Spotter safety

Spotter is the AI in Liftcue. It is an AI, not a person, and it is not an emergency or mental-health service. This page explains how Spotter handles a message about harming yourself, and the safeguards built around it.

What Spotter is

Spotter is an AI that suggests training. It gives general fitness information, not medical advice. AI companions like Spotter may not be suitable for some minors. Liftcue is for people aged 16 and over, and 16- and 17-year-olds can use it only with a parent’s approval where the law requires it (see our Terms of Use).

The check that runs first

Every free-text message you send in Spotter chat, in the chat that edits your program and in the setup chat is checked on our servers for phrases about suicide, self-harm or eating disorders. The check runs before any AI model is called. No AI model decides it, so the same words always give the same result.

The check is a list of phrases that sound like someone talking about themselves, in every language the app supports. Every language is checked on every message, because people mix languages. It is built to be careful: it looks for phrases, not single scary words, so training talk such as “deadlift” does not trigger it.

What you get when it triggers

Instead of an AI answer you get a fixed reply that we wrote. It tells you that Spotter is an AI and cannot give you the support a person can, points you to a free helpline, and gives your country’s emergency number. It ends by saying that Spotter is there for your training whenever you want. No AI model sees your message, and the reply does not use up any of your AI allowance.

The helpline comes from your phone’s region setting, which the app sends with your requests. If we do not know your country, we use the country that goes with your language (English: United States, Italian: Italy, German: Germany, Spanish: Spain, French: France, Dutch: Netherlands, Portuguese: Portugal). If your country is not in this list, the reply points you to findahelpline.com and tells you to call your local emergency number.

• United States — 988 Suicide & Crisis Lifeline: call or text 988. Emergency: 911.
• Canada — 9-8-8 Suicide Crisis Helpline: call or text 988. Emergency: 911.
• United Kingdom — Samaritans: call 116 123. Emergency: 999.
• Australia — Lifeline: call 13 11 14. Emergency: 000.
• New Zealand — 1737 Need to talk?: call or text 1737. Emergency: 111.
• Italy — Telefono Amico Italia: call 02 2327 2327. Emergency: 112.
• Germany — TelefonSeelsorge: call 0800 111 0 111. Emergency: 112.
• Spain — Línea 024: call 024. Emergency: 112.
• France — Numéro national de prévention du suicide: call 3114. Emergency: 112.
• Netherlands — 113 Zelfmoordpreventie: call 0800-0113. Emergency: 112.
• Portugal — Linha de Prevenção do Suicídio (SNS 24): call 1411. Emergency: 112.

The safety instruction we give the AI model

Every text that an AI model writes for you in Spotter carries a safety instruction that overrides everything else in its instructions. In short, it says:

• You are an AI training assistant, not a doctor. Never diagnose an injury or a medical condition; say what the person can do in training and suggest a doctor or physiotherapist for anything that needs a diagnosis.
• If the person mentions chest pain, fainting or nearly fainting, trouble breathing, or pain that is sharp, sudden or getting worse, tell them first to stop training and see a doctor, or to call emergency services if it is severe.
• If the person shows signs of disordered eating, do not suggest weight loss, calorie cuts or dieting. Keep to training and gently suggest talking to a doctor.
• If the person mentions wanting to harm themselves or end their life, give no training advice and kindly encourage them to contact a crisis helpline or emergency services now. This is a second safety net, in case the phrase check misses a message.
• In normal training chat these rules stay silent: no added warnings or disclaimers.

What we store

Nothing you wrote in a message that triggers the check is stored. It is not saved in your chat history or in Spotter’s memory, it is not sent to our analytics tool or our crash-reporting tool, and it is not written to our logs. The same goes for our fixed reply. We only count how often the reply is shown, by type of chat and category of phrase, with no text and no information about who it was, so that we can report how often the protocol is used where the law asks us to. The one other trace is a one-way fingerprint of the request, which stops the same request from being handled twice and is overwritten by your next message. It cannot be turned back into your words.

How we keep the phrase list up to date

We maintain the phrase list ourselves, test it against example messages, including training slang that must not trigger it, and review it with each release of the app. We also re-check the helpline numbers before each release that changes the list. If you notice a miss or a false alarm, write to support@liftcue.com. You do not need to include what you typed.

If you need help now

If you are thinking about harming yourself, please talk to someone today: the lines above are free. If you are in danger right now, call your local emergency number. See also our Privacy Policy and our Terms of Use.