> For the complete documentation index, see [llms.txt](https://help.clausebuddy.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.clausebuddy.com/introduction-to-genai/introduction-to-genai.md).

# Introduction to GenAI

The philosophy behind ClauseBuddy is that **the future is hybrid**: **human + traditional software + GenAI**. The new type of Artificial Intelligence ("Generative AI" or "GenAI") is here to stay, and will bring a lot of changes to the legal sector. However, not even the brightest and fastest GenAI can replace human lawyers and traditional database technology is also more relevant than ever for legal professionals.&#x20;

When working with GenAI, you will notice that it is very creative, so that it can produce brilliant results. Yet it is also highly unpredictable and will sometimes produce quite inferior results (often referenced as "hallucinations". As former lawyers ourselves we know that this will turn off some legal lawyers, because the results are often far from perfect — but if you avoid the technology, you'll miss out on all the brilliance it frequently offers you and the speed at which it offers those results.&#x20;

The question is therefore how to get most of the brilliance but avoid the bad results. **GenAI is a new software tool on your legal toolbelt that any modern legal professional must master, just like any other tool.**&#x20;

By default, ClauseBuddy integrates with the GPT4.1 "Large Language Model" (LLM), as [published by Microsoft](https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy?context=%2Fazure%2Fcognitive-services%2Fopenai%2Fcontext%2Fcontext). However, that's just the default, because ClauseBuddy is "LLM-agnostic" and actually allows you to use most of the custom LLM's on the market, such as the EU-based [Noxtua from Xayn](https://xayn.com), or [Anthropic Claude](https://claude.ai).&#x20;

{% hint style="info" %}
See also: [Custom LLMs](/admin/custom-llms.md) if you want to bring your own LLM.
{% endhint %}

***

## **Features**

ClauseBuddy offers GenAI-based features and buttons across the entire app, in roughly all modules — for example, also the traditional Quality Library hosts many GenAI functionality. However, several modules in ClauseBuddy (such as Doc Chat, Polish Text, etc) work entirely on the basis of AI. Without AI, these modules would not work.&#x20;

### **Drafting clauses**

Drafting clauses with the LLM is easy: you go to the *Draft New Text feature* in the ClauseBuddy menu, click on *Draft new text* again and enter a query. The software will then come up with an answer after a few seconds, which you can insert into your document, ready to be inserted in the same layout.&#x20;

<figure><img src="/files/9DnnfpSSXVA0v6gNKn62" alt="" width="563"><figcaption></figcaption></figure>

### **Drafting other legal texts**

Above we talked about contract clauses, because those will be the most frequently produced paragraphs of text. However, you can also ask the LLM to draft other types of text — from short summaries of some legal rule, to grammatical enhancements of certain fragments to text, to even poems (*legal poems*, of course!).

### **Rewriting clauses**

Instead of drafting a new clause, you may want to rewrite existing text:&#x20;

<figure><img src="/files/5eYkozvAeFO11fGAljbZ" alt=""><figcaption></figcaption></figure>

In both the Draft New Text and Rewrite Text modules, ClauseBuddy will automatically take into account the terminology used in the active document.

You can then use the Rewrite Text to make selected text more in favour of one of the parties if desired.

Keep in mind that the numbering you see in the preview is not final. The numbering will be adjusted based on where the clause will be inserted in the document.

### **Creating summaries**

You can ask the LLM to create a summary of some selected text. You can do so going to the *Summarise* module, selecting text within your MS Word document (or pasting text when you are using ClauseBuddy outside of Word), and then clicking on the blue *Summarise* button.

By default a summary is 'Simple' — i.e. a short, single paragraph. But it is also possible to produce more structured summaries like so:

<figure><img src="/files/2B7SvhbjJrr704VmJ2hA" alt="" width="563"><figcaption></figcaption></figure>

Selecting Configure *> Create new summary type* opens a summary format editor. Here, you can create your own summary type. For instance, a summary rendered as a table with columns for the *Date*, *Party*, and *Content* of the clauses in the selected text (or uploaded document).

<figure><img src="/files/IAuYF9XhTHeFpHHcsXgE" alt=""><figcaption></figcaption></figure>

### **Drafting full documents**

You can instruct the LLM to generate an entire outline of a document, and then interactively add clauses to them — either from your own clause library, or generated by the LLM.

Read more about it on our [dedicated page](https://help.clausebuddy.com/compose-document/) for Composing full-Documents.

### Reviewing documents

The [Playbooks](https://help.clausebuddy.com/playbooks/) feature is also 100% GenAI-focused. They allow you to create custom rulesets ("playbooks") to automatically review legal documents against your internal standards.

<figure><img src="/files/8M2MZUipZCU0yKkPB2PM" alt="" width="249"><figcaption></figcaption></figure>

***

## **Confidentiality & compliance**

GPT4.1 and its successors has surprised the legal world, and many legal teams and law firms have published warnings about being careful with this technology — particularly as regards the hallucinations that may happen for very factual answers (much less a risk for typical contract clause drafting)

ClauseBuddy uses Microsoft Azure's version of GPT4.1 by default, which does not reuse your data to train the AI (as opposed to the free version of the LLM which potentially causes confidentiality leaks). As [explained on Microsoft's technical website](https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy?context=%2Fazure%2Fcognitive-services%2Fopenai%2Fcontext%2Fcontext), Microsoft only reserves the right to look into your data to investigate abuses or technical failures.

{% hint style="info" %}
For more in-depth information about this topic, see the section on "confidentiality" in our [blog post "Generative AI - fall update"](https://www.clausebase.com/post/generative-ai-fall-2023-update).
{% endhint %}

{% hint style="warning" %}
Warning. You should be aware that GenAI technology is  100% stable from a technical perspective. Accordingly, you will sometimes get errors from the service, for inexplicable reasons.

Also, you should be aware that Microsoft (similar to OpenAI) applies content filtering on the prompts and the answers. Any profanity, sexual content, semi-illegal content, etc. will therefore likely be refused by GPT4.1.
{% endhint %}

## General remarks about AI and “hallucinations”

Tools such as ChatGPT and ClauseBuddy, which rely on LLM technology, can be extremely powerful, but they also come with certain limitations. One important issue to be aware of is the potential for inaccurate or even entirely fabricated output, a known phenomenon often referred to as “hallucinations”.

Several factors influence the likelihood of such hallucinations:

* **Length of the document:** The longer the document the AI has to process, the greater the chance of inaccuracies. You can compare it to a reader losing focus over time.
* **Location of the answer:** Answers located at the beginning or end of a document are typically easier for the model to retrieve reliably than those buried in the middle.
* **Fragmentation of relevant information:** When the answer must be pieced together from many different parts of a text, the risk of errors increases.
* **How the question is phrased:** LLMs are designed to always provide an answer, even when unsure. Strongly implying that an answer *must* exist can lead the model to produce confident-sounding but incorrect information.
* **Domain-specific knowledge:** While LLMs have broad general knowledge, they typically lack detailed or up-to-date legal expertise, especially if such information wasn’t publicly available during training.

These limitations are not specific to ChatGPT — they apply equally to all AI tools that rely on the same underlying models, including ClauseBuddy. This is also why such tools are not suitable for deep legal research. There's a real risk that the AI will confidently fabricate a legal case or misinterpret legal doctrine in a way that sounds convincing but is ultimately incorrect.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://help.clausebuddy.com/introduction-to-genai/introduction-to-genai.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
