AI Business Guides
ChatGPT for Business: How to Set It Up So It Actually Helps
By Scott McIntosh, Founder of DigitalTreehouse · Updated November 2026

Most businesses "use ChatGPT." Very few have set it up. The difference is the gap between an employee pasting a customer email into a free account on their phone and a workspace where every team member has the company's voice, policies, and documents at hand, where nothing sensitive leaks into a training set, and where the ten best prompts your business relies on are one click away for everyone.
This guide is the setup we walk clients through at DigitalTreehouse. It takes about a day of focused work, costs less than one employee lunch per person per month, and it is the single highest-return AI project most small businesses have not done yet.
Why setup matters more than the subscription
MIT's Project NANDA found that workers at more than 90 percent of surveyed companies were already using personal AI tools for work, even at companies whose official AI pilots had failed. The report called it the "shadow AI economy."
For a small business that means three things are almost certainly true right now:
- Your team is already using ChatGPT.
- They are using it inconsistently, with no shared instructions, on accounts you do not control.
- At least some company or customer information has been pasted into a consumer account whose default settings allow the content to be used for model training.
Setup fixes all three. The rest of this article is the how.
Step 1: Pick the right plan (and understand the data policy)
OpenAI's pricing page currently lists six tiers: Free, Go, Plus, Pro, Business, and Enterprise. For a company, the decision comes down to one line in the comparison table.
Under "Privacy," the Free, Go, Plus, and Pro plans all say content is used to train our models, opt-out available. Business and Enterprise plans do not use your workspace data for training by default; OpenAI's enterprise privacy page is the primary reference.
That single distinction is why a business should not run on a pile of individual Plus subscriptions, no matter how convenient they were to expense.
The plans, briefly
| Plan | Who it is for | Data used for training? | Admin controls | Notes |
|---|---|---|---|---|
| Free | Individuals trying it out | Yes, unless opted out | None | Limited uploads, memory, and deep research |
| Go | Individuals, longer conversations | Yes, unless opted out | None | May include ads |
| Plus | Individual power users | Yes, unless opted out | None | Projects, scheduled tasks, custom GPTs, connectors |
| Pro | Individual researchers and developers | Yes, unless opted out | None | Highest usage limits |
| Business | Teams of 2 or more | No, by default | Admin console, SAML SSO, GPT management, company knowledge, workspace agents | Priced per user per month; annual and monthly billing |
| Enterprise | Larger or regulated organizations | No, contractually | Everything in Business plus SCIM, data residency in ten regions, enterprise key management, compliance logs | Custom pricing, contact sales |
Our recommendation for almost every business under 150 people: ChatGPT Business. It starts at two seats, it is priced per user per month at roughly the cost of a consumer Plus seat depending on billing term (check the pricing page for the current figure, since OpenAI has adjusted it more than once), and it unlocks the features that make a shared setup possible: an admin console, single sign-on, workspace-shared custom GPTs, "company knowledge" that connects the model to your files, and workspace agents.
If you are in healthcare and need a HIPAA business associate agreement, or you need data residency or key management, that is Enterprise. Everyone else can wait.
One more note on privacy. Even on Business, "not used for training" is not the same as "safe to paste anything." Your data is still transmitted to and processed by OpenAI under its terms. Write a one-page policy (Step 4) that says what may and may not go in.
Step 2: Set up the workspace the right way
Once the Business workspace exists, do these in order. Total time: about two hours.
Turn on SSO or at least domain verification. Everyone logs in with their work identity. When someone leaves, their access ends with their email.
Invite every person who will use it, including the ones who "don't do AI." The point is to bring shadow usage inside the fence. If someone has a personal Plus account for work, have them move over and cancel it.
Set workspace-level custom instructions. ChatGPT lets each user set instructions about who they are and how they want responses. At the workspace level, you want a short, shared baseline. Ours for a client typically covers:
- Company name, what it does, who it serves, and where.
