ChatGPT for SMEs: Your Business Is Probably Using It Wrong

ChatGPT in the workplace — it is always the same story. A team member asks a question, gets a generic answer, and concludes that "AI is not quite there yet." The problem is not the tool. It is that it is being used like a souped-up search engine, when it could become a genuine persistent work assistant.

Projects, memory, Custom Instructions, custom GPTs: most SMEs are unaware of these features or do not know how to use them. The result: they capture maybe 10% of the tool's value. This guide shows you concretely how to move from "I ask ChatGPT a question" to "ChatGPT knows my business and helps me every day."

The "Enhanced Search Engine" Trap

Watch how most people use ChatGPT at work. They open a new conversation, type a question ("What is ISO 9001?"), read the answer, close the window. The next day, they do it again. No memory, no context, no personalization.

It is exactly like using a smartphone only to make phone calls. It works, but you are missing 90% of the value.

The real power of ChatGPT for an SME is not answering one-off questions. It is becoming an assistant that:

• Knows your business — your products, your processes, your industry jargon
• Remembers your preferences — your tone, your deliverable format, your regular contacts
• Integrates into your workflows — not an extra task, but an accelerator on existing ones

To get there, you need to move out of question-and-answer mode and configure the tool properly. Here is how.

ChatGPT Projects: Your Business Knowledge Base

The most underused feature of ChatGPT in a professional context is Projects. Launched by OpenAI to bring structure to work inside the tool, it lets you create dedicated workspaces with reference files and custom instructions.

Concretely, a ChatGPT Project is:

• Up to 25 reference files — PDFs, Word docs, Excel, code — that ChatGPT automatically consults when answering your questions
• Custom instructions — "You are an expert in responding to RFPs in the construction sector", "Always cite internal sources in your answers", "Format deliverables according to our brand standards"
• An organized conversation history — all discussions tied to that project are grouped and accessible

How to Create an Effective Project in 10 Minutes

1. Identify a specific use case — not "my whole company" but "preparing client meetings" or "drafting quotes"
2. Select 5 to 15 key documents — the files you actually consult daily, not your entire document archive
3. Write clear instructions — describe the role ("You are my sales assistant"), the context ("We sell X solutions to the Y sector"), and the expected format ("Reply in bullet points with product references")
4. Test with 3 questions you already know the answer to — this is the best calibration check. If the answers are accurate and relevant, you are ready

Custom Instructions and Memory: the Tool That Learns Your Habits

Beyond Projects, two features transform ChatGPT from a generic tool into a personalized assistant: Custom Instructions and memory.

Custom Instructions: Your Permanent Briefing

Custom Instructions are a block of text that ChatGPT reads before every conversation. They apply globally, not just within a Project. This is your way of telling the tool who you are and how you want it to respond.

Examples of effective Custom Instructions for an SME:

• "I run a 30-person SME in the construction sector based in the south of England. Our clients are local authorities and property developers."
• "Always respond directly and actionably. Skip generic introductions. If you are not sure, say so."
• "For emails, use a professional but not distant tone. First-name basis with colleagues, formal with clients."

In 5 lines, you have just transformed a generic chatbot into an assistant that understands your context. Every response will now be calibrated to your reality, without needing to re-explain each time.

Memory: ChatGPT Remembers What You Tell It

The memory feature lets ChatGPT retain information from one conversation to the next. When you mention that your company uses Salesforce, that your sales cycle runs three months, or that you prefer tables over bullet points, ChatGPT stores that and applies it going forward.

This is not trivial. It means the tool becomes more relevant over time, much like a team member who grows into the role. After a few weeks of use, you no longer need to state your context: ChatGPT already knows it.

Custom GPTs: Build Your Own Specialized Assistants

GPTs (formerly Custom GPTs) are one of ChatGPT's most powerful features for businesses. The concept: you create a specialized assistant with its own instructions, its own reference documents, and even connections to external tools via Actions.

Unlike a Project, which is personal, a GPT can be shared with your entire team (on Team or Enterprise plans). That is the difference between "my assistant" and "the sales team's assistant."

Concrete GPT Use Cases for SMEs

Actions: Connecting ChatGPT to Your Tools

Actions let a GPT call external APIs. In plain terms, your GPT can query a third-party tool, send data, or automatically trigger an action — for example, pulling a client's details from your CRM, creating a ticket in your project management tool, or querying an internal database.

In practice, configuring Actions requires technical skills (API specification, authentication). It is not plug-and-play. But for SMEs that want to go beyond conversation, it is a significant automation lever.

What ChatGPT Does Badly (and Nobody Tells You)

Articles about ChatGPT tend to list features without talking about the real friction points. Here is what we observe with the SMEs we work with.

