ChatGPT for business and what to build instead

By Luca Planchais · Published 17 September 2026 · 6 min read

ChatGPT has become a normal part of how individual people work: drafting, summarising, thinking through a problem out loud. It has not become, on its own, a business system, and treating it as one is where most of the disappointment and most of the risk shows up. This is a practical guide to where it genuinely helps, what to check before using it with real business information, and when the honest answer is to build something else.

Good uses for individuals

Drafting is the clearest win. A first draft of an email, a job description, a proposal outline or a social post is faster to produce and often easier to edit than to write from nothing, and the person using it stays fully in control of what actually gets sent.

Thinking things through works well too. Talking through a pricing decision, a difficult message to a client, or the structure of a report, with something that asks reasonable follow-up questions, is genuinely useful preparation before a real conversation or a real piece of writing.

Summarising and reformatting is another strong, low-risk use: turning a long document into a shorter brief, converting notes into a structured list, or rewriting something in a different tone for a different audience. Research and explanation, checked afterwards rather than trusted blindly, rounds out the list of everyday uses that hold up well across most roles.

These uses share a pattern worth noticing: a person starts with the output, checks it against what they actually know, and edits before it goes anywhere near a client, a candidate or a colleague. That habit is what keeps a general-purpose tool useful rather than risky, and it is worth stating explicitly to a team rather than assuming everyone applies it by default.

Data and confidentiality settings to check first

Before anyone pastes client information, financial figures or anything confidential into ChatGPT, check which product and which settings are actually in use, because the answer is not the same for every plan.

OpenAI's own published policy states that ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu and the API platform do not use your prompts and outputs to train its models by default. Business data on these plans is described as remaining confidential and owned by the organisation, with encryption at rest and in transit. On the free or personal consumer product, training participation is controlled by an account-level setting, and it is worth every business checking that setting directly rather than assuming it, since defaults and product tiers change.

Two practical checks before rolling ChatGPT out more widely: confirm which plan your team is actually using, since a personal account someone signed up for individually may have different data handling to a business licence bought centrally; and confirm what your own client contracts or data protection obligations say about pasting third-party information into any external tool, which is a business decision independent of what OpenAI's policy allows.

Where it stops

The limits matter as much as the capabilities, and they are the reason a lot of "we tried ChatGPT for that" projects stall.

No access to your CRM. Standard ChatGPT cannot read your pipeline, update a client record or check what stage a deal is at, unless a developer has deliberately built and connected that access, which most individual usage has not.

No approvals. It cannot enforce who is allowed to sign off a quote, release a document, or authorise a spend. It can help draft the message that requests approval; it cannot be the approval step.

No memory of your process, unless built. A fresh conversation does not know your company's specific policies, your tone of voice guide, or how your business actually handles a particular type of enquiry, unless that context is supplied every time or a system has been built to carry it forward automatically.

These are not flaws so much as a description of what a general-purpose assistant is: a capable, stateless tool a person operates, not a system that runs part of a business on its own.

A short example makes the boundary concrete. Ask ChatGPT to draft a reply to a candidate enquiry and it will produce a perfectly usable email, once you have supplied the details of the role and the candidate yourself. Ask it to check whether that candidate has already been contacted this week, update their record, and notify the consultant who owns the account, and it cannot, because none of that lives anywhere it can reach. The first task is drafting; the second is a connected workflow, and the two need different tools even though both start with typing a request in plain English.

When a custom system or a per-person assistant is the right next step

A few signals reliably indicate that ad hoc ChatGPT use has reached its limit and something more deliberate is worth building.

The same prompt keeps getting typed out. If several people are independently writing near-identical prompts for the same recurring task, that is a sign the task deserves a proper workflow or a small custom tool, not repeated manual effort dressed up as flexibility.

The task needs live business data. The moment a task genuinely needs to read or write to a CRM, a booking system or a database, general-purpose ChatGPT usage stops being enough, because that connection has to be built and secured deliberately.

A decision needs a record. Anything that needs an audit trail, a consistent process every time, or a way to prove what happened and why, needs a system built for that, not a conversation that disappears from someone's chat history.

One person needs ongoing, personal support across email, calendar and messaging. This is a different shape of problem to a company-wide workflow, and it is closer to what a per-person AI assistant such as SuperIntern is built for: drafting and scheduling inside one individual's accounts, rather than a shared business process.

Where the need is a genuine business system, connected to your data with rules, approvals and reporting built in, that is what a custom AI system is for, and it is a different build to giving a team access to ChatGPT and hoping the gaps close themselves. If the need spans several roles across a business rather than one workflow, it is worth reading how to build an AI workforce for how those pieces fit together.

The practical split

Most businesses that get real value end up running two things side by side rather than choosing one: individuals keep using ChatGPT for drafting, thinking and summarising, because it is genuinely good at that and does not need to be built or maintained. Alongside it, the business builds or commissions the specific, connected systems that need to touch real data, enforce a process or run unattended, because that is work a general-purpose chat tool was never designed to do on its own.

Questions people ask