- Tone: plain English, confident, specific, no hype, no em dashes, no filler.
- Formatting defaults: short paragraphs, lists only when they add clarity.
- Standing rules: never invent statistics, cite sources when asked for facts, ask a clarifying question rather than guessing when a request is ambiguous.
Create the folder structure with Projects. Projects let you group chats with files and instructions. Start with five: Sales, Marketing, Customer Service, Operations, and Leadership. Each gets its own instructions and its own reference files. (Projects and shared projects are available on all plans, so even a Plus user can do this piece; the workspace-shared version is what Business adds.)
Turn on memory, then decide what it may remember. Memory makes the assistant better over time. It also means a chat about a confidential matter can surface later. Business workspaces can control this; set the policy deliberately.
Step 3: Build the five custom GPTs every business needs
A custom GPT is a saved configuration: instructions, reference documents, and optional tools, wrapped in a name so anyone on the team can use it without knowing how to prompt. On Business, GPTs can be shared to the whole workspace and managed by an admin. This is where the setup starts paying.
Build these five first. Each one takes 20 to 40 minutes.
1. The Voice GPT
Purpose: Write anything in the company's voice. Feed it: 10 to 20 of your best emails, your about page, two or three pieces of content you are proud of, and a short "voice guide" (three things we always do, three things we never do). Instructions: "You write as [Company]. Match the attached samples in tone, sentence length, and vocabulary. Never use the following phrases: [list]. Ask one clarifying question if the goal is unclear."
Every other GPT on this list should reference the same voice guide.
2. The Customer Reply GPT
Purpose: Draft replies to customer emails, reviews, and messages. Feed it: Your policies (refunds, cancellations, warranties, hours, service area), your FAQ, and 20 real examples of good replies. Instructions: "Draft a reply. If the answer is not in the attached policies, say so and recommend escalation rather than inventing an answer. Match the customer's tone. Keep it under 150 words unless the situation requires more."
3. The Sales Follow-Up GPT
Purpose: Turn call notes or a transcript into a follow-up email, a CRM summary, and next steps. Feed it: Your service descriptions, pricing tiers, two or three proposal examples, and your objection-handling notes. Instructions: "From the notes provided, produce (1) a follow-up email in our voice that references specific things the prospect said, (2) a three-line CRM summary, (3) a list of next actions with owners and dates."
4. The Proposal and Scope GPT
Purpose: First drafts of proposals, statements of work, and quotes. Feed it: Your template, three past proposals, your pricing rules. Instructions: "Draft in the attached structure. Never state a price outside the attached pricing rules; if the request falls outside them, flag it. Mark every assumption you make in brackets."
5. The Internal Help GPT
Purpose: Answer "how do we do X here?" for the team. Feed it: SOPs, onboarding docs, tool guides, the org chart. Instructions: "Answer only from the attached documents. If the answer is not there, say who to ask."
On Business, "company knowledge" connectors can link the workspace to your Google Drive, SharePoint, or other stores so the GPT reads current documents rather than a stale upload. Use that where it is available; it removes the maintenance burden.
Step 4: Write the one-page usage policy
This is the document that keeps you out of trouble, and it fits on one page. Ours has five sections.
Allowed. Drafting, summarizing, brainstorming, research with verification, rewriting, translating, analyzing data you are permitted to share.
Not allowed without approval. Customer personal data beyond what is needed for the task, employee HR matters, unreleased financials, contracts under negotiation, anything covered by an NDA, passwords or credentials of any kind.
Never. Pasting credit card numbers, Social Security numbers, health records, or credentials. Presenting AI output as fact without verification. Sending AI-drafted customer communications without a human reading them.
Verification rule. Any statistic, quote, legal claim, or product specification produced by the assistant is checked against a primary source before it leaves the building.
Accounts. Work happens in the company workspace only. Personal accounts are not used for company work.
Have everyone read and acknowledge it. Then revisit it quarterly, because the tools change.
