ChatGPTAI chatbotHow-toSmall business

Can You Embed a Custom GPT on Your Website? What Works Instead

No. Custom GPTs cannot be embedded, visitors must sign in, and OpenAI set 11 December 2026 as their retirement date. Here are the routes that work.

PepoChat TeamPublished Last verified 13 min read
A bearded man in a checked shirt reads his phone at a bright office desk beside an open laptop

Short answer

No, you cannot embed a custom GPT on your website: OpenAI says GPTs are not a way to put ChatGPT on an external site, visitors must sign in to chat, and Chrome blocked the one iframe we tried from a local test page. OpenAI set 11 December 2026 as the standard retirement date for GPTs. To answer visitors, use the OpenAI API or a website chatbot trained on your content. PepoChat is one, free for 100 AI replies a month.

You built a custom GPT that knows your prices, your policies and your tone, and it answers better than your FAQ page does. The obvious next step is to put it on your website. That is where most people get stuck, because ChatGPT has no embed button and no snippet to copy.

This guide is for the owner or support lead in that spot. It sets out what OpenAI allows, checks each workaround, prices the real options and shows how to move a GPT's instructions and files into a website chatbot. Prices and rules checked 21 November 2026, on OpenAI's own Help Center and platform docs. PepoChat is our product; we include it where it fits and say where other tools are the better choice.

Can you embed a custom GPT on your website?

No, you cannot embed a custom GPT on your website. OpenAI's help page on GPTs in ChatGPT says GPTs "are not a way to embed ChatGPT in an external website or application" and sends anyone building an assistant into a product to the API. Until retirement you can link to a GPT, but visitors must sign in to chat.

What you might tryDoes it work?Visitor needs a ChatGPT account?Status on 21 November 2026
Link to the GPTYes, until retirementYesGPT pages inaccessible from 11 December 2026 (standard date)
IframeNoNot applicableBlocked by ChatGPT's headers
Unofficial embed toolUnreliableDepends on the toolRisky under OpenAI's terms
GPT ActionsNot for thisYesLets a GPT call your API, not the reverse
OpenAI API with your own chatYesNoSupported, needs a developer
OpenAI ChatKitYesNoAvailable, needs your own server
Website chatbot on your contentYesNoNo code

The link is the only official way to send a visitor to a GPT, and it now has an end date. That changes the decision for anyone who planned to rely on it.

What happens to custom GPTs on 11 December 2026?

OpenAI set 11 December 2026 as the standard retirement date for custom GPTs, the date GPTs and their pages become inaccessible, according to its retirement and migration FAQ. The FAQ says the transition affects all ChatGPT plans. A link to your GPT on your website, in an email or behind a QR code stops being useful at that point.

DateWhat happensWho it applies to
26 October 2026New GPT creation endedAffected Enterprise workspaces, except FedRAMP
11 December 2026Standard retirement: GPTs and their pages become inaccessibleAll plans, unless your notice says otherwise
28 February 2027Later retirement dateFedRAMP workspaces and Enterprise workspaces with an approved deferral

New GPT creation and publishing are not available on personal accounts (Free, Go, Plus and Pro), so a business without a Business, Enterprise or Edu workspace cannot start a new GPT, and inside those workspaces it depends on settings and permissions. OpenAI's FAQ tells account holders to follow any different timeline in their own notice, so check yours.

OpenAI's replacement is plugins, packages of skills and connected apps that run inside ChatGPT and Codex. Migration turns a GPT's instructions into a skill and copies its knowledge files. Custom Actions do not transfer, public sharing settings do not carry over, and people who used the GPT do not get the plugin automatically. Migration and plugin access differ by account or workspace. A plugin keeps the workflow alive for your team, but it puts nothing on your public website.

