experts to turn their knowledge into solid articles.
• Customer case studies: question the client and the teams who ran the project, then write the case up.
• User feedback: gather rich qualitative input at scale, without scheduling any calls.
• HR feedback: collect internal feedback that no one ever takes the time to go and get.

The time saved is real, but it is not the main point. What matters is the richness of the information collected, the kind you almost never get otherwise. Because usually, you do not do it: you have to get organized, call or book a meeting, ask the questions, take notes or run a transcription. All of that takes time. Raconte does it for you, automatically.

That, at heart, is the conviction behind the product: there is no faster way to get the information you need than a good old conversation.

Want to go further? The getting started guide has you send your first interview in a few minutes.