First, an important update on custom GPTs
According to OpenAI’s help centre, new GPT creation and publishing are no longer available on personal ChatGPT accounts, including Free, Go, Plus and Pro. Existing GPTs can still be used, and edited where plan and permissions allow. Business, Enterprise and Edu workspaces can still create GPTs if their settings permit it.
OpenAI also says custom GPTs are scheduled to retire on 11 December 2026, with a later date only for Enterprise workspaces that have an approved deferral. It recommends moving workflows to plugins, which bring reusable instructions, reference files and connected apps together. Existing conversations with GPTs stay accessible after retirement.
For most personal users today, the closest equivalent is a ChatGPT Project: a space that keeps related chats, files and instructions together. The points below apply to GPTs, Projects and files uploaded in a normal chat alike.
ChatGPT is a trademark of its owner. Memory Clone is not affiliated with OpenAI or ChatGPT. Features change often; this comparison reflects public information checked on 6 October 2026. Follow the notices in your own ChatGPT account for current options.
What a custom GPT is
A custom GPT is a configured version of ChatGPT. It has a name and description, instructions that apply to every conversation, and optional knowledge files that give it source material. OpenAI’s help page lists up to 20 knowledge files per GPT, each up to 512 MB. GPTs can be created and edited on the web, while the mobile apps support using them.
Projects work in a similar way. You can upload files, add project instructions that override your global custom instructions, and keep related chats in one place. The number of files a project can hold depends on your plan.
What really happens when you upload a WhatsApp export
People often describe this as “training” ChatGPT on their chats. That is not what happens. The model itself does not change. Your file becomes reference material that ChatGPT reads from while it writes a reply.
- Text is extracted. OpenAI says that, outside visual retrieval in Enterprise, document files support text-based retrieval: ChatGPT extracts the digital text and discards images.
- There are size caps. Text and document files are capped at 2 million tokens per file, so a very long chat history may need splitting.
- Files are stored in your account. Files in a chat follow that chat’s retention. Knowledge files stay until you delete the GPT. After deletion, OpenAI says files are removed from its systems within 30 days, with some security and legal exceptions.
- Training is a separate setting. On personal accounts, the “Improve the model for everyone” control decides whether new conversations can be used to train OpenAI models. By default, OpenAI does not train on content from Business, Enterprise or Edu workspaces.
- Memory is separate again. When Memory is on, ChatGPT can remember preferences and details from your chats to personalise later answers. It does not keep every detail.
Our guide turning WhatsApp chats into an AI chatbot explains what a WhatsApp export contains and why its format matters.
Where ChatGPT does well
- Flexibility. It can summarise a long chat, find a half-remembered conversation, or help you write a message, all in one place.
- General knowledge. It can explain, research and draft about almost anything, not just your own history.
- Everywhere you are. It runs on the web and in mobile apps, and Projects can be shared with others.
- Quick experiments. For a one-off question about one chat, uploading a file is often all you need.
If your goal is analysis, summaries or general writing help, a general assistant may simply be the better tool.
Where it falls short for simulating one person
One person, many relationships
Nobody writes the same way to everyone. You might send your sister voice-note jokes and your manager careful full sentences. A GPT or Project has one set of instructions and one pool of files. Unless you build and maintain a separate one for each contact, style and private details from one conversation can easily appear in another.
Whose words are whose
A chat export mixes both sides of the conversation. Group chats add many more voices. A general assistant has to work out, every time, which lines belong to the person you want to simulate, and it can mix them up.
Excerpts, not habits
Retrieval finds passages that look relevant to the current question. That is good for recalling facts, but writing habits such as message length, emoji use or how someone says goodbye are spread across thousands of messages. They are hard to capture from a handful of excerpts.
Other people’s messages
A full export includes everything the other person wrote to you. Uploading it stores their words on a cloud service. Think about whether they would be comfortable with that, especially for sensitive chats.
How Memory Clone approaches the same job
Memory Clone is a Windows desktop app built only for this task. It imports WhatsApp, Messenger, Instagram and Discord exports plus your own notes, and builds a memory index on your PC.
- Per-relationship style. The clone writes to each contact the way the person wrote to that contact. Private facts from one relationship are kept out of other conversations.
- Full-archive recall. Memory Search finds exact dated memories in the complete local index, and replies can draw on it.
- Local first. Originals stay on your PC. Cloud AI processing is off until you switch it on for a clone, and then only a bounded selection of excerpts and style statistics is sent. Memory Clone does not train models on your content.
- Honest framing. Every clone is presented as an AI simulation. The app is in alpha and is still developing.

Side-by-side comparison
| Aspect | ChatGPT custom GPT or Project | Memory Clone |
|---|---|---|
| Main purpose | General assistant you can customise | Simulating one real person from their messages |
| How your chats are used | Reference files ChatGPT reads during a conversation | Local index plus style learned per relationship |
| Relationship separation | Not built in; one instruction set per GPT or Project | Built in for each contact |
| Where original files live | Stored in your OpenAI account | On your Windows PC |
| Training on your content | Controlled by your data settings | No training by Memory Clone |
| General knowledge and tasks | Broad | Focused on the person and their memories |
| Platforms | Web and mobile apps | Windows desktop app |
| Availability for new builds | New GPTs not available on personal plans; GPTs scheduled to retire | Available (alpha) |
Which should you choose?
ChatGPT is probably enough if you want summaries, analysis or writing help, you mainly use a phone, or you only need to ask questions about one conversation.
Memory Clone fits better if you want an AI that writes like a specific person to many different people, keeps each relationship distinct, and leaves the original archive on your own computer. See how style learning works or download the app.
If you do use ChatGPT with chat exports: review your data controls, remove chats you would not want stored online, and delete files you no longer need. Our local-first privacy guide has more ways to share less.
If the chats belong to someone else, ask them first. Read our consent guide and the Acceptable Use Policy.
常见问题
Can I train ChatGPT on my WhatsApp chats?
Uploading a chat does not retrain the model. ChatGPT reads the file as reference material during a conversation, and whether your conversations help train future models depends on your data controls.
Can I still create a custom GPT in 2026?
OpenAI’s help centre says new GPT creation is not available on personal plans such as Free, Go, Plus and Pro. Business, Enterprise and Edu workspaces may still create them, and custom GPTs are scheduled to retire on 11 December 2026.
What is the alternative to a custom GPT in ChatGPT?
For personal use, Projects keep files, instructions and chats together. OpenAI recommends plugins as the replacement for custom GPT workflows.
Does Memory Clone use ChatGPT?
When you switch on cloud processing for a clone, Memory Clone sends a bounded selection of excerpts through its server to an OpenAI model under API terms that exclude training. It is a separate product, not a custom GPT.
Is a custom GPT bad for imitating a person?
Not bad, but general. It works for simple cases, while keeping many relationships separate and capturing long-term writing habits takes extra structure.
资料来源
核实于 。第三方产品和帮助页面经常变化,请点击链接查看最新详情。
- Creating and editing GPTs (OpenAI Help Center)
- Custom GPT retirement and migration FAQ (OpenAI Help Center)
- Projects in ChatGPT (OpenAI Help Center)
- File Uploads FAQ (OpenAI Help Center)
- Data controls in ChatGPT (OpenAI Help Center)
- Memory in ChatGPT (OpenAI Help Center)
App menus change over time, so labels on your device may differ slightly. Memory Clone creates an AI simulation from the material you provide; it is not the person.