A GPT link is a weak support channel because the customer leaves your site, needs a ChatGPT account and ends up in a conversation you cannot see. Our guide to using ChatGPT for customer service covers the wider picture. These are the gaps that matter when you already have a GPT and are deciding what to do with it:

  • Sign-in comes first. Public GPT pages may be visible without an account, but OpenAI says "users are prompted to sign in before chatting." A shopper with a quick question about delivery has little reason to create an account with another company.
  • You cannot read the conversations. OpenAI states that "GPT builders cannot view individual conversations" people have with their GPTs. Wrong answers stay invisible, and nobody on your team can pick up a complaint.
  • There is no human handoff. A GPT has no team inbox, no escalation status and no alert when a customer needs a person.
  • Your brand stays outside. The chat happens on chatgpt.com, in ChatGPT's interface, without your domain or your colours.
  • Every chat starts cold. GPTs do not use saved memory or past conversations, so a returning customer explains the problem again.
  • Live data needs development. An order lookup means building and hosting an Action, and the customer is still on chatgpt.com when it runs.
A woman with long dark hair smiles at her smartphone on a grey sofa in a brightly painted living room
A visitor checking your returns policy from her sofa has no reason to create a ChatGPT account first.

Does an iframe work for a custom GPT?

An iframe does not work for a custom GPT, because ChatGPT tells browsers not to show its pages inside other websites. We ran this test once, on 21 November 2026: one public GPT page in an iframe on a blank test page saved as a local file, opened in headless Chrome. Chrome refused to load it and logged this error, shortened here:

Framing 'https://chatgpt.com/' violates the following Content Security
Policy directive: "frame-ancestors 'self' chrome-extension://..."
The request has been blocked.

The frame-ancestors rule lists who may display the page in a frame: ChatGPT itself and four browser extensions. ChatGPT's home page also sends X-Frame-Options: SAMEORIGIN, an older header with the same effect. Our test page was not a public site, but the rule names no other origin, so a browser that honours it should block the frame on any ordinary website, whichever site builder or plugin you use.

Are unofficial GPT embed tools safe to use?

Unofficial tools that promise to embed a custom GPT are best avoided, because the ones that really connect to ChatGPT have to work around OpenAI's rules. OpenAI's Terms of Use, in the version for users outside the EEA, Switzerland and the UK, say you may not "share your account credentials or make your account available to anyone else". They also forbid you to "automatically or programmatically extract data or Output" or "circumvent any rate limits or restrictions". A widget that relays your visitors through one ChatGPT login runs into all three.

The practical risks are just as plain. Such a tool can stop working whenever ChatGPT changes, your customers' messages pass through a company you may have no agreement with, and the GPT behind it disappears at retirement anyway.

One question sorts these tools quickly: does it run on the OpenAI API with its own key, or does it need your ChatGPT login, a session cookie or your GPT's link? The first is an ordinary chatbot product. The second is a wrapper, and you should walk away.

Can GPT Actions put your GPT on a website?

GPT Actions cannot put a GPT on your website, because they work in the opposite direction. An Action lets a GPT inside ChatGPT call an outside service through a REST API you describe. That is useful for looking up an order from within ChatGPT, but the customer is still on chatgpt.com and still signed in. Actions also do not transfer in OpenAI's migration to plugins, so any Action you depend on has to be rebuilt whichever route you choose.

Can you build a website chatbot with the OpenAI API?

Yes, the OpenAI API is the official way to put an OpenAI model on your website, and it needs a developer. You send each conversation to the Responses API, OpenAI's current interface for generating replies, show the answer in your own chat window and keep the history yourself or with the Conversations API.

Three dates on OpenAI's deprecations page catch people who follow older tutorials:

OpenAI productDeprecation noticeShutdown dateReplacement
Assistants API26 August 202526 August 2026Responses API and Conversations API
Agent Builder3 June 202630 November 2026Agents SDK or ChatGPT Workspace Agents
Reusable prompt objects3 June 202630 November 2026Prompt text in your application code

ChatKit is OpenAI's embeddable chat interface, and OpenAI's deprecations page, read on 21 November 2026, says it remains available after Agent Builder's shutdown. It is the closest thing to an official embed, but it is a component for developers and needs a server of your own. For new apps, OpenAI's guide points to a server-side agent of your own, built with the ChatKit SDKs and the Agents SDK. The guide describes no way to load an existing GPT, so your instructions and files still have to move.

