ChatGPT Projects or custom GPTs: what should your company implement?

Choose a ChatGPT project for an evolving case and a custom GPT to distribute a defined way of completing tasks. A team preparing a proposal for one customer needs shared context for that case. Employees preparing similar proposals for different customers may need common instructions and approved service descriptions.

Separating those needs makes documents easier to maintain and reduces mixing of customer information. It does not immediately require an application built through the API.

How projects and custom GPTs differ

OpenAI describes projects as spaces grouping conversations, files and instructions. A custom GPT has configuration for a specific purpose; it does not use a user's saved memory or previous conversations. Each new conversation starts afresh.

As documented on 30 September 2026, personal Free, Go, Plus and Pro accounts cannot create or publish new GPTs. Business, Enterprise and Edu workspaces allow creation according to permissions. Existing GPTs remain usable, while editing depends on the account. Check this before buying a course that requires building a GPT on a personal Plus subscription.

Example: one proposal and a shared departmental method

Suppose a service company is preparing a proposal for a customer who changes requirements several times. A project can hold the brief, confirmed decisions and working drafts. The case owner maintains the current version. Project instructions identify authoritative prices and how to flag open questions.

A separate GPT could help the department structure proposals consistently. Its shared knowledge should contain material appropriate for all authorised users, such as service descriptions and working rules. Employees provide individual customer details in the relevant conversation according to company policy.

This is a proposed working arrangement to test on your account. Do not assume the GPT's creator can see user conversations. OpenAI states that builders cannot view individuals' conversations with their GPTs. Keep the register of approved proposals in the appropriate company system if you need one.

Establish access before adding documents

A shared project gives participants access to its context. Review sharing and memory settings before placing different customers' documents together. An instruction saying “do not show customer A's data” does not replace access control.

Start with one project for a case or a team whose members have the same access scope. Test an account that should have access and one that should not. If you use invitation links, check who can join.

For a GPT, assign configuration editors and define how new material is approved. Connecting an app or API action adds another access boundary: some information may reach an external service. Assess the connection separately from answer quality.

Test updates on a small case set

Prepare tasks with expected results: a complete brief, a missing deadline, conflicting requirements and an old price list. Check whether the tool identifies gaps or fills them with assumptions. Record the documents and instruction version used in the trial.

Then replace one source, such as the price list. Remove the outdated uploaded copy where necessary and repeat the trial in a new conversation. Do not assume an uploaded copy stays synchronised with its original. For a project, also check whether earlier conversations introduce contradictory decisions.

The process owner should be able to rerun this check after the next change. Without someone maintaining current information, a well-prepared first version can gradually become unreliable.

Budgeting for preparation

An illustrative four-person ChatGPT Business Standard budget on monthly billing is 4 × USD 25 = USD 100 per month, before taxes and additional usage. OpenAI pricing, checked on 30 September 2026.

Budget separately for organising sources, writing instructions and testing. A modelled ten hours at EUR 25 per hour represents EUR 250 of internal effort. This is a planning scenario, not an implementation quote. Syntalith training starts at EUR 600 net per day; configuration and additional preparation need an agreed scope.

If the output must enter the CRM automatically, appear on a customer website or run to a schedule with defined monitoring, consider an OpenAI API integration. A custom GPT operates in ChatGPT and is not itself a website widget.

Describe the task and who needs to use it. We will help choose a project, a custom GPT or an integration around the workflow and its data requirements.