Custom GPTs are configured versions of ChatGPT. Before planning a new one, check eligibility: OpenAI currently limits new GPT creation to eligible managed workspaces. Personal accounts can still use existing GPTs. Check GPT availability.
OpenAI has scheduled custom GPT retirement for December 11, 2026. Qualifying Enterprise workspaces with an approved deferral have a February 11, 2027 date. Existing GPTs remain usable until their retirement date. Check your account or workspace notice before investing in a new GPT. Retirement schedule and migration guidance, checked October 6, 2026.
Custom GPTs run inside ChatGPT. For an assistant on your own website, consider a chatbot platform or a developer-built API integration. OpenAI explains the distinction.
Agentkit documents website embedding, conversation logs and customization. Plan limits and integration access vary by tier. Agentkit plan documentation.
This guide breaks down the differences so you can choose the right tool for your situation.
What Custom GPTs Are
A custom GPT combines instructions, reference files and optional tools inside ChatGPT. Creating one now requires an eligible Business, Enterprise or Edu workspace and the necessary permissions. OpenAI's creation guide.
What you can do with a custom GPT
- Write detailed system instructions to shape behavior
- Upload up to 20 files for the GPT to reference
- Add custom Actions to call external APIs
- Share through the options permitted for your account and workspace. Sharing options.
- Use it within the ChatGPT web app or mobile app
These are documented capabilities, not tests conducted for this comparison. GPT configuration.
Where custom GPTs have limits
GPTs run in ChatGPT and cannot be embedded as a website widget. Builders cannot view individual user conversations. Those limits do not mean GPTs lack external integrations: Actions can call APIs, while eligible workspaces provide sharing controls and usage reporting. GPT scope and conversation access, GPT Actions, workspace reporting.
What Chatbot Platforms Offer
For website deployment, assess the installation route, ongoing configuration and usage limits. Agentkit publishes script and iframe embedding options; its browser SDK works with the script embed. Embed documentation.
Core capabilities of a chatbot platform
- Website embedding. Use the installation snippet from your agent's Channels page. Embed documentation.
- Customization. Basic customization is included across plans; removal of the Agentkit branding starts at Standard. Plan details.
- Conversation records. All plans include conversation logs. Plan details.
- Integration access. API access and integrations start at Hobby. Plan details.
The Comparison: Custom GPT vs Chatbot Platform
These are published specifications for a website-deployment decision, checked on October 6, 2026. This comparison was written by Agentkit and was not a head-to-head product test.
| Feature | Custom GPT | Chatbot Platform (Agentkit) |
|---|---|---|
| AI models available | Current model options depend on user access; retired models may be replaced | Multiple providers on Hobby and above; check the current catalog |
| Website embedding | Runs inside ChatGPT; not embeddable as a website widget | Script or iframe embed; browser SDK on script embed |
| Branding | Uses the ChatGPT interface | Agentkit branding removal on Standard and above |
| Analytics | Enterprise/Edu workspace reporting includes GPT adoption metrics; builders cannot view individual user chats | Conversation logs on all plans; check the reports you need |
| Team collaboration | Managed-workspace sharing can grant chat, settings-viewing or editing access | Published seat limits: Free/Hobby 1, Standard 3, Pro 5 |
| External integration route | Actions can call external APIs; this does not make the GPT itself an embeddable API assistant | API access and integrations on Hobby and above |
| Auto-retrain | Knowledge files are managed in the GPT editor | Automatic website-source retraining on Standard and above |
| Pricing | Check managed-workspace eligibility and current plan terms; personal subscriptions do not enable new GPT creation | Free $0 (50 msgs), Hobby $29.99/mo, Standard $119.99/mo, Pro $399.99/mo |
Sources: GPT overview, GPT editing, GPT sharing, workspace reporting, GPT Actions, Agentkit plans, Agentkit embedding.
Choose the delivery environment first: a GPT inside ChatGPT, a managed website chatbot, or a developer-built API assistant.
Looking for the welcome message field in the GPT builder?
OpenAI documents Conversation starters as example prompts shown when someone opens a GPT. Its configuration guide does not list a separate welcome-message field. The documented flow starts when the user sends a message. GPT configuration, using a GPT.
Agentkit's suggested messages guide distinguishes a welcome message from clickable suggestions. Use the greeting to introduce the bot and the suggestions to show useful questions.
When Custom GPTs Win
An existing GPT, or a new GPT in an eligible workspace, may suit the following uses. Allow time to migrate and test a replacement before the retirement date in your notice.
Internal team tools
In an eligible workspace, a custom GPT can help colleagues use shared reference material. Check creation and sharing permissions before building it. Users sign in to ChatGPT to chat with a GPT. Access requirements.
Examples:
- Internal knowledge base assistant for your support team
- Onboarding helper for new employees
- Research assistant trained on your industry reports
- Content writing assistant with your brand guidelines
Personal productivity
An existing custom GPT may still suit personal work inside ChatGPT. Personal accounts cannot create new GPTs under OpenAI's current availability rules. Current GPT availability.
Prototyping and experimentation
If your workspace permits GPT creation, build a small prototype with reference files and test it in Preview before sharing it. Creating and testing a GPT.
