How to Build a Custom GPT for Your Small Business (A Real Step-by-Step Guide)

The short version: A custom GPT is a version of ChatGPT you train with your own instructions, tone, and knowledge so it responds like a member of your team rather than a generic AI. Small businesses can build one in under two hours with no coding, and the productivity gains are measurable from week one. The hard part is not the build, it is knowing exactly what you want it to do before you start.

Why I built my first custom GPT and what happened

About eighteen months ago I was spending roughly three hours every single week writing first drafts of client-facing emails. Proposals, follow-ups, polite-but-firm late-payment nudges, onboarding sequences. Three hours. I am a one-woman business. Three hours is a Tuesday.

I had been using ChatGPT already, but the generic version kept writing in a voice that sounded like a bored American corporate intern. I would spend half my time editing out the filler words and reinjecting my actual personality. The output was saving me maybe twenty minutes a week, not three hours.

Then I built a custom GPT I called "Lilach Email Brain." I spent about ninety minutes setting it up. The following week I got through my email drafting in forty-five minutes. That single tool is now part of my daily workflow, and it is one of the reasons I have been able to rebuild my consultancy at a pace I could not have managed with generic AI.

That is what a well-built custom GPT does. It does not replace you. It replaces the worst, most repetitive version of your working day.

What a custom GPT is

A custom GPT is a configuration you create inside ChatGPT (you need a Plus or Team subscription, currently $20 or $30 per month respectively). You give it a name, a set of instructions, an optional knowledge base of uploaded files, and a defined set of capabilities. Once built, it lives in your ChatGPT sidebar and anyone you share it with can use it.

It is not a separate AI model. It is the same GPT-4o engine underneath, but it runs through a filter of your instructions every single time. Think of it like hiring a brilliant generalist and giving them a very thorough induction document. The intelligence is there already. You are just pointing it in the right direction.

For small businesses this matters because you are not paying for custom model training (which costs tens of thousands of pounds). You are paying $20 a month and investing your time once.

What small businesses are using custom GPTs for in 2026

Before you build one, it helps to see what is working in the real world. According to the data I covered in my AI Search Demand Report 2026, which analysed 3.4 million searches, the most searched AI use cases for small businesses in 2026 cluster around customer communication, content creation, and internal knowledge management. Custom GPTs are the practical answer to all three.

Here is what I see working most consistently:

• Customer FAQ bots: Upload your full FAQ document, your returns policy, your pricing sheet, and your shipping times. The GPT answers customer questions in your brand voice without anyone on your team touching it.
• Proposal and pitch drafters: Feed it your service descriptions, your case studies, your pricing tiers. Ask it to draft a proposal for a new client brief. It comes back with something 80% done in three minutes.
• Social content assistants: Give it your brand guidelines, your tone of voice document, your content pillars. It drafts captions, threads, and newsletter sections that sound like you wrote them.
• Onboarding assistants: Upload your internal processes, your style guides, your systems. New team members or contractors can ask it questions and get accurate answers instantly.
• First-line customer service: This is where a lot of small businesses are starting, and it connects directly to something I have written about before: the real experience of doing customer service work is exhausting and repetitive. A custom GPT does not replace human judgment for complex complaints, but it handles tier-one queries well.

Before you build: the one question most people skip

Most guides jump straight to the builder interface. I want to stop you here for five minutes.

The question you must answer before you open the builder is this: what is the single most repetitive thing my business does with words?

Not "what could AI help with." The specific, daily, time-consuming task that you or someone on your team does with text, over and over, that produces similar outputs each time.

If you cannot name it specifically, your custom GPT will be too broad. A custom GPT that does "marketing stuff" is useless. A custom GPT that drafts Instagram captions for a UK-based artisan bakery in a warm, slightly cheeky tone using only ingredients the bakery stocks, drawing from an uploaded product list, is useful.

Specificity is everything. The more precisely you define the job, the better the output.

Step-by-step: how to build your first custom GPT

Step 1: Log in and open the builder

You need a ChatGPT Plus account ($20/month) or a Team account ($30/user/month). Go to chat.openai.com, click "Explore GPTs" in the left sidebar, then click "Create" in the top right corner. You will land in the GPT builder, which has two tabs: "Create" (a conversational setup wizard) and "Configure" (the manual setup). Use Configure. It gives you full control.

Step 2: Write your system instructions

This is the most important part and where most people underinvest their time. The instructions box is where you tell the GPT who it is, what it does, what it never does, and how it sounds.

