You've probably had this chat. A support bot greets you with "Hi, I'm Sarah! How can I help you today?" You ask a real question and it stumbles, so you ask if you're talking to a person, and it dodges. The whole thing feels slightly off, like talking to someone wearing a mask. That discomfort has a name borrowed from robotics: the uncanny valley.
In chatbots, the uncanny valley isn't about looks. It's about a bot that acts human enough to raise the expectation, then fails to deliver, leaving the user unsettled and distrustful. The fix isn't making the bot more human. Often it's the opposite.
What "uncanny" feels like in a chat
The uncanny valley shows up as a small, nagging wrongness. The bot uses a human name and casual tone, so your brain relaxes and expects a human. Then it repeats itself, misses obvious context, or answers a question you didn't ask, and the illusion cracks.
That crack is worse than if the bot had never pretended. A visitor who knows they're talking to an assistant forgives clumsiness, because their expectations were set correctly. A visitor who thought they were talking to a person feels tricked, and tricked is a hard feeling to recover from in a customer relationship.
The goal isn't a bot that fools people. It's a bot that's genuinely helpful and honest about what it is.
Stop pretending to be human
The single biggest cause of uncanny chatbot moments is a bot pretending to be a person. It backfires, so don't do it.
A few specific things to avoid:
- Fake human names and personas. Giving the bot a first name and a "personality" invites people to treat it as human, then feel deceived when it isn't. A simple "assistant" framing sets honest expectations.
- Claiming feelings. "I'm so happy to help you today!" from a program rings hollow. It doesn't feel warm. It feels scripted.
- Dodging the "are you a bot?" question. When someone asks directly, answer directly. Evasion instantly turns curiosity into distrust.
- Fake typing delays meant to mimic a human. A little pacing can aid readability, but stretching it out to impersonate a person typing is a small deception people notice.
Being upfront that it's an AI assistant doesn't weaken the experience. It strengthens it, because now the user knows how to talk to it and how much to trust it. Honesty is the foundation everything else sits on.
Don't overcorrect into a robot
Here's the trap on the other side. Reacting against fake friendliness, some bots go stiff and mechanical, and that's its own kind of unpleasant.
A bot that answers every question with "PLEASE PROVIDE YOUR ORDER NUMBER TO PROCEED" or buries a simple answer in corporate boilerplate isn't honest, it's just cold. You can be clearly non-human and still be warm, clear, and easy to talk to. Natural language isn't the same as pretending to be a person. It's just good writing.
The sweet spot is a bot that talks like a helpful, plainspoken colleague who happens to be software. Friendly without faking emotion, clear without being robotic, and honest about being an assistant the whole time.
The habits that make a bot feel wrong
If you want to spot uncanny behavior in your own bot, watch for these patterns. Each one chips at trust.
| Habit | Why it feels off |
|---|---|
| Over-apologizing every message | Hollow and repetitive; nobody's that sorry |
| Excessive enthusiasm and exclamation points | Reads as a script, not a helper |
| Pretending to know when it doesn't | The confident wrong answer, the trust-killer |
| Ignoring what the user just said | Breaks the sense of an actual conversation |
| Long, hedged, throat-clearing answers | Feels evasive; people want the point |
Most of these come from a bot trying too hard, either to seem human or to seem thorough. Simpler and more direct almost always feels more trustworthy.
The habits that make a bot feel right
The good patterns are mostly about clarity and honesty, and they're easy to adopt.
- Say what it is early, in a light, non-awkward way.
- Answer the actual question first, then add detail if useful.
- Keep replies short. A wall of text feels like dodging.
- Admit limits plainly: "I don't have that, but I can get someone who does."
- Match the customer's tone without overdoing it.
- Offer a human handoff whenever the conversation gets stuck or sensitive.
None of these require sophisticated technology. They're writing and design choices you control. In SpideyChat you'd shape this through the bot's instructions and tone settings, telling it to be concise, honest about being an assistant, and quick to hand off, rather than chatty and evasive.
A quick before-and-after
Concrete beats abstract, so here's the same exchange handled two ways.
Uncanny version:
Customer: Are you a real person? Bot: Haha, that's a great question! I'm here and happy to help you with anything you need today! So, what can I do for you?
That reply dodges, over-emotes, and leaves the customer more suspicious than before.
Honest version:
Customer: Are you a real person? Bot: I'm an AI assistant for Northwind Coffee. I can answer questions about orders, shipping, and our beans, and I'll connect you with a human if you'd rather. What do you need?
Same friendliness, none of the discomfort. The second bot told the truth, set expectations, and got straight to the point without a single wasted word or hollow flourish. That's the whole formula, and it costs you nothing except the temptation to seem more human than you are.
Northwind Coffee makes a good running example. When they first launched a bot with a fake barista persona named "Bean," customers kept asking if Bean was real and getting cagey non-answers. Complaints went up, not down. They renamed it to a plain assistant, told it to be honest and brief, and the friction disappeared. People were fine talking to a bot. They just weren't fine being unsure.
The uncanny valley in chat is a trust problem dressed up as a personality problem. People don't dislike bots. They dislike being confused about whether they're talking to one. Set expectations honestly, write like a clear and friendly human wrote the script, skip the fake emotion, and hand off to a real person when it counts. Get those right and your bot lands on the near side of the valley, where it feels like a helpful tool instead of an awkward impersonation. Read a few of your own transcripts with this lens, and the awkward moments will jump out fast.