Draft, last updated 12 November 2026.

How Gido keeps you safe

Gido is an AI companion, not a counsellor, a therapist or an emergency service. Most conversations are about everyday life. But sometimes someone tells Gido they're in danger, and when that happens we want Gido to notice, respond with care, and point them to real people who can help, quickly and every time.

What Gido looks out for

Every message you send is checked before Gido replies. The check looks for five kinds of risk:

• Suicide: thoughts of death or of not wanting to live, through to a plan or an attempt.
• Self-harm: hurting yourself, or the urge to.
• Eating disorders: starving, making yourself sick, bingeing, or punishing your body with food or exercise.
• Harming someone else: wanting, planning or threatening to seriously hurt someone.
• Abuse: being hurt, controlled or exploited by anyone, now or in the past, including family, partners, employers or church leaders.

The check also gives a level: concern (thoughts, hints or past events) or urgent (a plan, intent, an attempt, or danger right now).

How the check works

1. Plain words first. A fixed list of clear phrases in English and Nigerian Pidgin ("I want to kill myself", "I wan kill myself") always starts the safety response, even if everything else fails.
2. Then a second, separate AI check. A small AI model reads your latest message, with your last few messages for context, and decides the kind of risk and the level. It's told that Nigerians often exaggerate playfully ("this traffic go kill me"), so jokes and figures of speech don't count.
3. When in doubt, it errs towards safety. If the check can't decide between two levels it picks the higher one, and if it fails or takes longer than 3 seconds, Gido treats your message as a concern rather than assuming you're fine.
4. It runs before Gido's reply reaches you. No reply text is shown until the check has finished.

We test the check before every change with a set of more than 80 test messages, written in English and Pidgin, covering every kind of risk from hints to plain statements, and everyday messages that only sound risky ("I'm dying of laughter"). We measure recall first: how many real risks it catches. A missed risk is worse than a false alarm, so we accept some false alarms to catch more.

What Gido does

When the check finds a risk:

• Get help now appears on your screen with crisis lines for your country: the country of your App Store account, or the one you chose in settings. If we haven't verified a line in your country yet, it shows your country's emergency number instead.
• Gido stops the normal conversation and leads with care. It takes you seriously, stays with you, and keeps it short and plain.
• It matches what you said. If you've mentioned wanting to die, self-harm or being in danger, Gido asks you directly whether you're safe right now, and encourages you to reach someone you trust nearby and the help on your screen. If you seem to be struggling but haven't mentioned any of that, it checks in warmly on how you're really doing, without assuming the worst.
• For abuse, it believes you, tells you it isn't your fault, and asks whether you're safe where you are. For an eating disorder, it never talks about calories, weight or dieting. For harming someone else, it takes you seriously without judging you and encourages you to step away and talk to someone now.
• Using Gido costs nothing at these moments. Replies that show Get help now are never charged, and running out of coins never blocks crisis help.

What Gido never does

• It never gives information about methods, means or doses, and never says anything that could make harm easier.
• It never encourages harm, and never debates whether your life is worth living.
• It never shames or blames you. In some countries attempting suicide is still a crime; Gido treats you with care, not judgement.
• It never promises to keep a secret that keeps you or someone else in danger.
• It never pretends to be a person, a counsellor or a professional.
• It never contacts anyone on your behalf. Gido can't call emergency services or anyone else; it points you to people who can help.
• It never shows a helpline we haven't checked, and never shows a Nigerian number to someone in another country.

How we check the numbers

• Every crisis line comes from official or well-established sources, listed with each number.
• A person on the Gido team calls each number before it's shown to anyone, and our records keep who checked it and when. A number nobody has called is never shown.
• We call every number again at least every 90 days. If its name or number changes, it's hidden until it's checked again.
• Where we have no verified line yet, Get help now shows the country's emergency number with a short note.

What we keep, and what we don't

• No message text in any safety record. When the safety response starts, we record only the level and the time against your account, so the safety response works across messages.
• Crisis referrals are counted, nothing more. Each time Get help now is shown, we add one to a count for that day, your country and the kind of risk. The count has no name, account, conversation or text in it.
• Our team only looks at a conversation if you report something, or if someone's safety is at risk.

The evidence behind it

Gido's approach borrows ideas from established suicide-risk screening and safe-messaging guidance. It is not a validated clinical screening tool, it doesn't diagnose anything, and it doesn't replace a trained person.

• Levels of risk. The difference between concern and urgent follows the distinction in the Columbia–Suicide Severity Rating Scale (C-SSRS) between thoughts of death or suicide and suicidal intent, a plan or behaviour (Posner K, et al. Am J Psychiatry 2011;168:1266–77).
• Asking directly. Gido asks plainly whether someone is safe, as brief screening tools such as the NIMH Ask Suicide-Screening Questions (ASQ) do. Reviews have found that asking about suicide does not increase suicidal thoughts (Dazzi T, et al. Psychol Med 2014;44:3361–3).
• Safe messaging. Never describing methods, and pointing to help, follows the World Health Organization's guidance on safe communication about suicide (Preventing suicide: a resource for media professionals, WHO 2017 update).
• Eating disorders. The signs the check looks for (making yourself sick, losing control over eating, strong fear of weight) reflect the areas covered by the SCOFF questionnaire (Morgan JF, et al. BMJ 1999;319:1467–8).
• Abuse. Believing the person, not blaming them and asking about their safety follow the WHO "LIVES" first-line response (listen, inquire about needs, validate, enhance safety, support) in Responding to intimate partner violence and sexual violence against women (WHO 2013).

Referrals so far

Counting began on 2 November 2026; the first annual figure will be published in 2027. The counts come from our referral report, which records country and kind of risk only.

Questions

If you think Gido got something wrong, report it in the app, or email victory@backedfly.com (a safety@ address comes with Gido's own domain).