A good instruction block covers six things:

• Role: "You are the customer service assistant for [Business Name], a UK-based independent bookshop."
• Tone: "Write in a warm, direct, slightly informal tone. No corporate jargon. Short sentences. Sound like a knowledgeable friend, not a policy document."
• Scope: "You help with order enquiries, book recommendations, event information, and general questions about the shop. You do not handle complaints about damaged goods. Direct those to [email address]."
• Format: "Keep responses under 150 words unless the question requires a detailed list. Use bullet points only when listing three or more items."
• What to avoid: "Never make up stock availability. If you are not certain, say 'I will check and get back to you.'"
• Escalation rules: "If the customer is upset or uses the word 'refund,' end your response with 'A member of our team will be in touch within one working day.'"

Write this out before you open the builder. Seriously. Draft it in a Word document or a Notes app first. The quality of your instructions determines the quality of every single output. I spent forty-five minutes on my email GPT instructions and it showed in the results from day one.

Step 3: Upload your knowledge base

Click "Add files" under the Knowledge section. You can upload PDFs, Word documents, text files, and spreadsheets. Here is what to upload depending on your use case:

• Customer service GPT: Your full FAQ, your current pricing, your delivery/returns policy, your contact details, your opening hours.
• Content GPT: Your brand guidelines, your tone of voice document, three to five examples of content you love that you have written yourself, your content calendar themes.
• Proposal GPT: Your service descriptions, your standard pricing tiers, two or three anonymised past proposals, your most common objections and how you answer them.
• Onboarding GPT: Your internal process documents, your systems guide, your style guide, your team org chart, your supplier contacts list.

Keep your files clean and clearly named. If you upload a messy 80-page PDF with tables that do not parse well, the GPT will struggle. Better to upload five clean, well-structured documents than one sprawling mess.

Step 4: Set the conversation starters

The builder lets you add four suggested prompts that appear when someone first opens the GPT. Use these. They guide users toward the right behaviour and reduce off-brief requests.

For a customer service GPT the starters might be: "Where is my order?", "What are your opening hours?", "Do you stock [book title]?", "How do I return something?"

For an internal proposal GPT: "Draft a proposal for a new social media client," "Write a follow-up email after a discovery call," "Summarise our services in two paragraphs for a new lead."

Step 5: Test it hard before you share it

Use the preview panel on the right side of the builder. Ask it easy questions and hard ones. Ask it things that are outside its brief and check that it handles them gracefully rather than inventing answers. Ask it something that requires information from your uploaded files and verify the accuracy.

The two tests I always run:

• The edge case test: Ask it something ambiguous or slightly off-topic. Does it stay in its lane, or does it start hallucinating information?
• The tone test: Copy a response into a blank document and read it out loud. Does it sound like your business, or does it sound like every other AI response you have ever read?

If it fails either test, go back to your instructions and be more specific. Ninety percent of the time the fix is adding one more specific rule to the instructions block.

Step 6: Share it with your team or embed it

When you publish the GPT you can set it to "Only me," "Anyone with a link," or "Everyone" (public). For most small business uses, "Anyone with a link" is right. You share the link with your team, your VA, your contractors. They open it in their own ChatGPT account and use it.

If you want to embed it on your website for customer use, you will need a third-party widget layer since OpenAI does not offer a native embed for custom GPTs at the consumer tier. This is a limitation worth knowing upfront.

The honest thing most articles will not tell you

Here it is: a custom GPT is only as good as the documents you feed it, and most small businesses do not have good documents.

I have helped four small business clients build custom GPTs in the past year. In every single case, the first conversation was not "how do we set this up." It was "let us write your FAQ " or "let us document your actual tone of voice because right now it only exists in your head."

The custom GPT build forces a documentation exercise that most small businesses have been avoiding for years. This is one of its hidden benefits: you cannot build a useful knowledge base without first articulating what your business is, does, and sounds like. But it is also the reason so many custom GPTs are mediocre. People upload whatever they have, which is often scrappy, incomplete, or outdated, and then wonder why the outputs are scrappy, incomplete, or outdated.

Before you start the build, spend a few hours getting your foundational documents right. A clean, current FAQ. A one-page tone of voice guide with three examples of good copy and three examples of what you want to avoid. An up-to-date service or product list with accurate pricing.

Do that first. Then build the GPT. The difference in output quality will be significant.

A word on using custom GPTs for customer-facing roles

If you are thinking about using a custom GPT to handle customer queries, it is worth understanding the full picture of what that role involves. I have written extensively about the demands and realities of customer service representative roles, and one thing that stands out is the sheer volume of repetitive, low-complexity queries that burn people out. Custom GPTs handle that tier extremely well.

