Bots are from Mars, humans are from Venus.

Remember the days of buying cinema tickets over the phone? Wasted minutes bellowing ‘Jumanji… JU-MAN-JI’ into your landline as the disembodied voice on the other end says ‘Sorry, I didn’t get that. Did you say Batman Forever?’

Thankfully, voice assistants have come a long way since then. By the end of 2017, Google's speech recognition had hit 95% accuracy. Chatbots have come a long way, too – but inevitably, error messages and bottlenecks still happen. And research shows that unhelpful messages can make customers' stress levels go through the roof.

So how do you stop your users getting fed up, and calling you instead? It’s all in your language. Find the right tone for your bot – and make it a master conversationalist – and you’ll save your customers from banging their heads against their screens.

1. Sound human. (Just not too human.)

The Turing Test might once have been the standard we held robots to, but we don’t want our bots to pretend to be human any more. A Goldsmiths study found that almost half of Brits think it's 'creepy' when a bot pretends to be real. We don’t want to feel duped – we want to know where we stand. In California, it'll be illegal for a bot to pretend to be human as of July this year.

That doesn’t mean we want to have to interpret robot-speak, though. In a conversation with a bot, we want the same things we want in any conversation: empathy. Warmth. Signs that the other person is listening. One of the most natural ways we show empathy is by mirroring the language of the person we’re speaking to.

So writing more like you speak will serve you well here. Use conversational words, contractions, and the odd discourse marker like ‘ok’ ‘well’ and ‘right’ to help build a warmer relationship with your users. (Of course, there are as many ways to sound ‘human’ as there are, er, actual humans. It all depends on your brand’s personality, and the persona you’re after for your bot.)

2. Set the parameters of your relationship.

One of the biggest reasons your users are getting frustrated with your bot might well be because their expectations don’t match up with what you can deliver. So from your first interaction, be totally clear on what your bot can – and can’t – do.

And pick the right points in the conversation to remind customers what those boundaries are.

Your intro message is the place to start. Nike does a nice job of this in theirs:

‘I can help with an existing order or get you set up with the hottest shoes & gear.’

As does Western Union:

‘Chat with us to send money, track transfers, check exchange rates/ fees, find agent locations, and more.’

Your bot not understanding a question is another good time to direct users to the right tasks – ‘Hmm, I can’t help with that. Here are a few things I can do…’

Set the parameters for the tone of your conversation, too. If you start on a subservient, apologetic note – ‘How may I be of assistance?’ – you’re priming your user to have less respect for your bot. (Plus, if your bot’s female, using language like that could reinforce harmful gender stereotypes.) Go for an adult-to-adult feel instead: ‘I’m here to help. What’s happening?’

3. Balance personality with getting on with the job at hand.

Your users might come to your bot because they have a job that needs doing. But they’re more likely to stick around, come back, and tell other people about it if they have a good time in the process.

In a Forrester study into the traits we want to see in a bot, ‘funny’ came fourth – right after polite, caring and intelligent. We want efficiency and helpfulness, sure – but sometimes, we want jokes and weird gifs, too.

Take the time to craft your bot’s character: what’s their back story? What are their hobbies? What other robots would they be mates with in the playground? Turn that character into a set of guidelines that anyone writing for your bot can follow – like the ones we did at The Writer for Vodafone's customer service chatbot TOBi.

Don’t let personality get in the way, though. No one wants your assistant to crack a joke at the expense of actually solving their problem. Save those snippets for the Easter eggs – hidden gems to reward you user with if they ask certain questions.

Ultimately, the smarter machines get at things like speech recognition and natural language processing, the less fed up we’ll get as consumers. But all of that technological brilliance is wasted if your bot’s just rubbish at having a conversation. And without things like body language or visual cues to help guide that conversation, it’s your words that need to do all the heavy lifting.

Pay more attention to those words, and your customers will feel more inclined to keep coming back – and less inclined to chuck their Alexa out the window.