Step 5: Build the prompt library
Custom GPTs cover the repeated jobs. The prompt library covers the recurring one-off tasks. Keep it in a shared doc (or inside the Internal Help GPT) and organize it by role. A starting set:
For everyone
- "Summarize this thread. List decisions, open questions, and who owes what by when."
- "Rewrite this for a customer who is frustrated. Acknowledge the problem in the first sentence."
- "Turn these bullet points into a clear email. Under 120 words."
Sales
- "Here is a company website and a LinkedIn profile. Give me three specific, verifiable reasons this business might need [our service], and one opening line that references one of them."
- "Here are my notes from a discovery call. What did I forget to ask?"
Marketing
- "Here is our last newsletter and this week's three topics. Draft the issue in the same structure and voice."
- "List 15 questions our customers ask before buying [service]. Group them by stage of the decision."
Operations
- "Here is a process description. Turn it into a numbered SOP with a checklist at the end."
- "Read these five vendor quotes. Build a comparison table with total cost, terms, and gaps."
Leadership
- "Here are this month's numbers and last month's. Write five sentences on what changed and two questions I should ask my team."
Prompts that work get promoted into GPTs. Prompts that do not get deleted. The library should shrink over time as the GPTs get better.
Step 6: Connect it to the tools you already use
ChatGPT on Plus and above can connect to internal tools through apps and connectors, and Business adds a connector registry and workspace agents. What this looks like in practice:
- Email and calendar: Ask the assistant to find the thread with a vendor, summarize it, and draft the reply. Have it check availability before proposing meeting times.
- Drive and documents: Ask questions of your own files without uploading them each time.
- Spreadsheets: The Excel, Word, PowerPoint, and Google Sheets extensions let the assistant work inside the document rather than in a chat window.
- Scheduled tasks: Available from the Go plan up. "Every Monday at 7 a.m., summarize last week's customer reviews and flag anything under four stars." "Every Friday, draft next week's social posts from the content calendar."
Turn on connectors one at a time, and only for the people who need them. Each connector is a new door into your data.
Step 7: Train the team in 90 minutes
Adoption is a habit problem, not a knowledge problem. The training that works is short and practical.
Session 1 (45 minutes, everyone). Show the workspace. Walk through the policy. Have each person complete one real task from their own work using a custom GPT while you watch. The goal is that everyone leaves having saved ten minutes on something real.
Session 2 (45 minutes, by role). Sales does the follow-up flow on a real call. Service drafts three real replies. Ops builds one SOP. Marketing drafts one real piece.
Then: A weekly five-minute "what did AI do for you this week" in your existing team meeting. The stories spread the habits faster than any training deck.
The Census Bureau's Business Trends and Outlook Survey found that businesses using AI were far more likely to rate their current performance as excellent than non-users. Correlation, not proof, but it points the right direction: the businesses that use it well are the ones that made it part of how work happens.
Step 8: Measure it
You cannot manage the value of something you do not measure. Three numbers, tracked monthly:
- Active users in the workspace (the admin console shows this on Business). If it drops, the habit is dying.
- Hours saved per week, self-reported in the five-minute weekly check-in. Rough is fine.
- One quality number per team. Customer reply time, follow-up emails sent within 24 hours of a call, proposals drafted per week.
Multiply hours saved by loaded hourly cost. Compare to the monthly bill. For a ten-person team at roughly $25 per seat, the workspace costs about $250 a month. If it saves each person one hour a week, that is 40 hours a month, or roughly $2,000 in recovered time at $50 an hour. The subscription pays for itself in the first week of the month.
The ten most common mistakes
- Buying individual Plus seats for a team. No admin control, no shared GPTs, and consumer training defaults.
- Skipping the policy. Someone will paste something they should not.
- No voice guide. Every output sounds like a generic assistant, and the team stops using it.
- Uploading stale documents into GPTs and never updating them. Use company knowledge connectors where you can.
- Treating output as finished. It is a first draft. Always.
