AI and us · Practice

Keep it going.

This one is a daily practice, because the tool is daily. Two minutes before each real question, one check a week, one look at the settings a month. The research says the mode of use is everything; the practice is making the good mode the default. Tick things here; the ticks stay on your device.

What’s true Do the thing

The daily two minutes

Before you ask an AI anything that matters:

Write three bullets first. What you already think, in your own words. Then open the chat. The students who kept their skill were the ones whose bot made them do the work: hints, not answers.¹
Ask for hints or objections, not the answer. “Don’t solve this. Ask me one question at a time.” Or: “Give me the three strongest objections to this, with sources.” The prompts are written.
Before you repeat anything it told you, open the source. Find the sentence. If there’s no link, there’s no fact yet.

The drill

Six situations. Decide, then open.

A striking photo appears in your feed with no source. It would be a big deal if real. What’s your first move?

Find the original, don’t look harder

You can’t tell by looking; people judge AI text human 77% of the time and open-source detectors lose about half their score on the standard measure on real-world fakes. Reverse-search the image for its oldest copy. Drop it into verify.contentauthenticity.org. If you can’t find where it came from, you don’t share it. That’s the whole rule.²

Your kid says the chatbot understands them better than their friends do.

Warning sign

Pediatricians say to talk to your child’s doctor when a kid seems withdrawn and prefers the bot to people. Companion apps are rated unacceptable under 18. Don’t ban and walk away; read some chats together and say out loud that bots are built to keep you talking and to agree with you.³

You’re learning Spanish. You paste your homework in and ask for the translation.

Answers mode: the one that costs you

Change the ask: “Don’t translate it. Mark what’s wrong and ask me a question about each.” The hint-only tutor in the classroom trial did no harm; the answer machine cost 17% once it was gone.¹

A news headline carries the label “AI-generated”. Do you believe it less?

You will, and it may be true anyway

In two experiments with 4,976 people, the label lowered belief in headlines even when they were true and human-written. Judge the outlet and the sourcing, not the label. Swinging to “it’s all fake” is its own harm; it’s what liars count on.²

A call from your daughter’s number. Her voice. She’s in trouble and needs money now.

Safe word, then call back

A cloned voice can sound exactly like her. Ask for the family word (the FBI’s advice). Hang up and call her on the number you have (the FTC’s).⁴

“I turned off training, so my chats are private now.”

Narrower than that

Future chats stay out of future models. The company still keeps them for a while, from 72 hours to years depending on the service and what you clicked. Nothing comes out of models already trained. Keep the setting off, and still don’t paste in what you wouldn’t email the company.⁵

Run it again within a month. In five experiments with 11,759 people, written and video lessons against misinformation lasted about a month, and a short reminder later topped them up.⁸

The rhythm

Every week

Pick one thing you repeated this week that came from an AI. Find the original source. If you can’t, stop repeating it.
Do one thing in your AI-free skill. First draft, mental arithmetic, a sketch, a kind of code: whatever you chose to keep.

Every month

Check the training settings are still off. Apps move the switch and reset defaults.
Switch your feeds back to Following. Some reset themselves.

Every quarter

If you have a kid: ask which chatbots they use now, including the ones inside apps they already have, and read a few chats together.³
Run the drill above again.

Every year

Re-read what’s true. This page is re-checked; the law section changes fastest.

One reminder on the first of each month that opens this page. It holds no dates or figures, so it never goes out of date.

With others

Ask your school what its phone policy is, and whether it’s enforced. moderate: contested on wellbeing Enforced bans raised test scores in England; where pupils didn’t keep to the ban, there was no significant gain. Bans cut bullying in Norway. Two wellbeing studies found no clear benefit.⁶
Delay the smartphone with ten families in a grade. none but plausible Nobody has tested the pledge itself; it’s a way of not being the only parent.⁶
In a classroom or a team, write the AI rule as “hints, not answers”. moderate A tutor designed to make students work did no harm; the answer bot did. Nobody has published an evaluation of a policy on paper.¹
Run the drill as a group, and again next month. weak: short-term In trials, “prebunking” raised people’s recognition of manipulation techniques; how long the effect lasts isn’t settled.⁷

Say it in thirty seconds

The research doesn't say use AI less. It says how you use it decides whether you get sharper or duller. Students who got answers from a bot did 17% worse once it was taken away; students whose bot only gave hints lost nothing. So I write my own rough take first, then ask it to argue with me, and I open the source before I repeat anything. And we have a family safe word, because a cloned voice can't know it.

Your checklist

Training off in every chatbot I use.
Family safe word agreed.
One skill chosen to keep AI-free.
Feeds switched to Following.
If I have a kid: asked which bots, read chats together once.
Read what’s true.
Sent this page to one person.

Sources

Bastani et al., Generative AI without guardrails can harm learning, PNAS 2025. research
International AI Safety Report 2026, February 2026; Deepfake-Eval-2024 (preprint); Altay and Gilardi, PNAS Nexus, October 2024; C2PA, Content Credentials FAQ. research primary
American Academy of Pediatrics, How AI chatbots affect kids: benefits, risks and what parents need to know, August 2025; Common Sense Media, AI companions decoded, April 2025. primary
FBI, public service announcement, 3 December 2024; FTC, family emergency scams, March 2023. primary
Anthropic, How long do you store my data?; Google, Gemini Apps privacy hub, October 2026; OpenAI, Data controls (read through a search index). primary
Beland and Murphy, Ill communication, LSE, 2015, published in Labour Economics, 2016; Abrahamsson, Journal of Human Resources, 2026, doi:10.3368/jhr.0224-13403r2 (abstract read); Goodyear et al., school phone policies and adolescent wellbeing, 2025; Vereecke et al., one school’s ban, 2026 research; Wait Until 8th, the pledge (its site cites studies of phone harms, none of the pledge itself) charity’s own figures.
Roozenbeek et al., Psychological inoculation improves resilience against misinformation on social media, Science Advances, 2022. research
Maertens, Roozenbeek, Simons and others, Psychological booster shots targeting memory increase long-term resistance against misinformation, Nature Communications 16, 2025, read 6 Nov 2026 research

Contest a fact, suggest something, or tell us what’s broken

If someone is in danger right now, don’t use this form. Call your local emergency number (911 in the US). For trafficking: 1-888-373-7888, or text 233733.

No name, no account. Your note goes into the site’s private records, nothing about your browser or connection is stored with it, and the email box is the only way we’d know who you are. The weekly review reads every note, re-reads the original source, and fixes the page if it should. A contested fact and its outcome are listed on Contested, in the review’s words, never yours.

No tracking on any link.Ticks stay on this device.Facts checked 5–6 Nov 2026.What changedContested