> ## Knowledge Base Index > Fetch the complete knowledge base index at: https://help.crisp.chat/sitemap.xml > Use this file to discover available pages before exploring further. > Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash). # How to create an AI Chatbot for Customer Service with chatGPT *Learn how to set up Hugo, the Crisp AI Agent, as a chatGPT-powered AI chatbot for customer support.* Hugo combines a supported **OpenAI chatGPT model** with your support content, business instructions, routing rules, and human handoff. This is sometimes described as a [ChatGPT customer support chatbot](https://crisp.chat/en/chatbot/chatgpt/), but Hugo manages the customer support experience while GPT powers language understanding and answer generation. [Discover Hugo](https://hugo.ai/en/) to learn what the AI Agent can do. ||| GPT model availability evolves over time. Always check **AI Agent → Agent → Settings** for the current options in your workspace. --- # ${color}[#0080dd](How to set up an AI chatbot for customer support with GPT) __To select an OpenAI GPT model in Hugo:__ 1. Sign in to [**Crisp**](https://app.crisp.chat). 2. Open **AI Agent → Agent → Settings**. 3. Under **Model**, select a supported GPT model or keep **Choose automatically** enabled. 4. Review **Region**, **Enable Hugo's built-in fallback model**, **Allow agent to process images**, and **Allow agent to transcribe audio messages**. When **Choose automatically** is enabled, Crisp displays the current default below the model selector. The automatic model is currently **GPT-5.1**, but this may change as Hugo's supported models evolve. ![Select an OpenAI GPT model in Hugo settings](https://storage.crisp.chat/users/helpdesk/website/-/8/7/a/e/87ae2703583ac800/cleanshot-2026-06-23-at-235428_hk1sqa.png =1000xauto) Choosing a model is only one part of the setup. Add reliable support resources from **AI Agent → Train**, define Hugo's behavior from **AI Agent → Guidance → Instructions**, and configure escalation rules for conversations that need a human. --- # ${color}[#0080dd](Why chatGPT works well for customer support) OpenAI GPT models are strong general-purpose options for customer support. They are particularly useful when customer wording varies, follow-up context matters, or Hugo must reason across several support resources before answering. __GPT models are especially useful for:__ * **Natural customer conversations** → understand different ways of asking the same question and maintain context across follow-ups * **Reasoning and troubleshooting** → work through ambiguous requests, product issues, and multi-step questions * **Broad support coverage** → handle FAQs, policies, product guidance, and technical support from the same AI Agent * **Flexible cost and quality** → choose a lighter model for high-volume support or a stronger model for more complex requests The quality of your training resources still matters. A more capable GPT model cannot fully compensate for outdated documentation, contradictory policies, or missing product information. --- # ${color}[#0080dd](Compare the GPT models available in Hugo) Hugo currently supports **GPT-5.5**, **GPT-5.4**, **GPT-5.2**, and **GPT-5.1**, alongside selected legacy GPT models. The price levels below represent relative Hugo credit usage rather than a fixed price per conversation. | GPT model | Intelligence | Relative price | Recommended use | | ---- | | **GPT-5.5** | Very high | $$$ | Complex support environments with interconnected sources, ambiguous requests, or deep product reasoning | | **GPT-5.4** | High | $$ | Advanced support that needs strong reasoning with lower relative credit usage than GPT-5.5 | | **GPT-5.2** | Medium | $ | Well-documented support questions and cost-conscious deployments; compare it against GPT-5.1 using real conversations | | **GPT-5.1** | Medium | $ | Standard customer support, higher conversation volumes, and a practical starting point; currently used for automatic selection | | See [which AI models Hugo supports and how to configure them](https://help.crisp.chat/en/article/how-to-choose-and-configure-hugos-ai-model-1m4mm57/) for the latest model, region, and relative pricing options. --- # ${color}[#0080dd](Test and launch your GPT customer support chatbot) Before going live, open **AI Agent → Evaluate → Playground → Compare model options**. Test each model with the same real customer questions, including common FAQs, ambiguous requests, follow-ups, questions involving several resources, and situations that should be escalated. Compare **answer accuracy**, **source usage**, **reasoning quality**, **response speed**, **escalation behavior**, and **credit usage**. Once you have selected the best configuration, deploy Hugo from **AI Agent → Agent → Activation**. For the complete training, guidance, testing, and activation process, follow [Getting started with Hugo AI Agent](https://help.crisp.chat/en/article/getting-started-with-hugo-ai-agent-w6gbux/). --- # ${color}[#0080dd](Frequently Asked Questions) *Still have questions which were not covered in this article? Here is a collection of the most frequently asked questions on this topic.* ###### ${color}[#F08820](Do I need an OpenAI API key to use GPT with Hugo?) No. Supported GPT models are managed directly through Hugo, and you cannot connect an external OpenAI API key. Select the model you want from **AI Agent → Agent → Settings**. ###### ${color}[#F08820](Does Hugo support every GPT model released by OpenAI?) Almost, Hugo exposes a selected list of supported models rather than every numbered GPT release. The models displayed under **AI Agent → Agent → Settings** are the source of truth for your workspace. At Crisp, we work towards evaluating and implementing new model releases frequently to always offer modern and competitive options. ###### ${color}[#F08820](Is the most powerful GPT model always the best choice?) Not necessarily. **GPT-5.5** is best suited to requests that need deeper reasoning, while **GPT-5.4** offers a stronger balance between intelligence and credit usage. **GPT-5.2** and **GPT-5.1** can be more economical for well-documented, high-volume support. Test them with your real customer questions before deciding.