TL;DR: WasThatTrue is user-triggered. It does not scan every claim in every YouTube video. You click Fact-check when something sounds off, choose the exact claim you want verified, and read a source-backed verdict. That design protects three things at once: viewer agency, privacy, and trust. Automatic fact-checking sounds comprehensive, but in practice it creates new failure modes that the user-triggered model avoids.
Most fact-check tools want to do more, not less. They scan every line of captions, score every speaker, and flag what they think might be wrong. User-triggered fact-checking is the opposite. You click when you care. Nothing else gets checked.
That choice is the core design decision behind WasThatTrue. It is also the part people ask about most: why not just automate it?
The short answer is that fully automated systems try to solve a problem viewers do not actually have. People do not want a tool that polices every video. They want a tool that helps when a specific claim catches their attention.
The longer answer is that auto-scanning brings real costs. It creates new false-positive problems, it reads as moderation, and it processes far more of your viewing than it needs to.
This post walks through the trade-off, why user-triggered design holds up better than it first sounds, and what happens when you actually click the button.
What Is User-Triggered Fact-Checking?
User-triggered fact-checking means the viewer chooses both when a check runs and which claim gets checked. The tool does not scan the whole video, score the speaker, or flag content on its own. It waits for a deliberate click, then runs a focused check on a single claim the viewer selected.
That is a small design choice on the surface and a big one underneath.
Most fact-check workflows make the tool the active party. It scans, decides what looks suspicious, and surfaces a warning. User-triggered design flips the roles. The viewer is the active party, and the tool is the answer engine.
If you are new to the broader workflow, the claim-level fact-checking guide walks through what a YouTube fact check should look like in plain terms. This post is about a narrower question: who pushes the button.
Why Auto-Scanning Every Claim Does Not Work
Auto-scanning sounds like the obvious feature. Check everything, surface everything, let the viewer see it all.
In practice, it runs into four problems.
First, coverage is harder than it looks. YouTube's own help page for fact-check information panels acknowledges that these panels do not appear for every search. A 45 minute video can hold hundreds of factual statements. No system, automated or human, checks all of them with the same care.
Second, auto-extraction is noisy. A 2025 paper on AI agents for YouTube describes a system that extracts claims from YouTube videos and checks them with retrieval-based evidence. The interesting result is not just the accuracy. It is how much filtering the agent has to do before it lands on a checkable claim. Most spoken sentences are not factual assertions. They are jokes, asides, questions, or opinions. An auto-system that does not filter aggressively will spam the viewer with verdicts on the wrong sentences.
Third, policy lag is real. YouTube's medical misinformation policy page notes that updates can lag behind changes in health guidance. Auto-systems built on rule sets inherit that lag. A fast-moving topic can look very different to the platform policy than it does to current evidence.
Fourth, the auto frame reads as moderation. YouTube's Community Guidelines remove certain misleading content with serious risk of harm. That is a platform decision about what people can post. A fact-check extension running similar logic on the viewer's screen blurs the line between informing and policing, and the viewer never asked for it.
The Trust Problem When a Tool Picks Targets
When a tool decides which claims are worth checking, it inherits the suspicion that comes with any moderation system. Viewers ask why this claim and not that one. User-triggered design sidesteps the question entirely. The viewer picks the target, so the tool only has to answer one job: was the chosen claim accurate?
Trust in platforms has been wobbly for years. The Reuters Institute Digital News Report 2025 found that audiences across markets continue to question how algorithms shape what they see. A tool that quietly chooses what to flag is the same problem in a smaller wrapper.
Stanford researchers have spent years studying how people verify online claims. Their work on lateral reading shows that verification works best when the reader is in the driver seat, actively checking reputation, evidence, and claims across other sources. The Civic Online Reasoning project is built around the same idea, treating verification as a skill you exercise, not a service you subscribe to.
User-triggered design fits that model cleanly. The click is the moment of intent. The viewer chose to check. The tool just has to answer well.
How Crowdsourced Notes Get Closer, And Where They Stop
Crowdsourced notes are the closest cousin to user-triggered fact-checking. On X, Community Notes lets users write and rate corrections that attach to a post. When a note clears the rating threshold, it shows publicly.
That model has two strengths. It is fast inside its native format, and it keeps corrections close to the original content. A reader does not need to leave the post to see what other readers flagged.
It also has two limits when you carry it over to long-form video.
First, latency is wrong for video. A Community Note can take hours or days to surface, because it needs ratings from contributors who agree across viewpoints. That works for a post that lives in a feed. It does not work for a claim you heard 90 seconds ago in a video you are still watching.
