A claim from a reel
“Mindfulness worked as well as an antidepressant for anxiety, with far fewer side effects.”
1Mindfulness here means an 8-week MBSR course: 2.5 hours a week, plus 45 minutes of practice a day. The drug was escitalopram, sold as Lexapro, at 10–20 mg.
2Noninferior, which is not the same as better: severity fell 1.35 points on the course, 1.43 on the drug. Side effects: 15.4% against 78.6%. JAMA Psychiatry, 2023
Fact-checking, in your DMs
You just did by hand what the bot does in a few minutes. Forward any reel and the claim, the verdict and the sources come back in the chat. Every check stays in an archive you can search later.
Free credits to start, and there is nothing to install
An example of a check, with the sources it actually rests on.
Forward a reel, a post or a link from Instagram or Telegram.
The bot reads the whole video, both what is said out loud and what is written on screen, then pulls out the factual claims.
The verdict comes back with its sources. Open them yourself: that is the point of the thing.
Closing the chat does not lose it. The check is written down, and you can ask about it later.
Where we look
A deep check searches in the language the material was made in, in English, and in the language of the country where the events happened. The same event gets reported differently in each.
what the people who made it read
the widest published record
where the event actually happened
When the zones contradict each other on the substance of a fact, we don’t quietly pick the convenient one. The verdict says Mixed, both versions go in the card, and the language of every source is named.
In your language
Searching and answering are separate things. The interface is translated into sixteen languages, down to the buttons, and the answers are not limited to those.
Say nothing and the check answers in the language of the material. Ask for Kazakh, Farsi or Brazilian Portuguese in your own words and it answers that way from then on, including languages where nobody has translated the buttons.
The central claims hold up against the sources found.
Part holds up and part does not, or credible sources genuinely disagree. The card carries both versions.
The central claims contradict the sources found, and the card says which ones.
Opinion, satire, personal experience. The bot says so instead of inventing a verdict.
What stays
Every check is written down the moment it finishes: the verdict, the reasoning, the sources, the caption, and whatever you wrote next to it. Saved reels pile up somewhere you never open again. These you can search, and ask about, months later.
The numbers check out. TAME trial, JAMA Psychiatry 2023: 276 adults with anxiety disorders randomised, 208 finished. The 8-week mindfulness course came out noninferior to escitalopram at 10–20 mg: severity fell 1.35 points on the course and 1.43 on the drug, a difference of 0.07 either way. Study-related side effects: 78.6% on the drug, 15.4% on the course. Noninferior means not worse. It does not mean better.
One entry, as you find it when you come back. It is yours, and one message deletes it, even when your balance is empty.
“what did I check about mindfulness?”
Ask for everything that turned out false, for one subject, or for the last few checks — in your own words, in the chat.
Entries link to each other by the people and concepts inside them, so related checks stay connected instead of sitting in a flat list.
Your account starts in whichever app you first wrote from. One tap connects the other side, and after that it is a single account: the same memory, the same balance, the same conversation, continued wherever you happen to be.
Where the reels find you. Forward one, get the verdict, keep it.
Where you can actually type. Question your memory, ask for a topic, top up.
If you never link them nothing breaks: everything still works in the app you started in.
The people who use it pay for it. There are no ads and no sponsors, we do not sell data, and nobody can order a verdict or order one changed.
Facts age. Every card carries the date it was checked, and a stale verdict is checked again rather than repeated.
Models run the checks, not editors, so a check can be wrong. Send one back and we review it again against current sources — how that works.
Credits pay for actual AI usage. Basic is usually about 0.3 credit. Deep is never more than 20 credits. The exact charge appears with the result. The prices themselves live on the pricing page, in your currency.
Buying happens in the chat: the payment page is personal, so its link arrives there.
Send it here before you repeat it. You start with six credits.
Send a reel on Instagram