Sign inJoin the waitlist

GPT-3 and ChatGPT AI chatbots: what each is, what is still available, and how to build one for your business

What the two names mean now, and three ways to get a chatbot that answers your customers.

Published by AI Knows Us · Updated 13 October 2026

A GPT-3 chatbot is a chat interface built on OpenAI's GPT-3 models, and a ChatGPT chatbot is one built on the models behind ChatGPT. The two are no longer the same thing to build on. OpenAI's deprecations page lists the original GPT-3 base models (ada, babbage, curie and davinci) and text-davinci-003 as shut down on 4 January 2024, so a new AI chatbot is built on a current OpenAI model, or on another provider's, through an API. If you only want to chat with an AI, use ChatGPT itself. If you want a chatbot on your own website that answers your customers, you build one on an API, and the rest of this page shows how.

What GPT-3 and ChatGPT actually are

GPT-3 is a large language model. The research paper was published in May 2020 and OpenAI opened an API beta in June 2020. The model completes text: you give it the start of something and it writes what comes next. On its own it was not a chatbot. Developers wrapped it in code that fed it a conversation and showed the reply.

ChatGPT is a product. It launched on 30 November 2022 and its first model came from the GPT-3.5 series, tuned to hold a conversation. The difference matters when you search for a "GPT-3 AI chatbot": most people who type it want a conversational assistant, and ChatGPT is the product that gave them one. Behind it, OpenAI has since moved on to newer models, so the name GPT-3 now describes history, not something you can sign up for.

TermWhat it isWhere it stands
GPT-3The 2020 language model family, reached through an APIBase models shut down 4 January 2024, per OpenAI's deprecations page
GPT-3.5The series behind the first ChatGPTSuperseded by newer models
ChatGPTOpenAI's consumer chat productLive, runs on current models
OpenAI APIThe route for building your own chatbotLive, billed by usage

Three ways to get a chatbot

1. Use ChatGPT directly

Open it, type, and read the reply. This is what most people searching "chat gpt ai chatbot" are after, and it needs no code. The free plan and the paid plans differ in which models and features you get, and OpenAI states the current limits on its own pricing page, which is the place to read them because they change.

2. Build a chatbot on the API

A website chatbot has four parts. Your page collects the visitor's message. A small server of yours sends it to the model together with instructions and any facts the bot needs. The model returns a reply. Your page shows it. The server step matters because your API key must never sit in code the visitor can read.

The instructions you send are called the system prompt. Write it like a briefing for a new staff member: who the bot speaks for, what it may discuss, how long its answers should be, and what to do when it does not know. A bot told to say "I will pass this to the team" when it lacks an answer behaves better than one left to improvise.

For cost, the API bills by tokens in and tokens out, which are roughly pieces of words. Our guide to what is free and what is billed in the ChatGPT API walks through the billing, and OpenAI's own documentation, which our guide to the OpenAI documentation maps, holds the current rates.

3. Use a no code chatbot builder

Many website tools offer a chat widget you paste onto your site, trained on your pages. You give up control of the prompt and the model in exchange for an afternoon's setup. The questions to ask any of them are the same as for a build: which model runs underneath, where the visitor's messages are stored, how you read the conversations, and how a human takes over.

Making a business chatbot answer correctly

A chatbot that invents prices is worse than no chatbot. Four habits prevent it.

  • Give it the facts in the prompt. Opening hours, services, delivery areas, return terms. Paste them as plain text and tell the bot to answer only from that text.
  • Retrieve, do not memorise. For larger sets of documents, the standard design stores your pages in a search index. For each question the server finds the most relevant passages and sends only those to the model with the question. This is often called retrieval, and it keeps answers tied to your current pages.
  • Define a refusal. Tell the bot what to say for medical, legal, financial and complaint topics, and where to send the visitor.
  • Read the transcripts weekly. Every wrong answer is a missing line in your facts or a weak instruction. Fix the prompt, then retest the same question.

Choosing between the options

You want toChooseSkill needed
Chat, draft and research for yourselfChatGPTNone
Answer customer questions on your site this weekA no code builderPaste a snippet
Control the model, the data and the lookAn API buildA developer, a few days

Start with the cheapest option that meets the need, and move up when the transcripts show you what you are missing.

Questions about GPT-3 and ChatGPT chatbots

Can I still use GPT-3 to build a chatbot?

The original GPT-3 base models were shut down on 4 January 2024, according to OpenAI's deprecations page. Build on a current model listed in the OpenAI documentation.

Is ChatGPT the same as GPT-3?

No. GPT-3 is a model family from 2020. ChatGPT is a product launched on 30 November 2022, first run on the GPT-3.5 series and now on newer models.

Do I need to code to make an AI chatbot?

No. A no code builder gives you a widget to paste on your site. Code is needed when you want control over the prompt, the model and where messages are stored.

How do I stop my chatbot making things up?

Give it your facts, tell it to answer only from them, and give it a fixed reply for questions outside them. Then read the transcripts and fix each wrong answer at its source.

Is the OpenAI API free?

It is billed by usage. Check OpenAI's current pricing page before you plan a budget.

The chatbot your customers already use

While you build a chatbot for your site, your customers are already talking to someone else's. A shopper types "which accountant in Pune handles startups" into ChatGPT, and the reply names a few firms before they ever reach a website with a chat widget. That answer comes from what the model learned and what it finds on the web, and it is different on Gemini and Perplexity. AI Knows Us asks those assistants the questions your buyers ask and records whether your business is named, with a free scan that needs no sign in. Checking it takes five minutes and shows you how much work your own chatbot has to do after the visitor arrives.

Read next

AI Visibility Audit Deliverables for a UK Agency: The Five Artefacts, and Why the Zero Column Is One of ThemWhat a British agency should hand a client at the end of an audit, including the findings that did not survive a second reading.A Workflow Standard for AI Search Optimisation: Six Stations, and Who Owns Each OneThe execution sequence an Indian agency can hold a team to, measured against what our own site serves eleven named crawlers.The Multi Engine AI Citation Audit Framework: One Column Per Engine, Never a Blended ScoreA reproducible audit design for an Indian agency, built from 1,260 readings on 16 October 2026 and the 597 that came eleven days before it.A Small UK Agency Against a Large Consultancy on AI Search Optimisation: What the Citation Data Says Each One Can OwnThree British agency domains took dozens of citations each in our readings, and none of them is a large firm. Here is what that result does and does not tell a ten person shop.Build or Buy an AI Visibility Stack: The Five Silent Wrong Answers an In House Build ProducesWhat it actually takes to run a multi engine tracker yourself, measured from the defects in our own build, with rupee prices for the alternative.AI Search Optimisation and SEO in India: What Each One Costs, and What Each Return Is Counted InMost of the production work is already inside your search invoice, two budget lines are genuinely new, and the two results are counted in units that cannot be divided by each other.

Or browse the whole archive.

See what AI says about you.

The first scan is free, it needs no account, and it asks the questions your buyers ask.

Free. No card. We ask 5 real buyer questions on 2 AI apps.

Already sure you want this? Join the waitlist. We open on 31 October 2026. It costs an email address. No card, and nothing is charged. Or talk to a person if you have a question first. Or read how we fix what it finds.