Skip to content
TruthRanking
Evidence verification for the open web

Everyclaimhasafirst source.
We find it.

TruthRanking traces what's being said online back to where it started, rates each claim against real evidence, and builds a documented case to correct it. Then it tracks whether the search results actually change.

Watch a trace
  • Claimsrated, never people
  • Correctionsapproved by a person
  • Evidencecaptured privately
“The school closes next fall”Contradicted“The agenda lists a parking study”Supported“Shared 1,240 times, so it's true”Contradicted“Enrollment is down 40%”Needs context“The district already signed a contract”Unverified“The council has failed us”Opinion“The council met on Oct 6”Supported“Nobody was told”Unverified

Live trace · fictional example

Start with what people see. End at what actually happened.

An AI summary repeats a claim. It cites a blog. The blog cites a post. The post misread a public record. Follow it down.

Riverbend is a fictional town. Every name, date and document here is invented.

  1. AI summaryOct 14

    “Riverbend Elementary will close next fall so the district can build a parking garage on the site.”

    cites → riverbendlocal.blog
    Indexed
  2. Local blogOct 9

    Repeats the post nearly word for word. No document linked.

    copied from → group post
    Copy · not corroboration
  3. Community group postOct 7

    “Saw it on the council agenda — they're closing the school for a garage.” Shared 1,240 times.

    earliest public mention
    First seen
  4. Primary recordOct 6

    Council agenda item 7b: “Parking study — district administration building.” No item mentions Riverbend Elementary.

    capture 7f3a…e19c · SHA-256
    Original source

Fifty copies of one post look like fifty sources. They're one source.

Search and AI engines repeat what ranks. A claim that's posted, reposted and summarized enough starts to read as settled fact, even when nobody ever checked the original.

Finding the bad result isn't enough. It has to be traced to its root, measured against evidence, and corrected where it started. Otherwise the engines just find it again.

Did it work?

Watch the results move.

TruthRanking records what search and AI summaries show before and after a correction, so you can see whether the source changed and whether the engines followed. Drag the timeline, or let it play.

  • Results that repeat the claim
  • Primary records
  • Published corrections

Simulation with fictional data. Real results depend on the platform, the publisher and the search engines.

riverbend elementary closing
Illustrative · fictional

AI summary

Riverbend Elementary will close next fall so the district can build a parking garage.

  1. 1

    riverbendlocal.blog

    Riverbend Elementary to close for parking garage

  2. 2

    valleycommunity.news

    Parents rally to save Riverbend Elementary

  3. 3

    social repost

    “They're closing the school for a garage”

  4. 4

    riverbendschools.org

    Riverbend Elementary · School profile

  5. 5

    riverbend.gov

    City council agenda, Oct 6 (item 7b: parking study)

  6. 6

    riverbendschools.org

    District facilities plan 2026–2030

Week 1 of 8

Claim spreading across results

Top 5 repeating claim

3/5

How it works

Six steps. One evidence trail.

  1. 01

    Monitor

    Choose the topics, names and places to watch. Paste any link you find: articles, posts, videos, petitions. Everything lands in one workspace.

    Watching

    • Riverbend Elementaryname
    • Riverbend school districtorganization
    • “school closing”phrase

    New links

    • riverbendlocal.blog/…captured
    • community group postcaptured
  2. 02

    Isolate the claim

    Each checkable statement is pulled out once and typed: factual, allegation, opinion or prediction. Mixed statements are split so each part is judged on its own.

    Statement

    “The school will close next fall so the district can build a garage.”

    Split into

    • School will close next fallFACTUAL
    • …to build a parking garageALLEGATION
  3. 03

    Trace it to the root

    Find the earliest source and mark what was copied from what. Reposts stay on the chain, but they never count as independent confirmation.

    Source chain

    1. AI summaryindexed
    2. Local blogcopy
    3. Group postfirst seen
    4. Council agendaoriginal
  4. 04

    Evaluate the evidence

    Link the documents, records and studies that bear on the claim. A rating counts only when a reviewer approves it, and “unverified” is never treated as “false.”

    Linked evidence

    • Agenda item 7bcontradicts
    • District facilities plancontradicts
    • Enrollment reportcontext

    Rating

    Contradicted · approved by reviewer

  5. 05

    Approve the correction

    Draft a reply, a correction request or a takedown case with the exact wording, the proof and the remedy. A second person approves it, and a person sends it.

    Correction draft

    • Where it goesblog editor
    • Wording in dispute“will close next fall”
    • Remedy requestedcorrection + link

    Approvals

    1 of 2
  6. 06

    Track what changed

    Recheck the source and record what search results show before and after. See whether the source came down and whether the engines moved with it.

    Results repeating the claim

    week 1 → week 7 · illustrative

    Source corrected

The vocabulary

Not true or false. What the evidence actually supports.

Every claim gets a type before it gets a rating. Opinions are labeled, not judged. A claim is marked contradicted only when specific evidence contradicts it, and that evidence is linked.

Claim types
Factual
Allegation
Opinion
Prediction

Cause-and-effect claims get their own check. “A parking study means the school is closing” is tested link by link, not as one sentence.

Evidence you can stand behind

If it gets taken down, you still have proof it existed.

Before any correction goes out, every source is captured privately: the full-page screenshot, the original HTML, the capture time and a SHA-256 fingerprint that shows the file hasn't been altered since.

Captures stay in your private workspace. They're never sent to a public archive, so correcting a claim never republishes it.

Full-page screenshot
Original HTML
Capture timestamp
Private storage
Capture record
Source
riverbendlocal.blog/school-garage
Captured
2026-10-14 09:12:44 MT
Files
Screenshot · HTML · metadata
SHA-256
f225b69355de62697134e589be5c28b39387975690923a3c5ae76a66b99133f7

Matches the original capture

Principles

What TruthRanking won't do.

The rules are built into the product, not left to good intentions. They're what make a correction credible to the platform or publisher reading it.

  1. 01

    Rate people

    Statements get rated. People don't. There are no honesty scores, no motive verdicts and no dossiers.

  2. 02

    Send anything by itself

    Every reply, report or takedown is a draft until someone who didn't write it approves it. Then a person sends it.

  3. 03

    Give anyone a free pass

    Clients, critics, agencies, experts and the press are all checked the same way, including the client's own statements.

  4. 04

    Manufacture agreement

    No fake supporters, no fake endorsements and no flooding platforms to bury legitimate criticism.

  5. 05

    Promise a removal

    TruthRanking builds the documented case. The platform or publisher makes the call.

Who it's for

For anyone a false claim is costing something.

Families and individuals

When a post about someone you love spreads, get the facts in order before you respond.

  • Private captures of every post
  • Plain-language verdicts
  • No public filing without your approval
A man at a kitchen table at dusk reading his phone with a concerned expression

Questions

Straight answers.

Pilot program

Bring us the claim. We'll show you where it started.

Pilots start with one topic and the links you already have. In the first week you'll see each claim isolated, traced and matched to the evidence that's still missing.

Sign in

mike@mmivip.com