No website deployment needed
If the intended users already work in ChatGPT, evaluate whether an existing or permitted workspace GPT can answer their questions. Test the actual reference material before relying on it.
When Chatbot Platforms Win
A managed chatbot platform may suit a team that wants a website widget without maintaining its own API application.
Website deployment
A GPT itself cannot be embedded in your website. OpenAI also supports developer-built assistants through its API, which is a separate implementation route. GPTs and API assistants.
To install Agentkit, copy the snippet from the agent's Channels page and follow the published embed instructions. Use the generated installation settings for your own agent rather than a copied hostname from this article. Test it on the intended pages before launch.
Connecting external workflows
GPT Actions can send API requests to external applications. That provides an integration route, but it does not establish that a particular contact form or CRM workflow is configured. How GPT Actions work.
Agentkit includes API access and integrations from Hobby upward. Verify the required destination, fields, authentication and failure handling before choosing a plan. Agentkit plan details.
Brand consistency
A GPT uses the ChatGPT interface. Agentkit offers basic customization, with Agentkit branding removal on Standard and above. Compare the actual settings against your brand requirements. Agentkit customization tiers.
Analytics and optimization
GPT builders cannot view individual user conversations. Enterprise and Edu workspace reporting does include GPT adoption metrics, so a blanket claim of no analytics is inaccurate. Builder conversation access, workspace reporting.
Agentkit includes conversation logs on all plans. Check whether its reports answer your questions before choosing a plan. Agentkit plan details.
Model flexibility
Model names and availability change. Check the options available to your users rather than choosing from a retired-model list. How GPT model availability changes.
Agentkit lists models from multiple providers for Hobby and above. Test representative questions with the models available to your plan; this article does not establish that one provider is more accurate. Agentkit plan details.
Team management
Managed workspaces support GPT sharing permissions, including access to chat, view settings or edit. Enterprise and Edu admins also control GPT access through workspace roles. GPT permissions.
Agentkit lists three seats on Standard and five on Pro. Seat limits.
Cost Comparison
Pricing depends on your use case and scale.
Custom GPT costs
Check workspace eligibility before comparing costs. A personal Plus or Pro subscription does not enable new GPT creation under the current rules. Existing GPTs remain usable, while editing still depends on the subscription and permissions. Current requirements.
Chatbot platform costs (Agentkit)
| Plan | Monthly | Messages | Chatbots | Team Seats |
|---|---|---|---|---|
| Free | $0 | 50 | 1 | 1 |
| Hobby | $29.99 | 2,000 | 1 | 1 |
| Standard | $119.99 | 12,000 | 2 | 3 |
| Pro | $399.99 | 40,000 | 3 | 5 |
These are published monthly prices. Check the selected billing interval, current allowances and final checkout terms before subscribing. Agentkit billing documentation.
The Hybrid Approach
One possible setup is to use both:
- Custom GPTs for internal team tools (support team knowledge base, content assistant, research helper)
- Chatbot platform for the customer-facing website chatbot (support widget, lead capture, product questions)
Use this setup only if each tool serves a distinct need. It creates two configurations to maintain, so test whether the extra work is justified. Plan the GPT-to-plugin migration for the internal tool before adopting this setup.
Decision Framework
| Your requirement | What to evaluate |
|---|---|
| A chatbot on your own website | A managed chatbot platform or a developer-built API assistant |
| An assistant used inside ChatGPT | Existing GPT until retirement or a replacement plugin; verify account access |
| External application workflows | The actual API/action route, permissions and failure handling |
| Conversation reporting | Which logs and metrics the selected plan exposes |
| Brand customization | The available settings and the tier needed to remove vendor branding |
Start with the requirement you cannot compromise on. Then test representative questions and the workflows you need. A count of checked boxes cannot replace those checks.
Moving an existing GPT to a plugin
A plugin packages reusable instructions, reference files and connected apps. OpenAI's migration route is to a plugin, separate from building a website assistant. Migration uses the latest published GPT version, so finish essential edits first. Draft changes do not transfer. The original becomes read-only after migration and remains usable until retirement.
Custom actions and sharing settings do not transfer. Rebuild required integrations, check who can access the replacement, and test familiar prompts before switching. For affected Enterprise workspaces, OpenAI plans to end new GPT creation on October 26, 2026. Follow your workspace notice for the applicable dates and migration access. Migration requirements.
Before moving to a website assistant
Treat migration as a new configuration, not a guaranteed copy-and-paste transfer.
- Choose which reference material and instructions the new assistant needs.
- Test representative questions and any external workflows before launch.
- Check the selected plan's limits and what happens when they are reached.
- For Agentkit, follow the published embed instructions and verify the installed widget on the intended pages.
Keep the previous setup available until you have checked the replacement. No migration time or answer-quality result was measured for this article.
Choose by deployment needs
Use a GPT when the intended experience belongs inside ChatGPT and the account has the necessary access. For a website assistant, compare a managed chatbot platform with a developer-built API solution. Evaluate the configuration, limits and maintenance work for your actual use case.
If you already have access to a GPT, try representative questions before choosing a different tool. To create a new GPT, first confirm workspace eligibility. OpenAI's current requirements.
