How the Nevent chatbot works

The Nevent chatbot is a virtual support assistant that answers your fans’ questions in real time from your website or landing page. This page explains how it works under the hood: where it gets its information from, what role each configuration layer plays, and why the system is designed this way.

What the Nevent chatbot is

The chatbot is an artificial intelligence assistant that you install on your event’s website to handle attendee questions without your team having to respond to every message. When someone asks about doors opening times, the lineup artists or how to top up cashless credit, the bot answers instantly with your event’s information.

It isn’t a fixed question-and-answer chatbot. The assistant generates natural-language responses by combining several layers of information that the promoter controls.

The four layers that make up every response

When a fan writes a message, the chatbot composes its reply from four layers that work together. Understanding each one helps explain why the bot answers the way it does and how to improve it.

1. Custom instructions

Instructions are the core of the assistant’s behavior. Here you define who the bot is, what it prioritizes and how it acts in each situation. For example, you can tell it to call itself “Official Demo Fest Assistant”, use an informal tone, prioritize information about access, or always offer the support email when it doesn’t know the answer.

The field accepts up to 5,000 characters of free-form natural language text. You don’t need to know how to code. Writing style, on the other hand, is managed separately in the voice layer.

2. The assistant’s voice

Voice defines how the assistant sounds when it writes: the vocabulary it uses, how long its sentences are, how formal it is. It’s generated from real conversations your team has had. The platform recommends providing between two and six varied conversations (common questions, incidents, complaints) with a minimum of 500 words in total.

The difference between instructions and voice is conceptual. Instructions say what to do; voice says how to say it. If you don’t configure voice, the assistant writes with a generic default style.

3. Reference documents

These are files you upload that the bot can consult to give more precise answers: the venue guide, the event FAQ, access rules, the stage map. Supported formats are PDF, TXT, MD and DOCX, with a maximum size of 10 MB per document.

When you upload a document, the platform processes it and sets it to “Ready” status. Only then can the chatbot read it. If the status isn’t “Ready”, the document isn’t used.

The logic is simple: the more complete and up to date the document, the more accurate the bot’s answers will be. A well-written FAQ covering your event’s common questions reduces the chance of the bot improvising or getting things wrong.

4. Nevent data sources

Data sources are the structured information that already exists in your Nevent account. Instead of manually copying that information into a document, the chatbot can query it in real time whenever the source is turned on.

There are up to eight sources grouped into three blocks:

Event information: The event’s name, description and main image; the venue’s physical location, dates and gate opening times; and the full lineup with set times, stages and featured artists.

User information: The attendee’s name, their assigned folders and current location; their loyalty points, tier reached and available benefits; and their ticket type, included experiences and purchased add-ons.

Additional information: Points of interest, service locations and venue areas from the map; and the list of sponsor brands with their links.

You can turn each source on or off independently. The reason for turning off sources you don’t need is practical: the platform notes that unchecking unnecessary sources optimizes token usage. At Demo Fest, for example, the sponsors source is turned off, which explains the “7 of 8 sources active” marker visible in the status panel.

How the chatbot decides what to include in each response

When a question comes in, the assistant combines what it knows from the four layers and generates a natural-language response. It doesn’t look up a predefined answer; instead it reasons over the available information to build a reply appropriate to the context.

The language model used by the assistant at Demo Fest is Gemini 3.5 Flash Lite, visible in the configuration’s status panel.

What language the chatbot answers in

The chatbot detects the language the visitor writes in and answers in that same language. This capability is automatic and requires no extra configuration from the promoter.

In the Demo Fest inbox, active conversations have been observed in Spanish, Dutch, Japanese and Romanian, all handled by the same assistant.

The widget has a default language setting (Spanish, English, Catalan or Portuguese), but if the page where it’s installed declares another supported language, the widget adopts it automatically.

Tracked links and the push to sale

When the chatbot recommends an event or links to documentation, it can use a measurable short link (with the nevt.link domain) instead of the original URL. This makes it possible to count how many clicks the assistant generates and attribute them to specific conversations.

There’s a limit of 200 new links per day per venue. If that’s exceeded, responses go back to using the original URL with no tracking.

This measurement layer is what feeds the “Push to sale” section in the performance panel, where the promoter can see how many link recommendations the bot made, how many clicks they generated, and what percentage of conversations resulted in a click.

The inbox and human intervention

The chatbot isn’t a closed channel. From the inbox, any team member can see the status of every conversation and take control of any of them in real time. When an agent steps in, the bot stops responding and the agent writes directly to the fan.

Conversations have four possible states: bot in progress (the assistant is responding autonomously), in progress (a human agent has taken over), escalated (the bot couldn’t resolve the query and is waiting for human attention) and resolved.

This design lets the chatbot handle the high volume of repetitive questions, with human agents available for the cases that require judgment or information the bot doesn’t have.

Why this layered architecture

The separation into instructions, voice, documents and data sources responds to a specific goal: letting the promoter update each part without touching the others.

If the lineup changes, the information is updated in the platform and the chatbot reflects it automatically through the “Artist lineup” source, with no need to edit any document or rewrite the instructions. If the festival’s tone of voice changes, it’s enough to adjust the voice or the instructions, without touching the event data.

This modularity reduces maintenance work and lowers the risk of the bot working with outdated information.