68.6% say users should learn to tell true from false. 31.9% say they do not want to check

The Information Integrity Survey 2026, published by Dentsu Soken with the Japan Fact-check Center, sampled 5,000 people aged 15 to 69 across Japan, supervised by Professor Shinichi Yamaguchi of GLOCOM, International University of Japan.

Two of its findings sit side by side. Asked what should be done about malicious AI-generated disinformation, 68.6% said users themselves should build the ability to discern correct information. Asked about actually verifying false or mistaken information, the single most common response was that they do not want to check whether something is true, at 31.9%.

The majority believes discernment is the user's job. The largest single group says they do not want to do it. Both live inside the same population.

Reading this as apathy is easy and wrong. Another finding from the same survey clarifies it: fewer than 10% of respondents understood concepts like filter bubbles and echo chambers. Most people are not skipping a procedure they know. Nobody taught them the procedure.

And even for those who know it, verification on the open web is genuinely expensive. Open the link beside the summary, skim the whole article, hunt for the part that might support the claim. Several minutes per sentence, against the volume of information that arrives in a day. Seen that way, "I do not want to check" is not a confession of laziness. It is a correct cost-benefit calculation. People stop doing things that do not pay, and that is rational.

Any remedy that asks for more willpower leaves the calculation untouched. The variable you can actually move is the price.

The survey also asked where responsibility sits. Beyond the creators of malicious AI disinformation, respondents pointed to users who spread it (35.1%), the platforms it spreads on (27.2%), and the companies providing the AI (25.5%). Asked who should build a digital environment where information can be used safely, the most common answer was that they do not know or none apply, at 35.7%, ahead of public institutions (33.5%), mass media (30.4%), and platform companies (28.1%).

In other words, the social consensus has not formed. While people disagree about who is accountable, the information keeps flowing. So rather than waiting for that argument to resolve, it is worth starting with what you control. Lowering the price of a check is faster than settling the question of blame.

For what it is worth, respondents did name where these skills should be taught: school education at 62.3% and the home at 50.7%. Expectations of education are high, which is exactly why what gets taught has to be a concrete procedure rather than a warning to be careful.

Verification is expensive mostly because of searching. The claim is right in front of you, but the sentence that supports it could be anywhere, so you end up skimming an article.

Change that structure and checking becomes a different task. When each claim links to the exact passage it rests on, verifying is not a skim but a landing: read, compare, decide, in well under a minute. And you do not need to check everything. Scope it to three things: claims that carry a decision, numbers you will repeat, and anything that surprised you. That usually means one or two checks per answer.

The 31.9% who said they do not want to check are probably not asking for permission to stay uninformed. They are saying they do not want verification to be a second job.

That is the thinking behind Tatsulok. Answers come only from your own documents and curated collections, and every claim links to the exact passage behind it. When the sources do not answer, it says so rather than filling the gap with something plausible. Making a check cheap is the most reliable way to make it happen.

Repeat a few cheap checks and you get a byproduct: you start to learn which kinds of claims tend to lose their support, and which sources hold up. The discernment that 68.6% of people said users should build does not come from a lecture. It comes from small verifications repeated often enough to become intuition. You should leave more capable than you arrived, which is the whole point.

This week's drill, no signup needed. From something you read today, an AI answer or an article, pick the single claim you would repeat to someone else. Open its source and find the sentence that supports it. Just the one. "Check everything" does not survive contact with a working week. "Check one" does.