Can ChatGPT Help You Run Interior Design Projects?

TL;DR: ChatGPT can help you think and write about a project — phase planning, client emails, summarising contracts, talking through a decision. It cannot run one, because running a project means holding state: what is specified, approved, ordered, delivered, invoiced. A chat has no authoritative record and no memory of your job between sessions, so anything it produces still has to be maintained somewhere else. The useful setup is an assistant for language plus something that actually holds the project — or a connector so the assistant can act on the real thing.

The distinction that matters

“Running a project” sounds like one activity. It is two, and they have opposite requirements.

Thinking and communicating. What phases should this job have? How do I explain a four-week delay to an anxious client? What does this supplier quote actually commit me to? These are language problems, and a good assistant is genuinely excellent at them.

Holding state. Which of the two hundred items are approved. Which are ordered. What deposit was paid against which purchase order. What the client has actually seen. This is not language — it is a record, and it has to be authoritative, meaning when it disagrees with your memory the record wins.

ChatGPT is very good at the first and structurally cannot do the second. Not because it is missing a feature, but because a conversation is not a system of record.

Why “just keep the project in a chat” fails

Studios do try this, and it fails in a specific and predictable way.

There is no authority. If your chat says the sofa is ordered and your inbox says the supplier never received the PO, the chat has no way to know it is wrong. It will restate the incorrect fact confidently, because consistency with earlier context is exactly what it is built to do.

Divergence is silent. Nothing reconciles the conversation against reality. Every real action taken outside the chat — an email to a supplier, a delivery, a payment — is invisible to it unless you remember to narrate it back.

Recall degrades. Long conversations lose earlier detail, and memory features summarise rather than store. A project’s state is precisely the thing you cannot afford summarised.

It is single-player. If a colleague orders something, your chat does not know.

The tell is simple: if you find yourself re-telling the assistant what has happened so it can answer correctly, you are the system of record and it is the interface. That is a fine arrangement for one person on a small job, and it does not scale past that.

What it is genuinely worth using for

Four things, all high value and all language:

Correspondence. Client updates, supplier chasers, the difficult email about a delay or an overrun. Fluent, fast, and easy to make sound like you.

Planning and scope. “Draft a phase structure for a three-bedroom renovation with a six-month timeline.” A sensible starting framework to edit, which beats a blank page.

Summarising. A contract into obligations and dates. A twelve-page quote into the five things that matter. A long client email into decisions and questions.

Thinking out loud. Talking through a layout constraint or a client dynamic. Genuinely useful and consistently undersold, because it produces no artifact.

Custom GPTs and project features help here — they hold your tone, your standard scope language, your usual phase structure. That is continuity of instructions, which is real. It is not continuity of truth.

What has to live somewhere else

Anything that determines whether you spend money:

• The FF&E schedule — every item with vendor, price, discount, lead time, status
• Approval decisions — what the client actually agreed to
• Purchase orders, deposits, deliveries
• Invoices and what has been paid

For a solo designer with small projects, that “somewhere else” can honestly be a spreadsheet — our free FF&E schedule template is built for exactly that, and ChatGPT plus a good spreadsheet is a reasonable setup that costs nothing.

The thresholds where it stops working are about fifty line items or the second person. At that point the thing that has to change is where the project lives, not whether you use an assistant.

The third option: connect the assistant to the system

This is the part most answers to this question miss.

You do not have to choose between a chat with no project data and a platform with no chat. If your design software exposes a connector, the assistant can read and act on the real thing.

Atelier — our product, so weigh accordingly — publishes an MCP server at api.getatelier.design/mcp that ChatGPT and Claude connect to over standard OAuth, no API keys. Connected, the assistant works on real records: reading and updating projects, schedules and products, extracting a product from a supplier URL into your library, assembling pinboards from products you have chosen. It inherits your permissions, so it can only reach what you can.

That converts the assistant from a place you re-type things into an interface onto a system that holds state — which is the actual answer to “can ChatGPT run my projects.” It can drive something that does.

$39 per seat per month billed annually ($49 monthly), free trial, no card. It does not pick your products or design for you, does no 3D rendering, and is not an accounting system.

The honest recommendation

Keep ChatGPT. It costs almost nothing and it is excellent for the language half. Every studio should have an assistant.

Do not make it the record. The moment a decision costs money, it needs to live in something authoritative — a spreadsheet at small scale, a platform past the thresholds above.

Connect the two if you can. That gets the conversational interface without the divergence, which is the arrangement that actually holds up.

Disclosure: Atelier is our product, and this article says plainly that ChatGPT plus a spreadsheet is a reasonable setup below fifty line items and one person, which is a recommendation to not buy anything.