Do you actually know what you're paying for?

Most of us can say exactly what we get from our electricity bill or our phone plan. So when you pay for an AI tool every month, can you say what you're really getting? How many questions can you ask? Which version of the AI is answering you today, and was it the same one last week?

If you hesitated, you're not alone. Many Malaysians and businesses now pay for ChatGPT, Claude or Gemini, and very few can answer. The companies don't hide that limits exist. They just never tell you what the limits are.

"Usage limits" sound clear, but they aren't

Open any AI app and you'll see "usage limits". On Claude's free plan, for example, your allowance resets on a rolling five-hour window. But how big is the allowance? There's no number. You only find out when you hit the wall.

Behind the scenes, usage is counted in "tokens", small pieces of text the AI reads and writes. Longer questions and longer answers use more, so two people on the same plan can get very different amounts out of it. The company decides how it's all counted, and it could easily show a usage meter, the way your phone shows mobile data. As far as we can tell, OpenAI, Google and Anthropic all keep these numbers private. If it isn't a technical secret, who is it a secret from?


What if the AI you get today isn't the AI you got yesterday?

Imagine paying for a premium gym membership, and one day the good machines are quietly swapped for older ones. Same price, no announcement. Would you notice?

Something similar has been reported by AI users. People on several platforms have complained about being switched to a cheaper, weaker model mid-conversation without clear notice, and some posted what they found in public bug reports on GitHub. These are user reports, not proven company policy, and there may be reasonable explanations such as heavy demand. But ordinary users had to play detective to find out. And most of us would blame ourselves first: "maybe my prompt wasn't clear." That's what makes silent changes so hard to catch.

And what if you pay for a premium model but get a cheaper one?

This is the question that bothers us even more. Let's be very clear: this is a "what if", not an accusation. We have no evidence that any company is doing this. The point is that, today, you would have no way to find out either way.

On a normal chat app, you never see an itemised record of which model answered, how many tokens it used, or how that compares with last month. If the same kind of answer used double the allowance this month, who would notice? There's no public benchmark and no statement with line items.

Proper receipts exist only on the API, the developer version, which shows the model, token count and price for every request. The consumer apps stay opaque. It's not that someone is cheating. It's that the system makes it impossible for you to check, and a system you can't check asks for a lot of trust.


Surely there's a law for this? Not really, not yet

The EU's AI Act has a transparency rule, but it's mainly about telling people they're talking to an AI, not about publishing usage limits. New US state laws focus on similar disclosure and on protecting young users. Pricing falls under general consumer protection rules, written long before anyone talked about tokens. As far as we could find, no country has a law made specifically for this.

We're not lawyers, and we haven't checked how Malaysian consumer law treats it, so this isn't legal advice. But when no rule forces transparency, who should fix it: the companies, the regulators, or customers who ask harder questions?

For a small business, this is not a small issue

If you're one person polishing an email, vague limits are just an annoyance. But if your business runs daily work on AI, like customer replies, quotations, reports and automations, it's a different story.

What happens if the limit hits at 3pm on a busy Friday? If answer quality drops for a month and your team quietly spends more time fixing it? If the cost per task doubles and you only see it on the bill? In business language, that's an unmanaged dependency, and good digital transformation should reduce risk, not add one. Ironically, the businesses most excited about AI often ask the fewest questions, because nobody wants to be last to adopt. Yet "what exactly am I signing up for?" can save a lot of money and stress.

So what can you actually do about it?

Not everything, because the real fix needs more transparency from the companies. But you're not helpless. Start with these:

  1. Check which model is answering you. Most apps show the model name. Screenshot it when an answer feels off.
  2. Keep a simple log of the date, tool, model and task size for important work. It's the only "receipt" you'll have on a chat app.
  3. Consider the API for business use. It shows the model, token count and cost for every request.
  4. Ask your vendor or partner what's included, what happens at the limit, and how you'll be told about changes. Keep a backup tool too.

So, do you know what you're paying for?

Back to the first question. If you still can't answer it, you're in the majority, and that's the point. Opacity shows up whenever new technology moves faster than the rules, and the people caught out are usually the ones who trusted before asking the boring questions.


Five things Malaysian SMEs should check before paying for AI

You do not need to become an AI engineer to make a better purchasing decision. Start with these five checks.

1. Test the work your team actually does

Use representative tasks: a quotation, a customer email, a monthly report or a tender summary. A tool that writes an impressive poem may still struggle with your quotation format. Test what matters to your business.

2. Understand what happens at the limit

Does work stop? Can you switch models? Are additional charges available, and who can enable them? Find out before a busy afternoon makes the answer urgent.

3. Measure time saved after checking

Track how long the task takes from start to approved result, including corrections. A fast first draft is useful. A fast, accurate completed task is the outcome you are paying for.

4. Keep a small record of important work

For recurring tasks, record the tool, displayed model, completion time, corrections and interruptions. Repeat a few similar tasks periodically. This gives you a practical way to assess whether the service still meets your needs.

5. Plan for interruptions

Decide which tasks can wait, which can move to another tool and which need a manual process. Also check what business or customer information your team is allowed to upload. Convenience should come with clear working rules.

Before your next AI subscription, ask one better question

“Is this the best AI tool?” is a difficult question to answer.

Try this instead:

“Can this tool complete our work reliably, at a cost we understand?”

That question brings the discussion back to your business: your workload, your people, your deadlines and your customers.

AI can offer real value to Malaysian SMEs. Getting that value takes a little testing, a little measurement and a few questions before the subscription becomes part of everyday operations.

At Fiqmatic, we help businesses connect technology decisions with practical outcomes. If you are considering AI for your business, talk to us about assessing the workflow, expected costs and controls before you commit.

The monthly price is easy to see. Make sure the value is just as clear.