Comparison
A custom chatbot or a ChatGPT subscription: not the same purchase
One raises the productivity of every employee; the other answers on behalf of the company. They do not replace each other.
The confusion starts because both look like a chat window with an AI behind it, even though they solve entirely different problems. A public subscription to a general-purpose assistant is a personal tool: every colleague drafts faster, summarises a document, translates, writes an email or cleans up a spreadsheet. It knows nothing about your procedures beyond what you paste into the conversation, it cannot be put on your website in front of customers, and it leaves behind no record the company can audit. A chatbot built for you works the other way round: it is given a bounded library — procedures, datasheets, framework contracts, frequently asked questions — searches that library on every question, and answers only from what it found, with the source in view and a complete log of every exchange. The deciding criterion is who is asking. If an employee is asking, for their own work, the subscription is the cheap and correct answer. If a customer is asking, or if a wrong answer produces a complaint, you need an assistant that cannot step outside your own documents.
Let’s talk about your project
Message us on WhatsApp or send an email — you talk directly to a developer.
office@northdan.com · +40 752 070 247
Key takeaways
What the public subscription gives you
A small cost per user, available today, with no project and no implementation. A powerful model kept current by the vendor, strong at drafting, translation, synthesis and text analysis. Zero maintenance on your side.
What an assistant built on your documents gives you
Answers restricted strictly to the sources you upload, with a citation of the document and section used. It can live on your website, your intranet or inside an internal application, escalate to a human, and leave a full conversation history behind.
The difference in liability
When a general assistant invents a fact, the damage usually stops at the employee who checked it badly. When the chatbot on your website quotes a wrong price or lead time to a customer, the conversation turns commercial — which is why control over sources is not a technical nicety.
How the cost looks in each case
The subscription is a predictable monthly expense per seat. Your own assistant carries an implementation investment plus a variable monthly consumption proportional to the number of conversations — indicatively, it starts paying off from a few hundred repetitive questions a month.
The four questions that settle it
Who is asking — employees or customers? Employees tolerate an approximate answer and verify it; a customer treats it as the company's official position. Where must the answer come from — the model's general knowledge, or strictly from your documents? If you have your own procedures, warranties, lead times or prices, only the second is acceptable. What happens to the conversation once it closes: if you need to review it, measure it or turn it into a support ticket, a personal subscription offers you none of that.
The fourth question is about the data you feed in. Before you let the team paste customer files into a public tool, read exactly what the plan you bought says about the use and retention of your content, then write a short internal rule and circulate it. Most of the incidents we see do not come from the technology; they come from the absence of a written rule about what may be uploaded and what may not.
When you do not need us in order to start
If we are talking about ten documents and five people, you need no project at all: buy the team subscription, load the material into a workspace inside the platform and see what comes out over two weeks. The cost is small, the risk is small, and the experience gained there makes the next conversation far clearer for everyone.
Call us when the symptoms appear that a subscription cannot fix: the same customer questions repeated on the phone all day, a manual of several hundred pages nobody searches any more, a support team copying answers out of three different sources, or the need to expose the assistant outside the company. Then we build it, measure the rate of correct answers, and if the numbers do not hold up we tell you before you invest further.
Frequently asked questions
Can I put our company ChatGPT subscription on the website?
Not in the sense you would expect. Team subscriptions are individual access licences to the vendor's own interface, not components you embed in your page. A public-facing assistant needs an integration through the programming interface, with your own rules, your own source library and your own consumption budget.
How does a custom chatbot know it is not allowed to invent things?
By construction: it searches your library first, receives only the relevant fragments, and is instructed to say it does not have the information when nothing suitable turns up. The rate of correct answers is not assumed, it is measured against a set of real questions before launch and periodically afterwards.
How long does it take to launch an assistant on company documents?
The technical part is the fast one, a matter of days. The real time goes into cleaning the material: superseded versions, contradictory procedures, scanned files with no searchable text. A disorderly library produces a disorderly assistant, so the first stage is always an inventory of sources.
What do I do with confidential documents in either option?
You keep them out of public tools and place them only in systems where you set the access rights yourself. A custom assistant can filter by role, so that a colleague in the warehouse never receives information from commercial contracts — something a personal subscription cannot do in any form.
Similar pages
Related resources
Let’s talk about your project
Message us on WhatsApp or send an email — you talk directly to a developer.
office@northdan.com · +40 752 070 247