Designing conversations: the design behind the “Chiedilo a me” chatbot

Every design project is the result of a challenge — very often a shared one.

In a national context where the theme of interaction between people and the Public Administration is very much live, particularly given the role of digital technology, Regione Toscana decided to invest in a new way of understanding this relationship.

Starting from two essential questions: is it possible to simplify as much as possible the findability of documents and tools available to citizens? And if so, what is the most effective way to do it?

The answer was “Chiedilo a me” — a new virtual assistant that has transformed the complexity of information portals into a simple, direct dialogue experience, through careful conversation design work, where language itself is the true interface between people and technology.

This article is the first instalment of a two-part story:

• in this piece we will discuss the design of the conversational experience;
• in the next we will explore the technical solutions that make it possible.

The challenge

Regione Toscana found itself caught in the same knot as many other Public Administration bodies. The information — on sites, on portals — is there, but it is too often buried within pages, attachments and acronyms that change over time.

The brief we received was clear: it was necessary to train guides capable of leading users out of the digital labyrinth — making access to calls for applications, services, opportunities and in-depth content immediate.

By “immediate” we mean that the answer to a question must arrive in a single step. Generative artificial intelligence offers the key to achieving this: transforming search into a conversation — accessible, personalised and within everyone’s reach.

From pages to dialogues

Moving from website design to conversation design was like learning a new language.

It was no longer a matter of organising content and navigation paths, but of creating dialogue experiences: anticipating questions, managing ambiguity, maintaining a consistent tone.

It is an entirely different way of thinking about digital interaction: closer to people, more dynamic and, above all, more human.

The compass of needs: users at the centre

What do people look for when they turn to Regione Toscana? A question the institution first asked itself, and then asked us.

To answer it, we analysed the navigation data of the regional portal, mapping the real needs of citizens, businesses and public bodies.

Our analysis focused on the most visited pages, the most searched keywords and the most visited external links: more than 500 results that we tagged and grouped by regional topic.

The most requested topics? Health, Work and Institutional activities (calls for applications, legislation, opportunities).

These areas became our compass, guiding the selection of sources and the design of dialogues.

We then built an extensive inventory of utterances — the real phrases and questions of users — drawing not only from institutional sites, but also from valuable resources such as the guides and information gathered over the years by the URP.

It was an indispensable and substantial piece of work, aimed at fully immersing ourselves in the chatbot’s information context. Some numbers to give a sense of the scale:

• +10 sources examined, including thematic portals, guides and URP documents
• +300 utterances inventoried, for example:
  – “What training courses are available for me?”
  – “I am an expectant mother — who can I contact about the pregnancy pathway?”
• +250 attributes catalogued, for example:
  – Educational level: upper secondary school diploma, under-18 leaving middle school, etc.
  – Employment status: unemployed, inactive, student, etc.

The heart of the project: the art of Conversation Design

Conversation design is the design of the conversational experience, where language, tone and interaction come together to create a natural and useful dialogue.

Behind this expression lies a complex body of work, drawing on competencies in user experience, interaction design, accessibility and technology.

In designing “Chiedilo a me” we drew on experience gained in previous projects — in particular “Parla con noi”, the chatbot developed for the Scuola Normale Superiore (SNS) dedicated to student guidance.

That work was a valuable proving ground: during its design phase we collaborated with the Designers Italia team, testing the dedicated tool for designing a virtual assistant profile, which provided fundamental guidance in terms of usability and interaction design.

For “Chiedilo a me” we revisited and expanded those guidelines, integrating them with:

• Nielsen’s Heuristics, as a reference for evaluating user experience.
• Best practices on Prompt Controls developed by the Nielsen Norman Group, which suggest how to design interfaces capable of guiding users in constructing prompts, clarifying the system’s capabilities and limitations, and providing visual and textual tools to improve mutual understanding between human and AI.

From this combination emerged a chatbot-specific usability checklist, which we now use as an operational tool to ensure clearer, more transparent conversational experiences oriented towards user control.

To explore SNS’s “Parla con noi” further, you can read:

• La Normale citata da Designers Italia: il chatbot di orientamento tra i casi d’uso del nuovo kit per assistenti virtuali della PA — Michele Fiaschi, Head of Information Systems at the Scuola Normale Superiore
• Progettare un assistente virtuale per servizi pubblici digitali — Daria Cantù, interaction designer and member of the Designers Italia team

User Experience: designing the chatbot’s voice

We worked to create a coherent, accessible and transparent experience:

• Bot Persona definition: building a recognisable identity and an institutional yet empathetic tone of voice.
• Conversation rules: courtesy messages, uncertainty management, out-of-scope responses and hate speech prevention.
• Usability checklist: ensuring that every interaction is fluid and intuitive.
• Conversational controls: question suggestions and guided pathways to facilitate getting started and discovering the system’s capabilities.
• Source transparency: every response displays the official documentation it is based on.
• Feedback and continuous improvement: tools to collect user ratings and optimise responses.
• Prompt bombing prevention: strategies to correctly manage intensive or improper use of the system.
• Quality metrics: measuring conversation satisfaction and success to improve over time.