Second, there is no equivalent inside YouTube. Long-form videos do not have a public note layer. The comment section is not it: comments sort by engagement, not by accuracy, and they do not appear in the moment the claim is made.
User-triggered fact-checking solves the latency problem in the way crowds cannot. You ask the question now. You get the answer now. The crowd model still earns its place for slower, public-record corrections; it just is not the right fit for a viewer hearing a claim mid-video.
What Happens When You Click Fact-Check
You press the WasThatTrue button, the extension reads the captions around the moment you clicked, and it offers a short list of factual claims it found there. You pick the one that caught your ear. A verdict card comes back with the exact quote, a plain restatement, a label, and source links you can open.
That is two decisions for the viewer, not one.
The first is when to check. Nothing runs in the background. The extension stays out of the way until you ask.
The second is which claim to check. Most fact-check tools assume they can guess what you cared about. WasThatTrue does not. It surfaces what it found, and you choose.
A 2026 paper looking at about 2,500 statements submitted by 457 participants offers a useful reality check. When people are free to fact-check anything, they pick selectively. They do not want every line of every video checked. They want help on the lines that matter to them.
If you want the deeper version of what happens after the click, how the sourcing pipeline works explains the two-model design and why the source links on a verdict card are real. The product page for the YouTube fact-checking tool covers the rest of the experience.
The Privacy Bonus of User-Triggered Design
User-triggered design is a privacy decision before it is a moderation one.
An auto-scanning extension has to process everything. Every caption line, every video, every viewing session. Even if the data never leaves your machine, the surface area is large. Even if nothing is logged, the assumptions you have to make about the tool get heavier.
WasThatTrue only processes the claim you choose. The extension reads a window of captions near the click to find candidate claims, and it sends the one you select to the server for verification. Nothing else. No watch history, no continuous scan, no profile of what you have been watching.
That matters because YouTube viewership is now serious. Pew Research Center found that 32% of US adults regularly got news from YouTube in 2024. The number of hours flowing through the platform every day is enormous. A tool that scans all of that on your behalf is a tool with a lot of data about you, even when it means well.
The privacy-first approach page covers what the extension processes in full. The short version is: less than you would expect, on purpose.
Where Auto Might Make Sense, And Where Manual Wins
Auto fact-checking fits centralised moderation: platform-level work that decides what stays up, what gets a label, and what comes down. User-triggered fact-checking fits the viewer: a tool you reach for, not a system that watches over your shoulder. Both have a place. They answer different questions, and only one of them belongs in a browser extension.
The honest version of this argument is not that auto is wrong. It is that auto belongs upstream.
Platforms have a real role in removing content with clear, serious harm. They have moderation teams, policy lawyers, and appeal processes for that work. An extension running auto checks on your screen has none of that, and it does not need to. It is not a court of last resort. It is a viewer tool.
A viewer tool should do one job. Answer the question the viewer asked, with sources, when the viewer asked it.
Pick the Claim. Get the Sources.
WasThatTrue is built around a single moment: the second you hear something and want to know if it is true. The whole product flows from that click. Nothing runs until you ask. Only the claim you pick gets checked. The verdict comes back with real sources you can open.
If you have been waiting for a fact-check tool that respects your attention, your time, and your privacy, this is the one. It is free to start on the free and Pro plans, and the Free tier does not need a credit card.
Add to ChromeFrequently Asked Questions
What does user-triggered fact-checking mean?
User-triggered fact-checking is a workflow where verification only runs after a deliberate viewer action. The tool surfaces nothing until you press a button, and even then it checks just one claim you point to. The model is the answer engine, and the viewer is the one asking the question.
Why does not WasThatTrue scan every claim automatically?
Auto-scanning creates noise, reads as moderation, and processes more of your viewing than necessary. Most spoken sentences are not factual claims. A focused check on a claim the viewer actually cares about is more useful than dozens of verdicts on lines they ignored.
Is auto fact-checking the same as censorship?
Not by itself, but the two read similarly from the viewer seat. When a tool quietly decides what is suspicious and flags it, it inherits the suspicion that follows any moderation system. User-triggered design avoids the problem by leaving target selection with the viewer.
How is WasThatTrue different from YouTube fact-check panels?
YouTube panels appear above some search results, not inside videos. WasThatTrue runs at the moment a viewer hears a claim mid-video, on a claim the viewer selected, and the verdict card links straight to the source articles behind the rating.
Does WasThatTrue work without me clicking?
No. The extension never verifies claims on its own. It waits for a click on the WasThatTrue button, then offers a short list of factual claims pulled from captions near that moment, and runs the check on the one you choose. The current scope is covered on the supported use page.