Usage itself is cheap. Here is what 1,000 support replies cost on gpt-6-luna, one of the low-priced models on OpenAI's API pricing page, at Standard rates, assuming each reply reads 4,000 tokens (instructions, retrieved passages and history), writes 250 and makes one file search call:

Item, for 1,000 repliesQuantityOpenAI rateCost (USD)
Input tokens4,000,000$0.10 per million$0.40
Output tokens250,000$0.50 per million$0.125
File search calls1,000$2.50 per 1,000 calls$2.50
TotalAbout $3.03

The search calls cost more than the model in this example. File storage adds $0.10 per GB per day after the first free GB.

The bill is the small part. You also build and maintain the chat window, the content import, limits against abuse, an inbox for your team and the alerts that bring a person in. Our estimate is days for a prototype and weeks for something you would trust with customers.

What about plugins, workspace agents and ChatGPT Sites?

OpenAI's newer products do not give you a public website chatbot either, although each is easy to mistake for one.

OpenAI productWhat it doesWho uses itChatbot on your public site?
PluginsPackages skills and apps for a taskPeople in ChatGPT and CodexNo
Workspace agentsRuns repeatable team workflowsWorkspace members, in ChatGPT or Slack, or by APINot described
ChatGPT SitesPublishes websites and light appsPlus, Pro, Business and Enterprise usersNot described
ChatGPT BusinessShared workspace for staffYour team, from 2 usersNo

"Not described" means OpenAI's help page for that product, read on 21 November 2026, mentions no way to use it as a chatbot for anonymous visitors. A public ChatGPT Site can be opened without signing in, but its help page covers hosting a site and tools for ChatGPT, and says nothing about running your GPT for visitors.

ChatGPT Business costs $20 per standard seat a month billed yearly, or $25 billed monthly, from two users. It is a sensible purchase for your staff's own work, and it does not answer your customers.

Which way of adding a chatbot to your website should you choose?

A no-code chatbot trained on your content is the right choice for most small businesses, and the OpenAI API is the right choice when chat is part of your product and you have a developer.

RouteMonthly cost (USD)Setup and skillAnonymous visitorsUses your contentHuman handoff
Link to a GPT$0 for a GPT you already have; a new one needs a workspace, from $40 for 2 Business seats billed yearly. Ends at retirementMinutes, no codeNoYour filesNo
IframeNot applicableBlockedNoNot applicableNo
Unofficial toolVariesMinutes, no codeSometimesVariesVaries
GPT ActionsIncluded with the GPTHours to days, developerNoCalls your APINo
OpenAI API, own chatAbout $3 per 1,000 replies, plus developer timeDays to weeks, developerYesIf you build itIf you build it
ChatKit, own serverSame usage, plus hostingDays to weeks, developerYesIf you build itIf you build it
No-code chatbot (PepoChat)$0 for 100 replies, $39 for 1,000About an hour, no codeYesYesBuilt in

Prices from each vendor's own pages, checked 21 November 2026. Setup times are our estimates.

  • Move the GPT to a plugin if only your staff use it. Nothing about it needs to be on your website.
  • Choose a no-code chatbot if customers will use it, you have no developer, and you want answers from your own content with a person behind them.
  • Choose the OpenAI API if chat is part of your software, needs deep access to your own systems, or you already have a developer with time to maintain it.

The arithmetic for building is simple. At 1,000 replies a month, building saves about $36 against PepoChat Starter ($39 minus about $3). At 17,000 replies it saves about $448 against Scale ($499 minus about $51). Hosting and developer time come out of that saving, so the case for building gets strong only at high volume or with unusual requirements.

How do you move a custom GPT's instructions and files into a website chatbot?

You move a custom GPT by copying its instructions, splitting them into behaviour and facts, uploading its knowledge files, rebuilding any Actions and installing a script tag. The steps below use PepoChat, and the same order works in most chatbot tools. The running example is a made-up bike repair shop whose GPT has 6,500 characters of instructions, eight knowledge files and one Action that checks whether a repair is ready.

