Written by Marijn Overvest | Reviewed by Sjoerd Goedhart | Fact Checked by Ruud Emonds | Our editorial policy

ChatGPT Projects for Procurement: The Setup That Keeps Context Alive

The Context Problem Projects Solves

Every procurement professional who uses ChatGPT seriously encounters the same pattern within weeks of starting. The ChatGPT answers were impressive at first because context was being explained carefully. Then the explanations got tedious. Then the answers started feeling generic because the procurement professional stopped explaining the context properly. Then the tool became less useful than it was at the start.

Most procurement teams find that isolated experiments with ChatGPT only become a durable team capability when tool practice is paired with structured training. The AI Fundamentals for Procurement Teams program is built for exactly that transition, from individual curiosity to a procurement function that works differently.

The problem is not ChatGPT. The problem is that ChatGPT without scoped context has to treat every question as if asked by a generic procurement professional at a generic company. The output is competent but generic. Good for quick research; poor for work that is specific to a category or a supplier.

Projects are ChatGPT's answer to this. A Project is a persistent workspace scoped to a specific piece of work, a strategic category, a supplier renewal, a cross-functional initiative. The Project holds files, custom instructions, and conversation history, all scoped. Every interaction inside the Project benefits from the accumulated context.

The Procurement Tactics 2026 AI Readiness in Procurement survey found that procurement teams at the Deploying or Embedded stages of AI adoption disproportionately use Projects (or the equivalent in other tools). Teams that remain at Exploring or Experimenting typically use ChatGPT as a one-off chat tool. The difference is not the tool; it is the operating pattern.

What Goes in a ChatGPT Project

A procurement Project has four components. Each shapes the quality of the output.

Custom instructions. The role, scope, and output preferences for the Project. Example: "You are an analyst supporting the IT Services category at a European manufacturer. Outputs should be concise, British English, and structured as tables where relevant. Reference the performance data in the uploaded scorecard when discussing suppliers."

Uploaded files. The documents that define the Project's subject matter. For an IT Services category Project: the spend overview for the category, the supplier list with performance data, the contract register, the market intelligence brief, and the company briefing document. For a supplier renewal Project: the current contract, performance history, renewal correspondence, and comparison suppliers.

Conversation history. Every interaction inside the Project is scoped to the Project. The procurement professional can start a new conversation thread without losing the accumulated context from previous work. This is what makes Projects feel like a workspace rather than a chat tool.

Knowledge (on higher tiers). On ChatGPT Team and Enterprise plans, additional knowledge base capabilities let Projects reference content from connected systems. Most procurement teams start with file uploads and only graduate to knowledge base setup when the source content library becomes large.

Projects, Memory, and Custom GPTs, The Three-Layer Model

ChatGPT offers three distinct mechanisms for giving the AI context. They are not substitutes; they are complementary.

Memory is the always-on layer. It stores facts that apply across every ChatGPT conversation, the user's role, the organisation's structure, the procurement team's fiscal calendar, standard terminology. Memory applies to every conversation regardless of whether the conversation is inside a Project.

Projects are the workstream layer. A Project scopes context to a specific piece of work. Custom instructions, files, and conversation history inside the Project apply only to that Project. Project context layers on top of Memory inside that Project.

Custom GPTs are the capability layer. A Custom GPT is a reusable assistant, a configured version of ChatGPT that performs a specific function consistently across many Projects or standalone conversations. A "Supplier Risk Analyst" Custom GPT can be invoked from any Project or any one-off chat.

The full setup: Memory for the organisational baseline, Projects for each active workstream, Custom GPTs for the reusable capabilities. Teams using only one of the three extract some value; teams using all three in combination extract substantially more.

When to Create a Project, and When Not To

Projects are useful when the work is ongoing and context compounds over time. They are overkill when the work is one-off and the context is self-contained.

Create a Project for: a strategic category the procurement team manages on an ongoing basis; an active supplier renewal running over weeks or months; a major initiative like a consolidation programme or supplier rationalisation; a recurring deliverable like quarterly category reviews.

Use a standalone chat for: a quick research question; a one-off drafting task with no ongoing context; an experimental prompt that does not yet have a clear application.

A procurement professional with twelve to fifteen categories in their portfolio typically runs five to ten active Projects at any given time, not one per category, because some categories do not need the Project overhead, and not for every piece of work, because standalone chats are faster for quick questions.

Setting Up a Procurement Project in 15 Minutes

The first Project is the hardest because the components, briefing document, file selection, instruction design, need to be built from scratch. Subsequent Projects are faster because most of the work is reusable.

Step 1, Pick a real piece of work. The first Project should support something the procurement professional is actually doing. A current category. An upcoming renewal. An active initiative. The first Project that supports real work pays back within days.

Step 2, Draft the briefing document. Two or three pages covering company profile, procurement operating model, category structure, strategic priorities, risk appetite, and output preferences. This is the reusable artefact, once written, it goes into every subsequent Project.

Step 3, Gather the Project files. The spend overview for the category. The active supplier list with performance data. The market intelligence brief. Relevant current contracts. Usually four to six files.

Step 4, Create the Project in ChatGPT. Upload the files. Paste the briefing document content into the custom instructions field. Name the Project with a clear convention.

Step 5, Test with a real question. Ask something that should be informed by the Project context. If the output is still generic, adjust the briefing document or the files. The test-and-adjust loop usually takes a few iterations for the first Project; subsequent Projects rarely need this step.

Step 6, Use the Project for two weeks of real work. Every question related to the category or initiative goes into the Project rather than into a new chat. The Project compounds context; by the end of two weeks, the output quality is significantly better than the first answer.

After the first Project, each subsequent Project takes five to ten minutes to set up because the briefing document is already written and the file list pattern is established.

Frequently asked questions

What's the difference between a ChatGPT Project and a Custom GPT?

A Project is a persistent workspace scoped to a specific piece of work (a category, a renewal). A Custom GPT is a reusable assistant that performs a specific function (review contracts, score RFPs). Projects hold context; Custom GPTs provide capability. A procurement team typically uses both.

How many Projects should a procurement professional run?

Typically five to fifteen active at any time. Below five, the feature is under-used. Above twenty, Projects start duplicating each other and become harder to maintain.

Can whole procurement teams share Projects?

On ChatGPT Team and Enterprise plans, yes. Sharing is usually appropriate for category-level Projects and more selective for supplier-specific or highly confidential initiative Projects.