Where the conversation happens
Custom GPT: ChatGPT UI (and GPTs you publish there). Reply: your website widget and connected messaging channels.
Compare
Reply vs Custom GPT — your customers should not need a ChatGPT account
A Custom GPT is a tailored ChatGPT experience: instructions, optional files, and tools inside OpenAI's product. Reply is a multi-tenant RAG service you embed on your domain. Visitors chat on your site or WhatsApp; answers retrieve from your crawled content; chats stay scoped to your workspace.
- Your domain, your widget
- Tenant-isolated retrieval
- 7-day money-back on paid
The demo trap
“We uploaded the PDF to a GPT” is not a support channel
Internal teams love Custom GPTs because ChatGPT is already open. Buyers on your store will not log into ChatGPT to ask if COD is available in 560001. App Store listing, Plus/Team seats, and GPT availability are OpenAI's surface — not yours.
File search inside a GPT is also not the same as a crawl, hybrid search, exact FAQs, origin-allowlisted widget, and plan-enforced isolation between clients.
- Custom GPTs are for people who already live in ChatGPT
- Reply is for people who landed on your site or messaged your business number
- Agencies cannot safely share one GPT across clients without mixing knowledge
Decision guide
Use each tool for the job it was designed for
Keep Custom GPTs for internal operators
Drafting, research, and staff who already have ChatGPT seats. Great sandbox. Weak customer channel.
Use Reply for public Q&A on your properties
Widget script, optional WhatsApp/Instagram, knowledge updates via crawl and upload, handoff to your team.
Do not paste the same system prompt into both and call it done
Prompting is not retrieval with tenant scope, widget keys, and origin allowlists.
What actually differs
Embed, isolation, and “it is not on chatgpt.com”
Whose content is searchable
Reply queries are tenant-scoped. A Custom GPT's files live in that GPT; sharing the GPT shares the knowledge.
How knowledge gets in
Reply: site crawl, PDF/DOCX/MD/CSV, exact FAQs, hybrid search + rerank. GPT: files and instructions you attach in OpenAI's builder.
When it should refuse
Reply is built for support: decline or hand off when retrieval is weak. A GPT will often keep chatting unless you constrain it carefully.
Controls buyers care about
Reply: widget appearance, keys, plan limits, handoff, analytics. Custom GPT: OpenAI's GPT settings, not your dashboard.
Side by side
Reply vs Custom GPT
| Capability | Reply | Custom GPT |
|---|---|---|
| Runs on your website | YesYes | No |
| Visitor needs a ChatGPT account | No | YesYes |
| Tenant-isolated multi-client bots | YesYes | No |
| Website crawl as a first-class ingest | YesYes | No |
| WhatsApp on the same knowledge base | YesYes | No |
| Human handoff in your dashboard | YesYes | No |
| Best as an internal assistant | Possible, not the focus | YesYes |
Runs on your website
- Reply
- YesYes
- Custom GPT
- No
Visitor needs a ChatGPT account
- Reply
- No
- Custom GPT
- YesYes
Tenant-isolated multi-client bots
- Reply
- YesYes
- Custom GPT
- No
Website crawl as a first-class ingest
- Reply
- YesYes
- Custom GPT
- No
WhatsApp on the same knowledge base
- Reply
- YesYes
- Custom GPT
- No
Human handoff in your dashboard
- Reply
- YesYes
- Custom GPT
- No
Best as an internal assistant
- Reply
- Possible, not the focus
- Custom GPT
- YesYes
FAQ
Custom GPT questions
Not as a Custom GPT. OpenAI has other APIs; they still are not Reply's crawl-widget-handoff product. Custom GPTs remain inside ChatGPT.
Keep it for staff. Add Reply if customers must ask on your domain or WhatsApp without a ChatGPT login, with isolation and handoff.
This page is vs Custom GPTs (the ChatGPT product). Building your own RAG on the API is the “build it yourself” column on our other landing pages — months of work we already operate.
Chat completions are provided via Groq; embeddings via Together AI. Model IDs are configured in env, not advertised as a ChatGPT wrapper.