Interaction Design: giving form to dialogue

The interface is the face of the assistant, and every detail matters.

We defined:

• The behaviour of interactive elements (text box, buttons, icons) to make every possible action clear.
• Animations and transitions between states (initial, response in progress, active conversation) and pages (landing, chat, widget) to convey continuity and naturalness.
• The appearance of chatbot messages, including the micro-states of “thinking” during processing.
• Flows for reporting feedback and technical issues, with dedicated pathways for each case.
• The sources panel, designed to make the origin of information transparent.
• Error and success messages, clear and consistent with the overall tone of the conversation.

Accessibility: a dialogue open to everyone

Designing a virtual assistant for the Public Administration means guaranteeing accessibility to every person. It is not simply a matter of complying with regulations — it means building a bridge towards the citizen.

For this reason, the entire project was guided by the four cardinal principles of POUR accessibility: Perceivable, Operable, Understandable, Robust.

We followed the WCAG 2.1 guidelines and AgID requirements for the Public Administration, but went further, adopting an iterative and corrective approach to anything that was not working as intended. This allowed us to continuously refine every aspect of the user experience:

• Perceivable Interface: we attended carefully to visual contrast and the legibility of all graphic elements and text, to ensure that information is clearly distinguishable even for those with visual difficulties.
• Operable Interaction: we tested keyboard and screen reader interaction with care. The chatbot was designed to be fully usable without a mouse and to communicate every state or change to assistive technologies effectively, ensuring a fully inclusive interaction.
• Understandable Language: we adopted the principle of linguistic simplicity to make the dialogue comprehensible even to those with low digital or linguistic literacy.
• Welcoming Tone: an empathetic and neutral tone that welcomes and assists without judging — demonstrating that accessibility is also a matter of relationship.

The goal is simple: to ensure that “Chiedilo a me” can truly be of and for everyone, regardless of ability or the tools used. 

Behind the scenes: the technological solutions

Every effective conversation is born from a solid, reliable technological foundation.

In the case of “Chiedilo a me”, the challenge was to combine the fluency of natural language with the authority of Regione Toscana’s official information.

To achieve this balance, we adopted a RAG-based approach — Retrieval-Augmented Generation, one of the most effective techniques in generative artificial intelligence today.
In practice, the model does not invent responses, but constructs them from verified documents and sources, selected in real time.

Every conversation is therefore powered by official content, which is retrieved, synthesised and reformulated in natural language, maintaining clarity and coherence.

This approach allowed us to:

• guarantee the accuracy of responses, maintaining total control over sources.
• make the chatbot transparent, showing users the documents from which the information originates.
• optimise the model over time, through the analysis of conversations and user feedback.

In the second part of this story we will go into detail on these choices: how the RAG model works, how we managed intent classification, and how we monitor performance to continuously improve response quality.

The release and next steps

Since September, “Chiedilo a me” has been available on me.toscana.it — the personal space for Tuscan citizens that enables them to receive communications of interest (profiled newsletters) and access online services.

The release marked the first public milestone of a continuously evolving journey, the result of months of research, design, testing and engagement with users and stakeholders.

But the work does not stop at launch: a chatbot is a living organism that grows and improves over time.

For this reason, we have initiated a process of continuous monitoring and iteration, based on the analysis of metrics and real user feedback.

Our commitment is to advance an idea of artificial intelligence that is transparent, controllable and trustworthy.

• Transparency: always making the reasoning behind the bot’s responses comprehensible.
• Controllability: offering users tools to manage their own experience (history, profile, transfer to a human operator).
• Trust: building an ethical and coherent dialogue in which AI is at the service of people.

Conclusions

Bringing “Chiedilo a me” online was an important milestone, but the true result is the transformation this project has generated within us.

We have acquired a new mindset: a different way of thinking about and designing digital experiences — more centred on people, on language, and on listening.

Every conversation, every piece of feedback, every interaction has taught us something about the relationship between citizens and technology, and about how to make this dialogue more natural and transparent.

We have learned that designing conversations means designing relationships: building trust, interpreting needs, giving form to a shared language between people and institutions.

And perhaps this is precisely the most stimulating part of our work: knowing that every project, like every dialogue, never truly ends — but continues to evolve, to learn, to grow alongside those who use it.

By Marianna De Caro (Net7)