GPT fieldWhere it goes in PepoChatPepoChat limit
Instructions: tone and personaAgent instructions, How to sound and behave4,000 characters
Instructions: hard rulesAgent instructions, Never do this2,000 characters
Facts inside the instructionsKnowledge Base, Text & Q&A20,000 characters each
Knowledge filesKnowledge Base, Add New25 MB per file
Conversation startersWidget Customization, Default suggestions3 suggestions
ActionsCustom Actions10 inputs per action

Step 1: Copy everything out of the GPT

While the GPT is still available, go to chatgpt.com/gpts, select My GPTs, choose the GPT and select Edit GPT. Paste its name, description, conversation starters and instructions into a document, and list its knowledge files and Actions. Find your original files as well, because OpenAI's retirement FAQ describes no way to download a GPT's knowledge files.

Step 2: Split the instructions into behaviour and facts

Behaviour is how the agent should sound and what it must refuse. Facts are prices, hours and policies, and they belong in the knowledge base, where the agent can find them and you can correct them in one place.

The bike shop's 6,500 characters split into 1,800 of tone and persona, 700 of hard rules such as "Never quote a repair price that is not in the price list", and 4,000 of facts. In PepoChat, the first two go into the Agent instructions card on Widget Customization. The facts become 12 Q&A pairs under Knowledge Base, Add New, Text & Q&A. Whatever you paste, PepoChat's base rules still apply: the agent searches your knowledge base first, does not invent answers and offers a person when it cannot help.

A man in a shirt and tie takes notes in an open notebook beside a laptop during a phone call
Moving a GPT starts with notes: list what is behaviour and what is a fact the knowledge base should hold.

Step 3: Upload the knowledge files

PepoChat accepts PDF, DOCX, Markdown, HTML, TXT, CSV, JSON and common image files up to 25 MB, added one at a time. A GPT can hold 20 files of up to 512 MB each, so some need work first: save spreadsheets as CSV and slides as PDF, and split anything over 25 MB.

On the Free plan each uploaded file counts as one of your 5 knowledge sources, while text snippets and Q&A pairs count as none. The shop's eight files become the price list as a CSV (one source), the service menu PDF (one), three warranty PDFs merged into one (one), and the FAQ, opening hours and brand list rewritten as snippets (none). That is three sources.

Step 4: Add your website

Paste your address under Knowledge Base, Add New. A crawl covers up to 200 pages, three levels deep, and counts as one source, so the shop is now at four of five. Your site is often fresher than the files you gave the GPT. Paid plans re-fetch website sources weekly; on Free, click Sync website pages after a change.

Step 5: Rebuild the Actions

A GPT Action has to be rebuilt in whatever tool you move to. In PepoChat, Custom Actions call any HTTPS endpoint with up to 10 inputs the agent collects from the visitor. There is no OpenAPI import, so you enter the method, URL, headers and inputs by hand. The shop's repair check becomes one GET request with a ticket_number input. A hand-built action starts switched off, so click Test and then turn it on.

If your Action talked to Shopify, Stripe, Zendesk or HubSpot, check the prebuilt integrations first. Never connect refunds or payments, because an action can occasionally run twice.

Step 6: Test the agent

Click Test the agent on Widget Customization. Test chats are free and stay out of your analytics and inbox. Ask ten questions the GPT handled well, two it should refuse and one your content does not cover, then ask for a person and check that the agent offers one.

Step 7: Install the script tag

Open Widget Setup, copy your Organization ID and paste one tag before the closing </body> tag of your site. On WordPress, use the official plugin instead.

<script
  src="https://widget.pepochat.com/widget.js"
  data-organization-id="YOUR_ORGANIZATION_ID"
></script>

Then remove any old link to the GPT from your site. For volume, the shop's 30 chats a month at three AI replies each come to 90 replies, inside the Free plan's 100. At 60 chats that is 180 replies, which needs Starter at $39 a month.

Some GPT features have no equivalent here. PepoChat does not search the open web, generate images, run code or open a canvas. If your GPT earns its keep through those, keep it for your team as a plugin and use the website chatbot only for customer questions.

Is your data safe with a GPT, the API or a website chatbot?

Whether your customers' data is safe depends on the route, because OpenAI applies different defaults to consumer accounts, business plans and the API.