- Not verifying facts. The assistant will produce a plausible statistic with a straight face.
- One person "owns AI" and nobody else touches it. The value is in wide, shallow use.
- Building 30 GPTs. Build five good ones.
- No connectors. Copy-pasting between tools erases half the benefit.
- Expecting ChatGPT to run processes. It is an assistant. When you need something to happen automatically, every time, without a person starting it, you have outgrown the chat window (see below).
When you have outgrown ChatGPT
ChatGPT is the right first step because it is cheap, immediate, and it teaches your team what AI is good at. But it has a ceiling. It waits for a person to open a window and type. It does not watch your inbox at 2 a.m., respond to a lead in 90 seconds, read every invoice that arrives, or update your CRM after every call unless someone tells it to.
When you find the same task being done in ChatGPT every day, by the same person, with the same inputs, that task is a candidate for an automation or an agent that runs without anyone starting it. That is the point where an AI agency earns its fee: taking the workflows your team discovered in the chat window and turning them into systems that run on their own.
At DigitalTreehouse, roughly half of the automations we build start as a prompt someone was running by hand in ChatGPT every morning. The setup in this article is how you find them.
Frequently asked questions
Is ChatGPT Business worth it for a two-person company?
Yes, and the plan starts at two seats for exactly this reason. The value is not the model access, which is similar to Plus. It is the default exclusion of your data from training, the shared custom GPTs, and the admin console. A two-person firm that handles customer information should not be running on consumer accounts regardless of size.
Can I just opt out of training on a Plus account instead?
You can, and every person on your team would have to do it individually, verify it stayed set, and never use a personal account for work. The Business plan makes the safe setting the default and gives you a console to see who is using what. For a team, the admin visibility alone is worth the difference.
Should we use ChatGPT, Claude, or Gemini?
For a shared team workspace, all three offer business tiers with no-training defaults and comparable admin features. Choose based on where your documents live (Gemini integrates tightly with Google Workspace, Microsoft Copilot with Microsoft 365) and which model your team finds produces better first drafts for your kind of work. Many businesses run a primary workspace and give a few power users access to a second. The setup steps in this article apply to any of them.
How do I stop the team from pasting sensitive information?
The one-page policy, acknowledged by everyone, plus custom GPTs that make the right way the easy way. When the Customer Reply GPT already has the policies loaded, nobody needs to paste a customer's full account history to get a good answer. Design the tools so the safe path is the fast path.
What if my industry has confidentiality rules?
Law, accounting, healthcare, and financial services each have specific obligations. Business-tier plans with no-training defaults address the most common concern, but a HIPAA business associate agreement requires Enterprise, and some professional rules require client consent before using third-party AI on their matters. Check with your regulator or bar before rolling out, and write those constraints into the policy.
How often should the custom GPTs be updated?
Quarterly at minimum, and immediately when a policy, price, or service changes. A GPT loaded with last year's pricing will confidently quote it. Where your plan supports company knowledge connectors to live documents, use them so the GPT reads the current file rather than a stale upload.
We set it up and nobody uses it. What went wrong?
Almost always one of three things: no custom GPTs (so every use requires prompting skill), no voice guide (so output sounds generic and gets discarded), or no weekly moment where people share what worked. Fix those three and usage recovers. The training session is less important than the habit loop that follows it.
Sources
- OpenAI, ChatGPT Pricing (plan features, privacy comparison, and FAQ): chatgpt.com/pricing
- OpenAI, Enterprise Privacy: openai.com/enterprise-privacy
- OpenAI, How your data is used to improve model performance: openai.com
- MIT Project NANDA, The GenAI Divide (shadow AI finding), as reported by Space Daily and Fortune
- U.S. Census Bureau, Business Trends and Outlook Survey AI data (April 2026 release), as summarized by SelfEmployed.com
- Stocktwits via Yahoo Tech, ChatGPT tops 1B weekly users (August 6, 2026): tech.yahoo.com


