Can you embed a custom GPT on your website? What OpenAI says
OpenAI's support team says custom GPTs run inside ChatGPT and cannot be embedded in another website. What that means, the two ways to rebuild the same assistant on your own site, and what each one asks of you.
MagizAI Team6 min read
Not directly. OpenAI's support team said on its developer forum in August 2026 that custom GPTs are designed to run inside ChatGPT and cannot be embedded in another website. To put the same assistant on your site, you rebuild it: with OpenAI's API and a chat interface, or with a website chatbot trained on the same files.
We are not affiliated with OpenAI. ChatGPT and GPTs are its names for its own products, and what follows about them comes from OpenAI's own pages, read on 13 October 2026.
What a custom GPT is
A custom GPT is a version of ChatGPT set up for one job. OpenAI's developer documentation lists what you can give it: instructions, documents attached as knowledge, and connections to outside services through what OpenAI calls GPT Actions. An Action lets the GPT call another system's API from a plain-language question, for example to fetch data or file a ticket.
All of that is stored in the GPT, inside ChatGPT. That is where people use it, which is why the question of putting it on your own pages comes up at all.
Can I embed a custom GPT on my website?
No. A user asked exactly this on OpenAI's developer forum, and on 19 August 2026 the OpenAI Support account replied:
Custom GPTs are designed to run within ChatGPT and can't be embedded directly into an external website. For a website where your users can interact with an assistant, the supported approach is to build the experience using the OpenAI API and configure its instructions, tools, and context to match the behavior you need.
The reply points to OpenAI's ChatKit guide, covered below. So if a page offers you an embed code for your GPT, it is not describing an OpenAI feature. Ask what that tool really runs on your site, and whose account pays for it.
What OpenAI offers instead
OpenAI's route is to rebuild the assistant with its API. In practice that takes four pieces.
- The model and your instructions. You send them through the Responses API. Many older tutorials use the Assistants API instead, but OpenAI's migration guide says it was sunset on 26 August 2026 and is no longer available.
- Your files. The file search tool in the Responses API searches files you upload to a vector store. That store plays the part of your GPT's knowledge.
- A chat window. ChatKit is OpenAI's embeddable chat interface. Its guide has you create each chat session on your server, sign your users in, pass a unique ID for each of them, and add a script to your site.
- A server. The server holds your API key, creates sessions and hands the browser a client secret instead of the key. In the setup OpenAI now recommends, it also runs the agent itself.
One more date matters. ChatKit can also run on workflows hosted in OpenAI's Agent Builder. The ChatKit guide says Agent Builder is being deprecated and is scheduled to shut down on 30 November 2026, while ChatKit stays available. For new work it points you to your own server-side agent.
On cost, the API is priced per token. OpenAI's pricing page, checked on 13 October 2026, also lists file search storage at $0.10 per GB per day, with 1 GB free, and $2.50 per 1,000 file search calls in the Responses API.
The other route: a website chatbot trained on the same files
If you would rather not run a server, a website chatbot service does the same job from a dashboard. You upload the documents you gave your GPT, add your website address, write in your instructions and install the chat window with a line of code or a plugin. The service hosts the chat, stores the knowledge and calls the AI model for you.
What you leave behind is the GPT itself. Expect to copy your instructions across by hand and to replace any Actions with the integrations the service offers. What you can gain are things a business chat on its own site needs, such as a team inbox and a hand-off to a person, if the service you pick has them.
Trade-offs side by side
| Rebuild with OpenAI's API | Website chatbot service | |
|---|---|---|
| Where the chat appears | Your site, through ChatKit or your own interface | Your site, through the service's chat window |
| What you build | A server, user sign-in, the agent and its tools | Setup in a dashboard and one install step |
| AI models | OpenAI's | Whatever the service supports |
| Who pays for the AI | You, per token, on your OpenAI account | Included in the plan, or your own key where offered |
| Your files | Uploaded to a vector store through the API | Uploaded in the dashboard |
| Upkeep | Yours: updates, security, monitoring | The service's, apart from your content |
| Suits | A developer team that wants full control | A business that wants the assistant live without code |
What to look for in a website chatbot
- Answers from your files. It should show where an answer came from, and say so when your content does not cover a question.
- A way to reach a person. A visitor should be able to get to your team in the same chat, and your team should be told.
- Room for your instructions. You should be able to bring over the instructions you wrote for your GPT and keep its tone and limits.
- Model choice. If you liked how your GPT answered, check that OpenAI's models are on the list, and whether you can use your own API key.
- Data terms. Find out which AI provider sees your visitors' messages, under which terms, and where the service keeps conversations.
- Install on your platform. Check that your site builder, and your plan on it, allow custom code or have a plugin.
- The pricing unit. Seats, messages, conversations or resolutions: work out the cost at the volume you expect, not at the entry price.
Where MagizAI fits
MagizAI is one website chatbot built this way. It learns from your website pages and sitemap, PDF, DOCX and HTML files, questions and answers and plain text. It shows the source under an answer and says it does not know when your content does not cover a question. When a visitor needs a person, the chat moves to your team inbox with the whole conversation, and your team is alerted.
You can run it on OpenAI's models with your own API key, or pick another supported provider, and install it with one line of code or the WordPress plugin. The training guide shows how to add your files, and the install guide covers each site builder.