Insights

Shadow AI in hotels: when guest data ends up in a personal ChatGPT account

A made-up but plausible case: a three-paragraph complaint arrives by email, giving the guest’s name, room number, arrival date and their daughter’s nut allergy. The front desk agent wants to reply quickly and kindly. She pastes the email into ChatGPT, using her personal account on her phone, because the hotel doesn’t have one.

How often this happens is hard to say. I found no figure specifically for hospitality. But the numbers from the wider economy suggest that it does happen.

The conclusion I draw: a hotel that gives its staff no AI tools should assume that AI is being used anyway, just without rules and without a contract. Guest data is therefore better protected by an approved tool and clear rules. A ban on its own does little.

What the numbers say

In spring 2026, the German Economic Institute (IW) surveyed almost 5,000 employees. Of those who use AI at work, 29.1% use tools their employer does not provide. Among under-30s, the figure is 38.4%. In businesses with fewer than 50 employees it is 33.5%, and in large ones with 250 or more it is 25.5%.

The picture from the businesses’ side is consistent with this. According to the Institute for Employment Research (IAB), the research institute of the German Federal Employment Agency, 90% of businesses that use generative AI work with publicly available models. Of all businesses using generative AI, 21% have internal rules. In hospitality, 13 to 15% of businesses say they use generative AI at all. That figure says nothing about how many employees there use AI on their own initiative.

General-purpose tools are widespread in the hotel industry itself, too. In the AI Opportunity Study by the consultancy h2c, which surveyed 113 hotel chains worldwide, external tools such as ChatGPT account for 60% of day-to-day AI use, according to the chains. The study was sponsored by technology providers. And external tools do not automatically mean personal accounts.

Encouragement without tools

The IW survey shows a striking pattern. Where employers strongly promote AI skills, 44.5% of AI users turn to shadow AI. Where they don’t promote them at all, the figure is 21.3%. The survey does not prove cause and effect, as the authors themselves point out. One plausible explanation, though: people who are encouraged but not given a suitable tool go and find one. The authors sum it up like this (our translation): “Promoting without providing makes use uncontrollable.”

The obvious reaction would be a ban. The study explicitly warns against it (our translation): “Bans alone would mainly hit committed employees.” In a hotel, those are precisely the people looking for a faster way because the front desk is busy.

The real risk is the personal account. Anyone who pastes guest data into a personal account hands it to a service the hotel has no contract with and whose settings it doesn’t know. According to OpenAI, content in ChatGPT for individuals can be used to train its models unless you switch this off in the settings. Temporary chats are excluded. If you give feedback on an answer, you can still release that conversation for training even after switching it off. For its business offerings, OpenAI says it does not use the data for training by default.

The case against

The cost argument is the obvious objection: licences for everyone cost money, and a small hotel doesn’t need them. Then there is the worry that an official tool would only drive usage up.

I understand both arguments, but find them short-sighted. The numbers suggest that the use is already happening. What matters is whether it happens with a contract, rules and the right settings. Nor does everyone need a licence: it is enough to equip the roles that work with guests and with text.

What I would advise

1. Ask rather than assume. Ask your team openly, with no repercussions, what they already use AI for. The answers show where the need is, and often which processes are stuck.

2. Provide an approved tool. A business account with a data processing agreement, for the roles that need it. The agreement governs how the provider handles the data. Whether certain guest data may go into the tool at all is a separate question. Health information such as allergies is specially protected under the GDPR. In groups, head office decides which tool to use, in consultation with those responsible for data protection, and the properties receive it ready to use.

3. One page of rules. What may go in, and what may not? Guest names, room numbers, payment details and health information do not belong in a personal account, and in the approved tool only if data protection rules allow it. The page has to be short enough that someone at the front desk will actually read it.

4. Show how it’s done. Half an hour with real examples from your own hotel makes the rules tangible. It also fits the EU AI Act’s requirement to promote AI literacy among staff.

5. Turn recurring tasks into set processes. Tasks that many people do with AI again and again, such as replying to complaints or translating, should become set processes with approved sources. Once they are technically integrated, nobody needs to copy and paste any more.

What I would not do: ban AI and consider the matter closed. A ban with no alternative risks being worked around in day-to-day practice. And rules are still missing in many places: in a Bitkom survey from summer 2025, only 23% of German companies with 20 or more employees had any rules for generative AI at all.

What this means for the guest

The guest whose daughter has the nut allergy doesn’t see which tool was used to handle their email. They notice whether the reply came quickly and whether it was right. Both are possible with a tool the hotel has approved and whose rules it knows.

That is why we at 5213 LABS would start with a question, before any tool comes into it: where is AI already being used in the hotel today, and for what?

Sources

Next step

Which workflow costs the most hours every week?

20 minutes. One workflow. A clear next step.

Book a first conversation