RouteUsed to train models by default?Who can read the conversations
GPT, used from a Free, Go, Plus or Pro accountMay be, unless the user opted outNot the GPT's builder
GPT, used from a Business, Enterprise or Edu accountNoNot the GPT's builder, but workspace admins can have access
OpenAI APINo, since 1 March 2023, unless you opt inYou, if you store them
PepoChat, built on the APIStandard API termsYour team, in the inbox

OpenAI's enterprise privacy page says it does not train on business data by default, that you own your inputs and outputs where the law allows, and that API inputs and outputs may be kept for up to 30 days to provide the service and identify abuse. The same page lists SOC 2 Type 2 audits for Business, Enterprise, Edu and the API, and says OpenAI can sign a Data Processing Addendum with Business, Enterprise and API customers.

Those commitments are OpenAI's. They do not automatically cover a tool built on its API, so read that vendor's own terms; our guide on where chatbot data goes lists the questions to ask. PepoChat uses OpenAI models under standard API terms and publishes no compliance certifications of its own, so do not collect card numbers, health details or government IDs in chat.

A GPT Action adds another party. OpenAI warns that "relevant parts of your input may be sent to the third-party service" and that it "does not audit or control how those services use or store your data."

Where PepoChat fits

PepoChat fits the business that built a GPT to answer customers and now needs that help on its own website. It is an AI support agent that runs on OpenAI models, answers from your knowledge base and loads with one script tag, so visitors chat on your site with no account. When it cannot answer, or a visitor asks for a person, the conversation moves to your team inbox with the transcript, and your team gets an email alert.

You can start on the Free plan, which includes every chat feature for 100 AI replies a month, with 5 knowledge sources and 20 MB of text. Starter is $39 a month for 1,000 replies, unlimited sources and 200 MB.

The limits matter if you are coming from a GPT. You cannot pick the model. PepoChat is built to answer customers from your content, so it is not a general assistant for writing or analysis. The small "Powered by PepoChat" line shows on every plan, and PepoChat publishes no compliance certifications. If your staff need a research or drafting assistant, ChatGPT is the better tool, and if chat is a core part of your software, build on the API.

What to do next

Frequently asked questions

Can I add ChatGPT to my website for free?
Not ChatGPT itself, because OpenAI offers no embed for it and the API is billed by usage. The free route is a website chatbot with a free plan that answers from your content. PepoChat's Free plan, for example, includes 100 AI replies a month, a chat widget and human handoff, with no credit card.
Can I train ChatGPT on my website?
Not in the sense most people mean. A custom GPT can draw on files you upload, which you replace by hand when your site changes. Website chatbots use retrieval instead: they index your pages, find the passages closest to each question and write the reply from them, so a corrected page changes the answers without any model training.
Do website visitors need a ChatGPT account to use a custom GPT?
Yes. OpenAI's Help Center says GPTs are available to all ChatGPT users, but people must be signed in to start a conversation. A public GPT page may be visible without an account, and visitors are prompted to sign in before chatting. A free ChatGPT account is enough, but it is still a step many customers will not take.
Is ChatGPT Business the same as a chatbot for my website?
No. ChatGPT Business is a shared workspace for your own staff, priced at $20 per standard seat a month billed yearly or $25 billed monthly, from two users, as checked on 21 November 2026. It includes plugins and workspace agents for internal work. It does not give you a widget that answers anonymous visitors on your website.
What replaced the OpenAI Assistants API?
The Responses API and the Conversations API replaced the Assistants API, which OpenAI shut down on 26 August 2026 after announcing its deprecation a year earlier. Assistants map to settings in your own code, threads map to Conversations and runs map to Responses. Tutorials written before the shutdown may still show the old calls, so check the date on anything you follow.
Does PepoChat use ChatGPT?
PepoChat uses OpenAI's models through the API under standard API terms, and it does not run on the ChatGPT product. Your visitors need no ChatGPT account, and you cannot pick the model. The agent answers from your own knowledge base and hands off to your team, so it is a support tool and not a general assistant.

Try this on your own site in ten minutes

PepoChat includes every chat feature on the free plan — 100 AI replies and 5 knowledge sources a month, no credit card.