If you built a custom GPT and it has sat frozen ever since, AI Central's verdict in The Best ChatGPT-5 Features is that Projects have quietly taken over the job. Projects keep memory between chats, hold more files, run deep research, sync edits everywhere, work properly on a phone and share with real permissions. GPTs still win on one thing, public distribution. The strongest setup uses both, with your GPTs running inside a Project.
Reviewed August 2026.
The build everyone abandoned
Custom GPTs arrived as the first genuine way to package a workflow inside ChatGPT and hand it to somebody else. Thousands of people built one. AI Central's account of what happened next is short.
Published them. Waited for monetization. It never came
The commercial upside never materialised, and neither did the product roadmap. AI Central's core charge against the format is that a GPT is finished the moment you publish it.
They never evolve, just a static knowledge base that can't adapt
Projects, which AI Central notes were upgraded heavily through 2025, went the other direction. AI Central calls them living workspaces rather than locked tools, and the mechanism behind that phrase is worth understanding, because it explains almost every other difference between the two.
They learn from your uploads and evolve with every conversation.
Memory is the whole argument
Start with what happens when you open a new chat. Inside a GPT, in AI Central's description, you start from nothing every time.
Each chat starts blank, so context gets lost and accuracy drifts
That is not a personality flaw in the model, it is a plumbing fact. A GPT's knowledge is whatever you froze into its instructions and its uploaded files. Anything you established in yesterday's conversation is gone, so today the model rebuilds your situation from a shorter prompt and fills the gaps with plausible invention. Drift shows up as confidently wrong specifics, the wrong pricing tier, last quarter's positioning, a client name that belongs to a different account.
Projects hold the thread instead. AI Central's framing is direct.
Project-only memory and retained context cut repetition and reduce hallucinations across chats
The phrase doing the work there is project-only. Memory scoped to a single workspace means one client's brief never bleeds into another client's draft, which is the failure mode that makes people distrust general memory and turn it off. It also removes the tax nobody counts, the four paragraphs of context you re-paste at the top of every session.
The file ceiling, and why it moves
Capacity is where the gap turns from philosophical to practical. AI Central puts the GPT limit plainly.
Lifetime limit for the entire GPT. Once you hit 20, you're done
Twenty files, ever, for the whole build. That is a hard stop rather than a rolling window, so a GPT that survives a year of real use eventually forces a choice between deleting knowledge and rebuilding from scratch. AI Central sets against that a Projects capacity that scales with your plan, roughly five on Free, twenty-five on Plus and up to fifty on Pro.
The strategic difference matters more than the numbers themselves. A fixed lifetime cap means your workspace gets worse as your work gets richer. A ceiling tied to your plan means the ceiling moves when you do.
One answer, or an investigation
AI Central's next contrast is about depth. A GPT gives one answer and stops, no investigation and no synthesis, just a single response. A Project can run deep research instead.
ChatGPT gathers and analyses data from multiple sources to deliver complete reports
That changes which questions are worth asking of which tool. A GPT suits a repeatable transformation where you already know the shape of the output, turn this transcript into three posts, rewrite this in our tone of voice. A Project suits a question whose answer shape is unknown, what are competitors charging, what actually changed in this market last quarter. If you have been putting research questions to a custom GPT and finding the answers thin, the tool was never built for that job.
Edits that actually reach every chat
A GPT does not auto-refresh. AI Central's description of the workaround will be familiar to anyone who maintains one, every edit means opening a new chat or rebuilding the GPT before the change shows up. Projects behave the way you would expect them to.
updates to files or instructions apply across every conversation
This is the difference between a build artefact and a live workspace, and it should decide which one you pick. If your instructions never change, a GPT's staleness costs you nothing. If your pricing, your positioning or your tone of voice changes every month, that refresh cost is the entire cost of ownership.
Where the work actually happens
AI Central is unsentimental about mobile. GPTs barely work there, with limited features and desktop dependent workflows, which is a real constraint given how much of anyone's day is spent away from a laptop. Projects do not carry that limitation.
Upload files from your phone and switch between models
Switching models inside the same Project is the underrated half of that. It means the workspace, not the model, is the thing you commit to, so a quick lookup and a heavy piece of analysis can run against the same files and the same memory without you reassembling context in between.
The one thing GPTs still do better
Distribution. AI Central credits GPTs with reach a Project cannot match, since a GPT can go public on the Store or be shared by link, which makes it ideal for discovery or lead generation. The trade-off is that its permission model is binary, a GPT is either private or public, with nothing in between.
Projects answer a different question. AI Central describes secure workspace sharing built for teams and clients, with granular permissions, so you can invite a teammate or a client with chat rights or with edit rights rather than handing over the whole build. One format is a shop window, the other is an office. Confusing the two is how people end up disappointed by both.
The combination most people miss
The most useful line in AI Central's comparison is the one about nesting. You cannot use a Project inside a GPT. You can use GPTs inside Projects.
combining their custom behaviours with shared memory and files
That is the setup worth copying. Keep the GPT as the public front door, the thing a stranger finds and tries, and keep the actual work in a Project where the files, the memory and the permissions live. Then call the GPT from inside the Project when you want its specific behaviour, and it operates on the project's context instead of starting blank.
What to do this week
- Move your most-used custom GPT's instructions and files into a Project, work there for a week, and count how often you still have to re-explain yourself.
- Run one Project per client, product or ongoing piece of work, so project-only memory stays clean instead of collecting everything you have ever done.
- Keep a GPT published only if it is doing a distribution job, discovery, lead capture or a public demo. If nobody outside your team ever opens it, it does not need to be a GPT.
- Send research-shaped questions to a Project running deep research, not to a GPT, which is built to answer once and stop.
- Before you invite anyone, decide whether they need chat rights or edit rights, because a Project can express that difference and a public GPT cannot.
Should I delete my custom GPTs?
Not automatically. AI Central's comparison leaves GPTs one clear advantage, public distribution through the Store or a shareable link, so a GPT that brings you discovery or leads is still earning its place. If it only ever serves you and your team, the twenty-file lifetime cap and the blank start to every chat make it the weaker home for that work.
How many files can I put in a Project?
AI Central's figures scale with your plan, roughly five on Free, twenty-five on Plus and up to fifty on Pro. The comparison sets those against a flat twenty-file lifetime limit for an entire custom GPT, which does not move whatever you pay.
Do Projects really hallucinate less than GPTs?
AI Central puts the difference down to context rather than to the model itself. A GPT starts every chat blank, so context is lost and accuracy drifts. A Project retains context and keeps project-only memory, which cuts repetition and reduces hallucinations across chats.
Can I use a custom GPT inside a Project?
Yes, and not the other way round. AI Central is explicit that you cannot use a Project inside a GPT, while GPTs can be used inside Projects, combining their custom behaviours with the project's shared memory and files.
Do Projects work properly on my phone?
Yes. AI Central rates mobile as fully supported for Projects, including uploading files from your phone and switching between models. GPTs get the opposite rating, barely usable on mobile, with limited features and workflows that assume you are at a desktop.