Home Getting Started on the Platform How to create ClaudIA's initial content (guide to uploading your first base)

How to create ClaudIA's initial content (guide to uploading your first base)

Last updated on Jul 05, 2026

When to use

  • You are building the first content base for ClaudIA in your project
  • You need a spreadsheet template and best practices to avoid common mistakes
  • You want to understand the relationship between IDS, tags, and content types (N1/N2/Interactive)

Prerequisites

  • Access to the project's Hub
  • Access to the helpdesk to extract tickets (if applicable)
  • (Optional) FAQs, training materials, and internal documentation to feed the base

About this article

This guide explains how ClaudIA uses content, how the Cloud Humans team builds the initial base, and the best practices for creating content that minimizes errors and improves retention.

What is ClaudIA?

ClaudIA is Cloud Humans' generative AI customer service agent. It was created to answer questions with more agility and empathy, using a content base from your company.

It is powered by large language models (LLMs) and trained with content extracted from tickets, FAQs, and your team's documents.

How does ClaudIA answer requests?

ClaudIA works with question/title and answer pairs — what we call IDS (Semantic Data Intelligence).

  1. The customer sends a request

  2. ClaudIA performs a semantic search in the IDS base

  3. It identifies the most relevant content and responds

Structure of each IDS

  • Question/title — how the customer triggers that content

  • Answer — text sent when the question is asked

  • Content type:

    • N1 — ClaudIA answers on its own

    • N2 — forwarded to a human

    • Interactive — triggers a Controlled Flow (spreadsheet, API)

  • Tag — categorizes the conversation

What is a TAG?

Each IDS receives a tag. The Tagger is the mechanism that defines the ticket's final tag.

How the Tagger decides

  1. It looks at which IDS contents were used in the conversation

  2. It applies the corresponding tag:

    • If N2 → tag of the section that was handed over

    • If Interactive → tag of the section that triggered the Controlled Flow

    • If N1 → tag of the most relevant content

If the conversation used more than one type of content, the priority order N2 → Interactive → N1 applies. Example: a ticket that used an N2 section receives that section's tag, even if an N1 content was more relevant in the conversation.

  1. If no content is used, the result depends on the project's default tag setting:

    • Default tag enabled → the configured default tag (usually NO_TAG) is applied directly

    • Default tag disabled → ClaudIA analyzes the conversation and tries to apply the tag of the most similar category in your base; the default tag (NO_TAG) is only used if nothing is found

    Learn more in How to reduce NO_TAG and increase how much ClaudIA classifies

  2. The tag is used to generate reports and metrics

We recommend writing tags in lowercase, without accents, without "ç", and separated by underscores (e.g., certificado_incompleto) — this avoids mismatches with your helpdesk and makes reporting easier.

How we build the content base

Step 1 — Extraction

We extract tickets from your helpdesk (~25% of the average volume).

Step 2 — Cleaning

We remove duplicates.

Step 3 — Clustering

We group similar questions to avoid duplicates and identify the most common questions.

Step 4 — Building the answers

Based on:

  1. Real answers from agents in the tickets

  2. Extra content provided (FAQs, training materials)

Step 5 — Spreadsheet generation

We generate a Google Sheets file with suggested questions and answers.

Step 6 — Client validation

You review:

  • Is the question clear and specific?

  • Is the answer correct and up to date?

  • Is the content useful or irrelevant?

  • What is the correct tag?

  • Is it N1 or N2?

Best practices for creating/validating content

  • Avoid questions that are too broad. ❌ "Can I change my order?" ✅ "Can I change my purchase order after it has been issued?"

  • Do not use greetings in the answers — ClaudIA already does this automatically

  • Avoid duplicated sections — it confuses the model and hurts performance

  • Use clear, well-written questions/titles — fix spelling mistakes

  • Point long instructions to external links: "You can renew your certificate by following the step-by-step guide at the link below: [link]."

  • Include specific instructions inside the IDS title: [If you decide to use the answer below, always send it in full without changing absolutely anything] How do I request a refund?

  • Add variations of the question in the same section: "I don't have my ID with me right now, can I still attend? / I won't have my ID on the day, is it okay to go anyway? / If I don't have my ID, am I still allowed in?"

Template for creation/validation

🔗 Content Validation Template

Notes