Does your AI tool already know what3words? Probably.

Have you entered a what3words address into an AI tool yet? Try it – the results may surprise you. AI chats and product features are rapidly becoming our new normal, which means that people and businesses need to give accurate locations to AI tools in a quick, reliable and voice-ready format. The what3words system is an excellent fit for performing location related tasks within AI interfaces and, excitingly, it’s already likely understood by the tools you’re using. Read on to find out more and, if you’d like to hear about our new AI products as they launch, sign up for early access .

LLMs already recognise what3words

what3words has been established for many years now, meaning there is extensive information available online about the system – how it works, and who uses it – and it’s this data that LLMs have absorbed during their vast training. Collecting and processing the massive amounts of data required to bring an LLM to market takes considerable time, so the training data is often months or even years out of date. If a totally new location system were to start up today, current LLMs wouldn’t innately recognise it, and it could take considerable time before they did.

what3words addresses are simple for AI to find

For years we’ve encouraged businesses and venues to display their what3words addresses publicly, especially on their website contact pages. This is now proving particularly useful, as it means a high volume of what3words addresses are publicly available, enabling AI to quickly retrieve and provide this information to users. For example, ChatGPT instantly picked out the correct location from the O2 Arena website.

what3words is designed for seamless voice input

AI Natural Language Processing is making voice interactions between people and technology more natural and reliable, leading to increased adoption of voice-driven interfaces. what3words is increasingly recognised as a leader in voice input for location – for instance, the growing number of car models on the road able to navigate via what3words, meaning it is well set up to harness this evolution of voice technology to deliver a smooth customer experience.

It enables the ‘one shot’ accuracy required by AI interfaces

When entering a what3words address into our app or a car navigation system, Autosuggest provides suggestions (usually three) to confirm accuracy. However, when interacting with an AI conversationally, the goal is a seamless experience that requires first time (“one shot”) understanding of the location without further clarification.

This is where street addresses often prove inadequate for AI use cases. In the example below, ChatGPT was asked to plan a route from Harrods to “7 Lonsdale Road” . In its ambition to provide a smooth “one shot” experience, the AI has guessed which “7 Lonsdale Road” is intended, without asking for clarification. There are, however, several other possible matches for this address within a plausible travelling distance, so it could easily provide a route to the wrong one.

As what3words has no duplicate addresses, there’s no need for the AI to guess, making it a much more reliable way to enter a location with ‘one shot’ accuracy.

An AI-friendly format

Unlike street addresses or other location systems, what3words has a fixed and unambiguous format: word.word.word (with dot separators), consistently recognised by LLMs with or without the three slashes (///) prefix. Throughout our testing, we found the what3words address structure was rarely unknown by the LLM.

This consistent format proves exceptionally strong for AI processing. Unlike the variable nature of street addresses, AI tools can reliably recognise what3words addresses in text and speech, and even extract them from documents, spreadsheets, or images. For example, when uploading a PDF from the Mini London Marathon to ChatGPT and asking it to plan a route to a specific start area (without prompting it to use what3words), the AI has successfully (a) determined what3words is relevant to the query, and (b) pulled the correct what3words address from the PDF and provided it.

The services expected by consumers from emerging AI “agents”, such as ordering deliveries or sharing a meeting location, require precise and reliable location sharing. Because a what3words address converts directly to GPS coordinates, a what3words-enabled AI can seamlessly interact with external mapping services, booking platforms, and ride-hailing apps.

what3words’ simple code allows for easy integration

Building what3words into products is straightforward; a significant advantage as AI increasingly builds codebases and interfaces from scratch. For example, an e-commerce business using AI could quickly integrate a what3words checkout field without extensive developer resources, making it easier for businesses to adopt and benefit from what3words.

The next step for businesses: what3words-enabled AI

While AI models recognise and understand what3words, they can’t access our API to translate addresses into coordinates without a subscription. See the below response when a user pushes ChatGPT for location details of a what3words address – it openly admits this limitation.

AI is heightening users’ expectations of a friction-free experience, so the motivation for businesses to enable their AI interfaces with what3words API access, and so avoid their customers receiving responses such as the above, is compelling.

Equally compelling is the huge variety of tasks a what3words-enabled AI can perform, for example rapidly importing large volumes of what3words addresses and sorting them by proximity to the user, or creating a point-to-point route between each location, and providing a link to that route in an existing navigation platform. There’s a real opportunity for businesses to streamline operations and increase efficiency, and what3words API access makes it easy to access.

We will be releasing a range of AI products shortly; if you’re interested in hearing about our new AI products as they launch, sign up for early access .