The honest case against chat with us

The banner appears bottom right as soon as the help page loads. A friendly cartoon head, a welcome message with an exclamation mark, three polite buttons offering to help with the three things you are not trying to do. You are trying to cancel a direct debit, or query a bill, or ask a human whether the thing you were sent is in fact the thing you ordered. You type the question. The bot replies with a link to the FAQ you have already read. You type it again. The bot asks you to confirm your account number, which the page already knows because you are logged in.

At minute four you type the word agent. The bot asks if you meant account. At minute six the bot admits that it cannot help with this, and offers to open a form. You open the form. The form asks for the question again. By the time any of this reaches a human, you have been on the help page for eleven minutes to achieve nothing, and the help page is counting your visit as resolved.

Why companies installed them

Everybody knows the answer, which is cost. A chatbot is a few thousand pounds a year to run and can field the first wave of common questions without a salary, holiday, or training. In the hands of a company that respects its customers, this is fine: the bot picks off the three things a lot of people ask about, and a human takes everything else promptly. In the hands of a company that would rather not hear from you, it is a maze dressed up as service.

The newer generation of chatbots, built on larger models, is more convincing. It uses full sentences. It apologises. It also, crucially, is still not allowed to do anything. The scripts restrict it to a small menu of canned responses, because a bot that can issue a refund can be talked into issuing a refund for free. The appearance has improved; the usefulness has not.

Four signs it will not help you

• The opening question is a drop-down of three bland options, none of which is something else.
• The bot asks for your account number though you are already logged in.
• There is no visible way to reach a human, or the words agent, representative, and complaint all route back to the FAQ.
• The bot ends every sentence by asking if it has resolved your query. Saying no restarts the loop.

How to escape

This is not a hostage situation and the companies are not hiding the exits on purpose. They are, however, not advertising them. A few stable tricks keep working.

Ring the number in the complaints policy. UK regulated companies in financial services, energy, telecoms and water publish a complaints procedure, and it includes a phone number. The number is sometimes buried on the complaints page specifically because it is not on the main contact page. Search for the company name and the word complaints procedure; the PDF or page you find will contain a route in.

Write it as a complaint. You can email the service, name the problem, use the word complaint, and say that under the company’s own procedure you expect an acknowledgement within a set number of working days. Companies in regulated sectors treat this differently from a chat message. The chat message may be ignored. The complaint is logged.

Know your ombudsman. If a complaint is not resolved after a company’s stated process (often eight weeks), you can go to the sector’s ombudsman. Citizens Advice has a guide. These bodies are free, independent, and quietly powerful; the possibility of a referral tends to focus minds.

Keep a receipt of the chat. Most chat widgets email you the transcript at the end, or let you download one. Download it. A transcript with a timestamp is useful if you later have to prove you tried to contact the company in good faith.

You did not fail at talking to the bot. The bot failed at being a service.

What a decent chatbot would do

A chatbot that is not a quiet refusal is easy to describe. It reads the account you are logged into. It can answer the common questions clearly, with numbers, not with links. It has a prominent, unmissable button that says talk to a person, and that button goes somewhere that answers within a stated time. It tells you plainly when it does not know, and routes you out. It does not call itself a team. It does not use the word journey. It respects your time.

These bots do exist. A couple of utilities and a couple of banks do the job properly. Notice them when you meet them. They are often the places that have been quietly losing fewer customers for some years.

The deeper irritation

The irritating thing about a chatbot is not that it is a bot. It is that it is a bot pretending to be a service, with a cartoon head and a first-name greeting, while an entire department of humans that used to pick up the phone has quietly been moved to a different floor. If a company wants to write a page that says our customer service is available by email, Tuesday to Thursday, within three working days, that is honest. That is a thing a customer can plan around. The bot that cheerfully asks how your day is going in order to delay you is not honest. It is a kind of lie.

Where this goes

The bots will keep getting more convincing. The pressure to install them will stay strong because they are cheap and the alternative is staff. The counter-pressure is only going to come from two places: the ombudsmen, who track the sectors where complaints keep escalating because the front line is a wall, and consumers, who move. Which? has a reasonable summary of consumer rights and runs a service that rates companies on how they handle complaints. Those lists are read by executives who do not want to be on them.

In the meantime: write it as a complaint. Keep the transcript. Say agent a lot. And when something as simple as a parking ticket appeal has its own cartoon head, remember that you are allowed to walk away from a service and come back with the number from the complaints procedure instead.