As technologies go, chatbots are not always popular among users. According to research from the online community, 54% of users in the U.S. found chatbots to be either “not effective” or only “somewhat effective.” Respondents expressed the most frustration with having to repeat information when being passed from a chatbot to a human agent but also reported chatbots “getting stuck and not knowing what to do next.”

These attitudes about chatbots play out in how digital strategists and implementation firms look at chatbots in the user experience mix. “What’s really interesting is how quickly customers are learning the fastest ways to solve problems with the infrastructure provided, particularly among younger cohorts,” says Jake DiMare, a Los Angeles-based digital strategy consultant. “Chatbots excel at understanding intent and filtering through piles of information to find the right answer to a question. This is very unlike search, which provides all the relevant results and expects the user to find the one that works.”

These factors are combining to drive much more chatbot development. Google entered the space when it acquired the popular API.AI chatbot platform in 2016. At the time of the acquisition, API.AI had 60,000 people developing chatbots on the platform. Since then, Google has renamed the platform Dialogflow and integrated it with Google’s cloud infrastructure. In a sign of how important multilingual and localization are to user experience, Google has increased its language support in the latest version. The original API.AI included 15 languages and a number of additional locales. The new version of Dialogflow does not expand on that number of languages significantly. Instead, Google focused its efforts on allowing developers to include more than one language in a single chatbot instance; now, developers can include up to 15 languages or locales.

With this increased language support, the potential applications have followed. KLM Royal Dutch Airlines is using Dialogflow behind chatbots that support customers during ticketing and check-in as well as a means to check status updates. The user flow can be initiated by the user during the ticketing process when the users are asked if they would like to receive trip updates via Facebook Messenger or other social media and messaging tools. Once they opt in, the users can then interact with KLM via Messenger throughout the trip.

Despite what some studies might suggest, chatbots are becoming more and more popular—at least among brands. Martha Senturia, senior user interface designer at PTP, says, “I think the global preference for chatting and texting is so strong, companies would lose a primary mode of communicating with their customers if they ignored it. But it should be done well to represent the brand professionally and appropriately or not done at all.”

But creating chatbots for multilingual audiences presents its own separate set of challenges. Senturia says the challenges include “understanding not only the language in each locale, but the written chat shorthands, abbreviations, use of emojis, and cultural considerations that are different depending on the user’s region. Knowledge of the spoken language is not the same as the written ‘chat’ language, and chatting conventions are not the same from region to region. Using the right degree of formality, appropriate conventions, and the correct writing system (that can differ even for areas where the spoken language is the same or similar) requires linguistic expertise and regional knowledge.”

One ambitious multilingual chatbot project is Visabot, a robot that helps people with U.S. immigration. One of the site’s applications enables U.S. citizens to help their spouse get a green card. It guides users through the process by collecting key information from open sources such as LinkedIn, helping the user gather supporting documents (such as birth and marriage certificates), and helping to make sure forms are filled out accurately and completely. As of this writing, Visabot supports English, Mandarin, and Hindi, and, to date, the site has supported more than 100,000 users. Its success has landed its CEO, 27-year-old Artem Goldman, a spot on Forbes’ most recent 30-under-30 list.

The more complex the chatbot, the more chatbot designers need to consider the user experience. Even something as (seemingly) simple as asking to change a seat can lead to frustration; something as complex as building a visa application has the potential to go further awry. If you want to avoid the user frustration described at the beginning of this article, you need to keep the user experience front and center.

“Chatbot users who need to type will want the chatbot to be easy to read and chat with—no matter the language,” says Senturia. “In regions where there are different levels of formality in the written system and/or register in the spoken version, the chatbot needs to aim for a balance of readability and respect for most users (i.e., Simplified vs. Traditional Chinese or informal vs. formal ‘you’ in Spanish). Users may not want to bother with punctuation, capitalization, diacritics, and so on, to save their hands from fatiguing; so they should be able to answer with the fewest symbols or letters as possible.”

She offers these other considerations: “In areas where the Roman alphabet is not the standard writing system or for multilingual users, minimizing the need to switch keyboards can make it easier on the user. For any language, the experience is visual, so it needs to be brief and easy to read.”

KLM, for its part, has been expanding on its chatbot efforts to include more complex ones, such as an application on Google Assistant that helps users pack for a trip. The application is intended to be both useful and somewhat playful; KLM even refers to its chatbot persona (“BB” for “blue bot”) as being “cheeky from time to time.” This element of humor can go a long way in helping users navigate chatbot-based applications—and by doing so, improve the acceptance of the technology as more organizations look to expand its usefulness and footprint.