What they do not handle well: genuine emotional distress, nuanced complaints, situations that require judgment calls your policy documents do not cover. A customer who is upset because their order arrived broken and is now threatening to leave a negative review is not a custom GPT situation. That is a human situation.

The smart model is tiered. Custom GPT handles tier one (tracking, basic FAQs, availability questions). A human handles anything that requires empathy, judgment, or retention thinking. This also connects to how customer service work from home is evolving in 2026: the human role is shifting from answering repetitive questions to managing escalations and building relationships. Custom GPTs accelerate that shift.

How to get more from your custom GPT over time

Most people build a GPT, use it for a week, and then forget to update it. This is a mistake.

Set a monthly reminder to review your GPT. Ask yourself:

• Have my prices changed? Update the pricing document in the knowledge base.
• Have we launched a new service or product? Add it.
• Are there questions the GPT keeps getting wrong or mishandling? Add a specific rule to the instructions to address that failure mode.
• Is there a new type of query that is coming up repeatedly that the GPT is not set up to handle? Consider whether to expand its brief or build a second, separate GPT for that use case.

I maintain three separate custom GPTs for my own business now: one for email drafting, one for social content, and one for client-facing educational content. Each does one job. None of them tries to do everything. That constraint is the secret to why they work.

Custom GPTs and your wider marketing stack

One thing worth saying: a custom GPT is not a social media strategy. I see small business owners building a content GPT and thinking they have solved their marketing. They have not. They have solved the drafting problem. The strategy problem, the audience problem, the distribution problem: those are still there.

If you are unclear on the difference between having a content tool and having a marketing approach, read my piece on why most businesses are doing social media marketing backwards. A custom GPT makes executing a good strategy faster. It does not create the strategy for you.

What it costs and what you get back

ChatGPT Plus: $20/month. ChatGPT Team (if you want shared GPTs across a small team): $30/user/month.

Time to build a solid custom GPT: ninety minutes to three hours depending on how much document prep you need to do first.

Time saved per week for my email GPT alone: roughly two hours and fifteen minutes, which I measured across four weeks by logging time before and after the build.

At my consulting rate that two-plus hours represents real money back in my working week. Even at minimum wage in the UK ($11.44/hour as of 2026), two hours a week is roughly $1,200 a year in time recovered for a $240 annual tool subscription. The maths work at almost any billing rate.

For a small team of five people each using a shared GPT for customer queries and content drafts, the savings compound quickly. Firms using Team accounts report cutting first-draft content time by 60 to 70 percent in the first month. That is not a marketing claim. That is what I hear consistently from clients.

The one thing to get right on day one

If you take nothing else from this article, take this: write your instructions as if you are onboarding a new member of staff who is brilliant but knows nothing about your business and will interpret every ambiguous instruction in the most literal possible way.

Do not write "be friendly." Write "use the customer's first name in the first line, keep sentences under fifteen words, and end every response with an offer to help further."

Do not write "avoid sounding too formal." Write "never use the words 'furthermore,' 'herewith,' 'please do not hesitate to contact us,' or 'I hope this email finds you well.'"

Specific beats general. Every. Single. Time.

There is also a useful parallel here to how the best customer service representatives working from home succeed: the ones who perform best are the ones who have been given the most specific guidance about what good looks like, not the ones handed a vague brief and told to use their judgment. Your custom GPT is no different.

Frequently asked questions

Do I need to know how to code to build a custom GPT?

No. The entire build happens through a text-based interface inside ChatGPT. You write instructions in plain English, upload documents, and click publish. There is no coding involved at any stage unless you choose to add custom API actions, which is an optional advanced feature most small businesses will never need.

Can I use a custom GPT to replace my customer service team?

Not entirely and not wisely. Custom GPTs handle repetitive, low-complexity queries well: order tracking, FAQs, availability checks, booking information. They are not reliable for emotional complaints, nuanced situations, or anything requiring real judgment. The smart approach is a tiered model where the GPT handles tier one and humans handle escalations.

How many documents can I upload to the knowledge base?

OpenAI allows up to 20 files per custom GPT, with a maximum file size of 512 MB per file. In practice, the quality of the documents matters far more than the quantity. Five well-structured, accurate, up-to-date documents will outperform twenty messy ones every time.

How often should I update my custom GPT?

Review it monthly at minimum. Update the knowledge base files any time your pricing, products, policies, or services change. Review the instructions quarterly and adjust any rules that are producing outputs you keep having to edit. A custom GPT that was perfect in January will be noticeably out of date by April if you do not maintain it.

I go much deeper on this in the AI marketing guide.

Related reading: How to Set Your Rates as a New Virtual Assistant and Ecommerce Virtual Assistant: What They Do and What They Cost in 2026.