How Clue is approaching AI technology

Over the last two years at Clue, we’ve been doing thorough research and development into how AI functionality could be safely added into the app. 

It’s not been a race for us, because we want to release purposeful features that are secure, medically accurate, and genuinely useful for people. So we’ve taken the time that was needed, and recently introduced Chat With Your Data to a small group of users as our first functionality. 

It’s a conversational feature powered by an AI model that has strict rules and guardrails set up by Clue. A user can ask questions about their cycle patterns and receive a written reply that’s based on their own tracking data in Clue. It’s designed to give insights that are easy to understand and personalized to the user.

A lot of hard work went into its development, and I’m proud of the team involved. But now more than ever, it feels important to take the time to talk about why and how we’ve carefully built this feature. 

Because we acknowledge that AI is a topic that causes some anxiety. It can feel scary, as it’s an unknown and new technology that most people don’t quite understand. And it’s that lack of understanding that can make us feel so uncomfortable. 

So for that reason, I wanted to take a step back and help raise that understanding. Because I want you to have all the information you need, so you can make the right decision for you on how to use (or not use) AI within Clue.

Because the choice is genuinely yours. As you’ll read more below, data processing for AI functionalities in Clue is entirely optional, and you can hide these features from your app if you want to.

Yet there are real benefits in using features like Chat With Your Data, and it’s worthwhile trying it out to see its potential. I’ve been working at Clue for 6 years, and I’ve seen how impactful it is once people start connecting the dots to better understand their body and health. 

Why we started exploring AI within Clue

Back in 2024, I came up with the idea of creating a health assistant within the Clue app that provides a safe and trustworthy space for people to ask about their reproductive health. Instead of turning to the internet, visiting websites where the source of information is unknown and generalized, people could turn to Clue. 

Yet I had to admit, while the app houses a library of science-based articles and personal tracking data that can reveal such valuable health insights, this information can sometimes be hard to access. 

This is when we started exploring the idea of using AI technology to help us offer more to our users. Instead of trying to decipher a graph of data, we wondered if people could instead ask the app things directly like “When do I usually have cramps in my cycle?” or “Which is the heaviest day of my period?”.

So different people from across Clue—data scientists, members of the Science Team, technical engineers, designers—put their heads together and came up with the concept of Chat With Your Data, which we’ve launched today. It’s not quite the full vision of a health assistant, but it’s a promising first step.

The principles behind Chat With Your Data

To explain the deeper thinking behind this feature, I want to share with you the principles we followed during design and development. They helped shape our vision in making sure this technology wasn’t just tech for the sake of it, but rather a medically safe tool that brings meaningful value to the Clue community.

Being useful:an approachable tool for insights 

We’ve heard through user feedback that people have all this tracking data in Clue, but they struggle to make sense of it. Yes, we offer visual graphs in the Analysis Tab, but over time we’ve learned from our community that this approach doesn’t work for everyone.

Having this user need in mind is what drove our design concepts –– how could we build a feature that provides cycle insights in a format that’s accessible to everyone? After research, we concluded the best bet is a format that mirrors a simple conversation. No jargon, no charts and legends, just the type of everyday language you use when chatting with your friends or family.

Chat With Your Data isn’t doing anything radical; it’s simply a different way of presenting your cycle data. By making the format more approachable, the intention is that more people will find patterns in their tracking data and get a deeper understanding of their health. 

Choice: it’s up to you whether you turn the feature on 

Alongside making the feature useful, we wanted to make the feature optional. Let me explain how:

If you’re not interested in using Chat With Your Data, we’ve designed the feature so it can be disabled. When you first open the chat, you’ll be asked whether you want to continue and grant your consent to this type of data processing. If you disagree, the feature will be hidden from your app, and the data processing will remain turned off.

Why did we do this? Because our commitment to data privacy and giving people granular control over how their data is processed is as staunch as ever. We’re mindful of the different feelings toward AI technology, and we respect that some people don’t want to engage with it. Instead of forcing a feature upon all users, we’re giving people the autonomy to decide for themselves how they’d like to use it.

Safety: ensuring the technology is reliable and accurate

There are two aspects to this principle—safe technology and medical accuracy. Let me dive into each:

The technology behind Chat With Your Data is a large language model (LLM) that the Clue Engineering Team has carefully restricted with rules and guardrails. The AI can’t just roam and do what it wants; instead, it can only use the “tools” that we’ve provided it with. These tools are software coded by Clue’s own engineers, and they limit the AI chat to performing only three types of tasks:

1. Understanding your question in natural language.

2. Fetching the relevant information from your tracking data.

3. Writing an answer in reply.

By limiting the AI to these particular “tools”, this works as a strict boundary on what can happen. If we didn’t write a tool for it, the AI can’t do it. 

Our Science Team has also been heavily involved in this feature from the start. They’ve defined the rules on what the chat can do from a medical perspective, and ensured that safety filters are in place to catch high-risk content. Both of these help prevent inaccurate information from being provided by the chat. Like all things in Clue, answers must be grounded in science, trustworthy, and medically accurate. 

Members of the Science Team and external medical experts also perform clinical evaluations of the chat’s performance, assessing responses (which have been anonymized for privacy) for any potential risks or harmful behavior. This isn’t a one-off process, but a regular practice that ensures the chat stays safe and reliable and is continuously monitored over time.

Data privacy: keeping your data private and secure

The guardrails set in place to limit the LLM are also designed to ensure data privacy and security are upheld.

Firstly, if you’d like to use the Chat With Your Data feature, you’ll need to give your consent to a new type of privacy setting that allows data processing for AI-powered features within the Clue app. If you give your consent, then health data you’ve tracked in the app, and any conversation data you enter into the chat (for example, the questions you ask) will be processed by the AI behind Chat With Your Data.

Now you know what data is being processed; here’s how it remains safe and private:

• The data processed by the AI does not include any of your direct identifiers, such as your name or email address. Instead, it is linked to a random user ID.

• Your data is fully controlled by Clue. Although the chat is powered by an LLM from Anthropic, it’s hosted on Amazon Web Services (AWS Bedrock) within Clue's own secure AWS cloud-based environment. Your data is always processed within this controlled Clue environment, and it is never shared with Anthropic directly or with operators of any other foundation models. 

• Your data can never be used to train the LLM or by Anthropic to train any of their other AI models. 

• All data processing remains within the European Union and is protected by strict European data privacy laws (the GDPR).

• You can withdraw your consent for AI-powered features at any time in your Clue privacy settings via the More Menu.

Data privacy has always been a pillar in everything Clue does, and features that work with AI technology are no exception.

Going forward

As we explore other potential ways that AI technology can improve Chat With Your Data and more features in the Clue app, these guiding principles will always be behind our approach. 

We care about doing this right, and care about doing it with medical accuracy. Because we’re committed more than ever to providing a service that delivers health insight and agency alongside data privacy and safe technology. There doesn’t have to be a compromise.

If you’re looking for more information on Chat With Your Data, please visit our FAQs or refer to Section 3.1.7 of our Privacy Policy.