Not every false claim is a deliberate lie. Some come from outdated information, missing context, honest mistakes or statistics presented in the most convenient possible way. But when you encounter hundreds of claims every day, the distinction does not always matter.
The effect is the same: something inaccurate enters your understanding of the world.
A podcast host references a study. A YouTuber explains a historical event. A politician quotes an economic figure. A livestream displays a chart for five seconds. A social post pairs an alarming headline with an unrelated image.
Most of these claims sound plausible. That is usually enough.
The internet made publishing information almost effortless, but verifying it remains inconvenient. Checking one claim can mean pausing what you are watching, identifying exactly what was said, opening several tabs, searching for the original source, comparing conflicting accounts and deciding which evidence deserves your trust.
Believing a claim takes seconds. Verifying it can take ten minutes.
That imbalance is the problem FactLens is designed to address.
Fact-check anything you watch, as you watch it
FactLens is a Chrome extension that checks factual claims while you consume content online.
For audio and video, it listens only after you start a session, transcribes what is being said, identifies externally verifiable claims, retrieves relevant sources and displays an evidence card inside a floating panel. The result appears alongside the content, so you do not have to stop watching, copy a quotation or move between several browser tabs.
The process follows a deliberate order:
transcription → claim detection → source retrieval → evidence verification → evidence card
The new FactLens experience makes this process easier to understand and control without changing that underlying sequence. Audio and video checking is separated clearly from image and post checking, while transcription, AI and search remain independent choices rather than being bundled into one opaque system. This distinction matters. A transcription service converts speech into text. A search provider retrieves evidence. An AI model analyses the claim and the available material. Each component has a different job, and FactLens lets you choose how each part of the pipeline works.
Evidence, not an answer you are expected to trust
FactLens is not intended to become another authority that tells you what to believe.
A label without evidence simply replaces one trust problem with another. Instead, FactLens shows the claim, the result, a concise explanation and the sources behind it. You can open those sources, examine the evidence and decide whether the conclusion is justified.
That is the principle behind the product:
You choose sources. FactLens provides evidence.
Source preferences give you direct control over retrieval. You can prioritise domains you consider useful, identify trusted sources and block sites you do not want included. Blocked domains are removed before evidence is presented, while preferred sources are given greater priority when they are relevant.
This does not make the result automatically correct. Search results can be incomplete. Sources can conflict. AI-generated summaries can misunderstand context. The purpose of FactLens is not to eliminate judgement but to make informed judgement less inconvenient.
Not every claim is simply true or false
Real-world claims rarely fit into a clean binary.
A statement may contain a correct central point but get an important detail wrong. It may use accurate numbers while omitting the context needed to interpret them. It may describe an event that has not yet been independently confirmed. It may not be a factual claim at all.
FactLens therefore supports more specific outcomes:
True when independent evidence supports the claim as stated.
Mostly True when the central claim is supported but a detail is imprecise or incomplete.
Misleading when technically accurate information is framed in a way that creates a false impression.
False when the available evidence directly contradicts the claim.
Unverified when there is not enough reliable evidence to reach a conclusion.
Opinion when the statement is subjective rather than externally checkable.
Larping when someone is performing a persona, exaggerating or “doing a bit”, unless the performance includes a concrete factual claim.
The label is only a summary. The evidence card is the part that matters. You can even create your own evidence cards and logics.
Misinformation is not limited to spoken words
Some of the most persuasive claims online are visual.
A chart can begin its axis at a misleading value. A screenshot can remove the surrounding conversation. An old photograph can be presented as evidence of a current event. A meme can combine a real person, an invented quotation and the branding of a legitimate news organisation.
FactLens can scan visible images and posts on a page, remove duplicates locally and let you review which items should be checked. With a vision-capable model, it can analyse screenshots, charts, memes, social posts and other visual material, then use search to retrieve supporting or contradictory evidence. Image checking is treated as its own workflow and does not unnecessarily pass content through an audio-transcription provider.
The user remains involved in the process. FactLens identifies visible material, but you choose what is actually submitted for checking.
Choose the tools behind the result
There is no single provider that is best for every user, every language or every type of content.
FactLens therefore separates its transcription, AI and search configurations. You can use supported commercial services, connect compatible custom endpoints or run parts of the workflow through services you host yourself. Audio and video sessions can use one configuration, while image and post checks can use a different vision and search setup.
This approach provides more control over cost, speed, privacy and model quality. It also avoids forcing every user to trust the same company at every stage of the verification process.
Provider keys remain on the device and are sent only to the service they authenticate. Audio is processed by the transcription provider you select; claims and relevant evidence are processed by your chosen AI service; and only selected images or posts are sent for visual analysis.
See what the system has actually checked
The redesigned experience also makes the extension’s activity easier to inspect.
Session history and analytics can show how many claims were checked, how results were classified, which subjects or speakers appeared most often and how frequently external APIs were used. This is not a score for deciding whether a person or source is trustworthy. It is a record of what FactLens processed and how the available evidence was classified.
That transparency matters because automated fact-checking should itself remain open to scrutiny.
A system that analyses other people’s claims should not hide its own behaviour.
Verification should not require leaving the conversation
You are going to encounter false, misleading and unverifiable claims every day. No extension can prevent that, and no model can determine truth perfectly.
What software can do is reduce the cost of checking.
FactLens brings the claim, the relevant sources and the evidence summary into the same place as the content itself. It allows you to choose the providers, shape the source rankings, inspect the supporting material and reach your own conclusion.
The internet does not need another black box handing down verdicts.
It needs better tools for seeing the claim, checking the evidence and deciding for yourself.