Hallucinations Persist

ChatGPT sometimes invents information with unsettling confidence. Numbers, legal references, product names that do not exist. The most recent models (GPT-4o, o3) are better, but the problem is not solved. Even OpenAI acknowledges that hallucinations remain a major challenge. For any critical content (legal, technical, financial), human verification remains essential.

Long Context Remains Approximate

Even with Projects and loaded files, ChatGPT tends to "lose" parts of long documents or to favor information at the beginning and end of a text (a known LLM bias). If your critical information is buried in the middle of a 50-page document, it may miss it. For precise long-document analysis, Claude from Anthropic is often more reliable with its 200,000-token context window.

Results Vary Query to Query

Ask the same question twice and you will get two different answers. That is the nature of language models. For tasks where reproducibility matters (standardized reports, calculations, quality processes), this is a genuine problem. The solution: very precise instructions and structured templates in your Projects.

Data Privacy Remains a Real Topic

With the free or Plus plan, your conversations may be used to train OpenAI's models (unless manually disabled). OpenAI's privacy policy is clear on this. With Team or Enterprise, data is not used for training. But in all cases, OpenAI is a US company: your data is hosted in the United States and subject to the CLOUD Act. For SMEs with data sovereignty requirements, Mistral's Le Chat (hosted in France) is an alternative worth considering.

Costs Escalate Quickly

One Plus user: $20/month. A team of 10 on Team: $250/month (roughly €230). Add the message limits that push some users toward the Pro plan at $200/month, and the bill becomes serious. Before deploying at scale, do a proper ROI calculation on time saved versus subscription cost.

ChatGPT vs. Claude vs. Mistral: Which Tool for Which Use Case

The question is not "which is the best" but "which fits my need." Here is an honest comparison based on our field experience.

Our pragmatic recommendation: start with ChatGPT if your needs are diverse (writing, research, images, automation). Test Claude if your primary use case is analysis of large documents. Consider Mistral if data sovereignty is a requirement, not just a preference. For a deeper comparison, see our full AI tools comparison.

ChatGPT Plans in 2026: Which One to Choose for Your SME

The right plan depends on three factors: number of users, daily usage intensity, and your privacy requirements.

Using ChatGPT at Work: 5 Rules for Concrete Results

Here are the principles we apply with our clients to maximize ChatGPT's value in the business. Not theory — rules tested in the field.

1. Contextualize Before You Question

Instead of asking "Write a follow-up email," prefix with context: "You are the sales manager of a medical equipment SME. The client is a public hospital that requested a quote 3 weeks ago. Write a professional follow-up email that is persistent but not pushy." The quality difference is dramatic.

2. Use ChatGPT in Conversation Mode, Not Query Mode

Do not chase the perfect answer on the first try. Start broad, then refine: "Good, but the tone is too formal. Cut it by 30%. Add a point about delivery timelines." ChatGPT excels when you guide it iteratively.

3. Give It Examples of What You Expect

If you want ChatGPT to write in your style, show it an example: "Here is an email I wrote last week that I was happy with. Write the next one in the same style." This is more effective than 10 lines of abstract instructions.

4. Verify Everything Critical

ChatGPT is an assistant, not an oracle. Numbers, legal references, technical data — verify them. Use it to accelerate production, not to replace your expertise.

5. Train Your Team, Not Just the Early Adopters

The biggest obstacle to ChatGPT adoption in an SME is not the technology. It is the fact that most team members do not know how to use it effectively. 20 minutes of hands-on training on Custom Instructions and Projects dramatically shifts both adoption rates and output quality.

Where to Start: a 4-Week Action Plan

If you want to deploy ChatGPT seriously in your SME, here is a realistic plan.

1. Week 1: Identify 3 priority use cases. Not the most ambitious — the most repetitive. Email drafting, document summarization, meeting prep, internal support. Cases where the time gain is immediate and measurable.
2. Week 2: Set up the environment. Create global Custom Instructions, one Project per use case, and a shared GPT if you are on Team. Load your reference documents.
3. Week 3: Pilot with 3 to 5 users. Your most motivated team members test on their real tasks. Collect feedback: what works, what does not, what adjustments are needed.
4. Week 4: Measure and decide. Compare time spent before and after on targeted tasks. If the gain is real, expand. If not, adjust the use cases or configuration before rolling out more broadly.

This is the same logic as for any AI project: start small, measure, then scale. Large rollouts without a pilot are the primary cause of failure.

Go Further

• Adobe Firefly for commercial images: when you need legally safe AI image generation for marketing materials
• Canva AI Magic Studio: the most accessible design tool for teams without a designer
• All our AI tool reviews: a catalog of tested and evaluated AI tools for businesses
• AI audit for your business: a structured diagnostic to identify where AI creates real value in your operations
• RAG systems: when you need AI that genuinely understands your internal documents at scale

Frequently Asked Questions