Give it the right scope
Check the full page, a selected passage, or text you paste privately. Narrow scope answers a narrow question more reliably.
FAQ · product guide · real checks
Read the score. Keep the question.
ClearWater is a reading instrument, not a truth machine. This guide shows what it checks, how to interpret a result, and where your own judgement must take over.
01 · Start here
One action. Three decisions.
A useful check should change how you read the passage, not replace reading it.
Read past the number
The band and score summarize the run. The explanation, marked fragments, and key observations show what actually drove it.
Return to the source
Check the cited passage, inspect the missing context, and decide whether the result survives contact with the original page.
02 · Result anatomy
The score is an index, not a probability.
A result combines a verdict band, a normalized score, and the observations that shaped it. None of those means “42% true.”
Example result band
Questionable
Truthfulness relative to the active criteria
0.42
Band
A compressed orientation
“Questionable” or “Suspicious” tells you where to start looking. It is not a final judgement about the author or publisher.
Score
Relative to this run
The decimal is normalized against the criteria, scope, model, and mode used for that check. Compare the evidence before scores.
Observations
The actionable layer
Contradictions, weak links, unsupported transitions, and marked fragments are the part you can verify against the text.
03 · Real checks
Three results where facts passed and the text still broke.
These are excerpts from real ClearWater runs supplied on 20 July 2026. Source identities are withheld or generalized because the examples teach result interpretation; they are not endorsements or accusations.
Real check 01 · article · focused analysis
Battery explainer
0.42
Questionable
What held
The check found real substance in the history of batteries, degradation mechanisms such as SEI and lithium plating, and descriptions of future technologies.
What broke
It surfaced an internal mismatch: Li-Po cycle counts of 1,500–2,000 and Li-ion counts of 500–1,000 were described as “roughly the same.” It also found that the article drifted into history and chemistry without returning to its stated question: why a charge does not last all day.
The useful output was not “do not trust batteries.” It was a concrete reading task: reconcile the numbers and ask whether the article ever answered its own headline.
Result excerpt condensed for layout without changing the score, verdict, factual-status badge, or identified contradictions.
Real check 02 · community full-page · GLM-5.2 · deep · high
Storage market outlook
0.55
Suspicious
What held
Technical sections on MAMR, HAMR, and glass substrates contained concrete material and were assembled into linked paragraphs.
What broke
The check reported tension between a “moves forward” headline and the page’s own market-share and shipment-timing data. It also marked a numerical transition from 3.6 to 3.3 that sat beside a later claim that density “increases.”
It additionally raised omitted competitive context as a question. That observation requires human verification: ClearWater can propose the missing comparison, but the reader must verify it against current sources.
A “Facts verified” badge records the fact-search outcome for that run. It does not certify completeness, framing, or future accuracy.
Real check 03 · full-page · Gemini 3.1 Flash-Lite
VPN recommendation guide
0.40
Doubtful
What held
The page made concrete, testable product claims about speed, server coverage, device limits, price, and privacy features. It also stated its preferred option clearly enough for the ranking logic to be inspected rather than guessed.
What broke
The run found an internal conflict: one provider was described as having the largest server network while the guide’s own comparison listed a higher server count for another provider. The ranking leaned on personal use and superlatives without exposing a reproducible benchmark.
It also grouped a hardware privacy product into the VPN list without clearly separating product categories. The useful result was not “never buy this provider.” It was a concrete reading task: verify the counts, separate product facts from affiliate-style ranking language, and ask what method produced “best overall.”
The source identity is generalized. Score, verdict, model, factual status, and both key-finding impacts are preserved from the supplied run.
04 · The honest boundary
What a result can carry. What it cannot.
ClearWater is strongest when its output becomes a testable question, and weakest when its number is treated as authority.
A result can help you
- Locate contradictions and weak transitions in the submitted text.
- Separate concrete claims from confidence, pressure, or filler.
- Corroborate claims that were actually made when fact-search is enabled.
- Turn a vague feeling of “something is off” into passages you can inspect.
A result cannot guarantee
- That every relevant fact or counterargument was included.
- That a model-generated observation is itself correct.
- That two runs with different scope, mode, or time are directly comparable.
- That a high score makes the page safe to believe without checking it.
05 · Questions
Short answers for normal use.
Open only what you need. Product behavior and privacy boundaries are stated directly.
01 What exactly is ClearWater?
ClearWater is a browser reading instrument. You ask it to inspect a full page, selected passage, or pasted text. It returns a criterion-relative score, a verdict band, an explanation, and marked fragments worth checking. It does not block pages or make decisions for you.
02 Does 0.42 mean the text is 42% true?
No. The score is a normalized trust index relative to the criteria, scope, model, and mode used in that run. Read the explanation and fragments first. Comparing two numbers without matching those conditions can be misleading.
03 What does “Facts verified” actually mean?
In that run, the fact-search path corroborated the claims it checked. It does not mean every sentence was checked, every source was independent, or every relevant fact appeared in the text. The real examples above show why factual corroboration and structural quality must remain separate signals.
04 Can ClearWater detect missing context?
It can sometimes notice that an argument needs context or propose a missing comparison. It cannot prove that nothing important was omitted. Absence leaves no passage to inspect, so verify proposed omissions against current primary sources.
05 When does it send text to the server?
Text is sent when you explicitly start a check. Opening the popup on a page may send that page’s URL to request its community score, but not the page text. The public service processes submitted text in zero-retention mode: it is not retained in server history after the result is produced.
See the Privacy Policy for the exact data flow and community-index controls.
06 Is community sharing automatic?
No. Community contribution follows the setting you choose: off, on, or ask each time. Custom pasted text remains private and is not contributed to the community index.
07 Do I need an email or password?
No conventional registration is required. ClearWater uses a random account key for balance and allowance. A separate random recovery token can reconnect an existing account after a reinstall. Keep the account key private: anyone who has it can use the associated balance.
08 What does a check cost?
A daily free allowance is available. Paid checks draw from a prepaid US-dollar balance, and deeper or longer runs cost more. ClearWater shows a quote before a paid check starts. Current payment and refund terms are listed in the Terms of Service.
09 What should I do when I disagree?
- Check whether the selected scope contains the disputed claim.
- Read the marked fragment in its original surrounding context.
- Verify factual observations against current primary sources.
- Rerun only if you intentionally change the scope or mode.
A disagreement is not a reason to obey the tool or ignore it. It is a reason to identify which observable would settle the question.
10 Where can I report a problem?
Product questions and result feedback: support@wecleantheinternet.com. Privacy questions: privacy@wecleantheinternet.com.
The useful loop