Picture this: it's gone 11pm, someone lands on your website with a question your chatbot hasn't been told the answer to. Most bots will guess. They'll cobble together something plausible-sounding, hand it over with total confidence, and move on. The visitor has no idea they've just been fed a made-up answer — until it turns out to be wrong.
That's the problem with a lot of AI chatbots. They're built to always have a reply, even when the honest answer is 'I don't know'. An AI chatbot that admits it doesn't know works differently — and that difference matters more than most businesses realise until they've lived through the alternative.
Chatbots are trained to be helpful, and helpful usually gets confused with confident. A bot that says 'I'm not sure' can feel like a failure to the people who built it, so a lot of systems are quietly nudged towards giving an answer rather than the right answer, or no answer at all.
The trouble is, a confidently wrong answer about your opening hours, your returns policy, or your pricing does more damage than no answer at all. People remember being misled far more than they remember being told 'let me check and get back to you'.
This is where it really bites. Someone's browsing late at night, they ask a slightly unusual question — maybe about a product bundle you don't actually offer, or a delivery option that's only available in certain postcodes. There's no one at your business awake to catch the mistake. If the chatbot invents an answer, that's the version of your business the customer walks away with.
An honest chatbot isn't a chatbot that knows everything. No chatbot does, and any business that claims otherwise is stretching the truth. Honesty here means something more specific:
That last point is the part people often miss. The value of a chatbot admitting it doesn't know isn't just in that one conversation — it's in what happens afterwards.
If a chatbot quietly says 'I don't know' and nothing else happens, you've avoided one bad guess, but you're still in the dark. A chatbot worth having will flag unanswered questions back to you, so you can see the pattern: maybe twelve people this month asked about a delivery option you haven't documented, or a warranty question you've never actually written an answer to. That's useful information about your own customers that you wouldn't otherwise have.
There's a common worry that admitting uncertainty makes a business look less capable. In practice, it's the opposite. People are used to dealing with AI now — they can usually tell when something's been generated rather than genuinely known. A chatbot that's upfront about its limits reads as trustworthy precisely because it isn't pretending.
Think about how you'd feel calling a shop and being told 'I'm not sure, let me find out and call you back' versus being given a confident answer that turns out to be wrong. The first is mildly inconvenient. The second costs you time, and possibly money, and definitely trust.
Being open that you're talking to an AI, and that the AI has boundaries, does something useful: it stops people expecting perfection. Once that expectation is set honestly, small gaps in knowledge don't feel like broken promises. They feel like exactly what was signed up for.
If you're weighing up chatbot options for your own business, it's worth asking a few direct questions rather than taking 'AI-powered' claims at face value:
If the answer to any of these is unclear, it's worth pushing for a proper explanation before you put it in front of customers. A chatbot that can't tell you what it's unsure about is a chatbot you're trusting blind.
Some tools bolt on 'transparency' as a checkbox. The more useful approach is building the whole thing around it: openly AI, upfront when it's guessing versus when it knows, and set up to report the gaps back to you rather than papering over them. That's not a lesser version of a chatbot — it's a more useful one, because it makes the unknowns visible instead of hiding them behind confident-sounding sentences.
That's the thinking behind Lyla. It's openly an AI chatbot, it says when it doesn't know something rather than guessing, and it reports the questions it couldn't answer back to you, so nothing useful gets lost at 11pm when no one's watching. If you'd like to see what that looks like on your own site, have a look at Lyla.