Why ChatGPT Is Not Enough for Your Business

The short answer: ChatGPT has never seen your business. It writes well, it knows a lot about the world in general, but it has not read your price list, your stock levels, your calendar, or the page where you list your Saturday hours. That is why ChatGPT for business, in the form most people use it - a chat window in a browser tab - is a great assistant for you and your team, but cannot answer your customers on your behalf. For that you need a bot that answers from your own content and lives on your own site.

This is not an argument against ChatGPT. A lot of daily work is faster because of it. But there is a real difference between a model that knows a bit about everything and a bot that knows everything about one company, and it is worth understanding before you spend time on the wrong one.

What ChatGPT is genuinely good at

If you need text, a summary or a first draft, a general model is still the fastest tool within reach:

• Writing: product descriptions, social posts, customer emails, page copy.
• Summarising: a long contract, meeting notes, a forty-message email thread.
• Translation and proofreading.
• Thinking out loud: five headline options, or a second opinion on a proposal.
• Explaining general topics that have nothing to do with your internal data.

What these tasks have in common is that they either need no internal data at all, or you paste the data in yourself. You are there to check the result. A visitor on your website is not, and that is where the problem starts.

Where a general model runs out of road

It does not know your content

The model learned from public text up to a certain date. It may have seen your site, a two-year-old version of it, or nothing at all. It has definitely not seen your PDF price list, your internal delivery terms, or the questions customers ask you over the phone. Asked something specific about your company, it has nothing to read from, so it produces the most plausible sentence. Sometimes that is right, sometimes it is not, and in both cases it sounds equally confident.

It does not have prices, stock or opening hours

Prices change. Items sell out. Holiday hours differ. A general model is not connected to any of those sources. A bot connected to your shop is: through the WooCommerce or Shopify integration it reads products, prices and stock levels, and a local product index helps it find an item even when the customer types it without accents, with a stray hyphen, or by SKU.

It cannot book an appointment

This is the biggest practical gap. ChatGPT can draft a message about how to offer a slot. It cannot see which of your staff is free on Tuesday at 11, hold that slot, and put a name in it. A bot connected to your calendar and your team takes it all the way: service, person, date, time, confirmation. The same system can cancel and reschedule when the customer asks.

It is not on your website

A visitor looking at your product will not open another site, copy the question over and come back. Support has to be where the visitor already is. That is why a website chatbot sits in the corner of the page and opens in one click, and why the same bot can also answer on WhatsApp, Messenger, Instagram, Viber and Telegram with the same knowledge and the same rules. Social channels start on the Standard plan, and Viber additionally needs a business account paid directly to Viber (around 100 EUR per month).

It does not keep conversations you can review

Conversations your customers have with ChatGPT stay in their accounts. You see none of it. Those conversations are the most useful by-product of the whole thing: what people actually ask, where they hesitate, and what they look for that you do not stock. In our panel the owner sees every conversation, a list of unresolved questions (the ones the bot could not handle), and how answer confidence moves over time. That list is the most concrete content plan you can get, because it comes from the customers themselves.

What "answers from your own content" actually means

1. You provide the sources: your website address, PDFs, spreadsheets, a list of questions and answers, documents from Google Docs.
2. The system cuts that material into smaller pieces and stores them so the right piece can be found by meaning, not just by exact wording. Each company's content sits in its own separate store, apart from everyone else's.
3. When a question arrives, the bot searches your material first, then writes the answer out of what it found.
4. If it found nothing, the rules explicitly forbid it from inventing an answer. It says it does not have that information and offers a contact or a callback instead of guessing.

One detail is rarely mentioned but worth knowing: your own instructions to the bot ("always mention free delivery above our threshold", "never negotiate on price") sit at the end of the rule list. The reason is practical - when a model gets two instructions that contradict each other, it follows the one closer to the end. That way your word is the one that counts.

There is also a nightly document quality check. If a file was read badly - a scanned PDF with no text layer, a font that drops local characters, an empty document - you get a warning. The bot finds very little in a file like that, and without the check you would only notice through a bad answer to a customer.

Why this changes the answers

Take a question that reaches you every day:

Do you have this model in size 42, and can it arrive by Friday?

A general model cannot answer that even approximately. It does not know which model, how many are left, or what your delivery times are. A bot connected to your shop knows the item, sees the stock, reads what your own page says about delivery, and answers specifically. The difference is not intelligence, it is access to facts.

The same applies to what the bot does not know. "I do not have that information, but I can put you through to a colleague or arrange a callback" causes no damage. An invented delivery date does.

AI for customer support, by industry

Same principle, different work. AI for customer support in an online shop spends most of its time on product search, stock checks and order status. For real estate agencies the visitor searches by size and location, then books a viewing for a specific unit. For clinics and hotels the weight is on appointments, services and prices, with careful wording: the bot gives information from your content and points to your people when the question calls for professional judgement.

Voice on the site, and its limits

The voice agent works as a microphone button inside the chat window on your site (your site needs to be on https). Audio goes straight from the visitor's browser to the model, our server never receives it. There are ten voices, adjustable speaking speed, and a session lasts up to one hour.

By voice a visitor can search your knowledge base, ask about services and prices, request details from a page of your site, check free slots and book an appointment. Some things still go through text chat: placing orders, order status and cancellation, payment links, stock checks, and cancelling or rescheduling an appointment. To be clear: the voice agent does not take or make phone calls. There is no phone number and no telephony. Phone calls are coming soon, but they are not live yet. On WhatsApp, Messenger, Instagram and Viber the bot is text only.

What it costs

Prices by plan, in EUR per month:

Voice minutes are included on Pro (200 minutes) and Business (1,000 minutes). SMS allowances are 50 on Standard, 200 on Pro and 500 on Business. Enterprise is by agreement, annual billing saves 12 percent, and the full breakdown is on the pricing page.

A few things depend on the plan: appointment booking is on Pro and Business (or as a 79 EUR per month add-on), removing our signature from the chat is also Pro and Business (or 29 EUR per month), and social channels plus order taking start at Standard. An extra bot is 29 EUR per month.

A message is used by each visitor question in the site chat, each message from WhatsApp, Messenger or Instagram, each visitor turn in a voice conversation, and each form submitted inside the chat. When the quota runs out the bot stops answering new questions. There is no automatic overage billing, so no surprise invoice. Top-ups exist: messages at 0.05 EUR each (100 messages for 5 EUR, 500 for 25), a 200-minute voice pack for 50 EUR as a one-off, and extra voice minutes at 0.25 EUR each. A top-up creates an invoice with a 7 day term, paid by bank transfer, and the quota is added once the payment is confirmed, so plan it before you run dry rather than after.

When ChatGPT is all you need

If you want help writing, ideas, summaries and translation, a general model is enough and there is no reason to pay for anything else. If you get ten visitors a day and answer every question personally, skip the bot: a human reply is better. A bot earns its place when the same questions repeat, when people write in the evening and at weekends, when enquiries go cold because nobody replied within half an hour, and when you want to see what customers are actually asking.

How to check without risk

Installation is one line of code on your site, and channels are connected by signing in with the company Facebook account. Add your website address and a few documents, let the bot learn the content, then ask it the twenty questions your customers ask most. If the answers are correct, you have your answer to the question in the title. The trial runs for 7 days and needs no card.