Where do the facts come from?

For a Connected brand, the canon is your verified brand record: facts confirmed about your business, kept current, with the record itself as ground truth. This public sample ran without one, so its ground truth was assembled from the brand's own published site plus live web verification, and every finding shows the sources it was checked against.

How current is a report?

Every report is a stamped run: the answers are captured live at run time and verified against the live web in the same run, and the date at the top is the date it happened. A report is a snapshot, re-run on demand, not a feed.

What if the engines say nothing about us?

That is a finding too. An empty or refused answer produces zero claims by design; the extractor never invents a claim just to have something to grade. Knowing the engines have nothing, or only something vague, is the honest starting line the rest of Jinn works from.

Why show a full sample instead of letting me run it once free?

Because a crippled run teaches you nothing and a blurred one insults you. We would rather show the entire depth of one real report and be plain that running it on your brand is paid. Gate the tool, never cap it.

Brand Fact Checker

What the engines get wrong about you.

AI engines are already answering questions about your brand, and some of what they say is wrong. The Brand Fact Checker captures exactly what they believe, checks every claim against verified ground truth, and writes the corrections.

Sample

One real run, rendered in full.

Liquid Death (liquiddeath.com)Run 2026-07-12 (UTC)

Liquid Death is one of the ten brands in Jinn's published engine-accuracy research. This is one real run, not a mockup: 6 engines, 3 questions, 2 samples each, 36 verbatim answers, 268 claims graded. What follows is a representative sample rendered exactly as the machinery wrote it; the full run is the paid deliverable. Separately, 9 findings are withheld from publication entirely: their corrections conflict with the brand's own published statements, and we do not publish contested accusations.

What went in

The exact inputs, snapshotted verbatim. Every engine got the same three questions a real buyer might ask.

Engines sampled: ChatGPT, Perplexity, Gemini, Claude, Claude with web search, Grok
Sampling: 2 independent samples per engine per question (36 answers total)
Question 1 (the company facts): Tell me about the brand Liquid Death. Who founded it and when? Where is it headquartered? Who owns it now and who is the CEO? What are its main products and roughly what do they cost?
Question 2 (the buyer lens): I'm considering buying from Liquid Death. What should I know about the company before I do — its reputation, any recent major changes or news, and anything controversial?
Question 3 (the staleness probe): What has changed at Liquid Death recently — leadership, ownership, pricing, or policies?

The run extracted 268 checkable claims from those answers. A checkable claim asserts something with a truth value: a founding year, a headquarters, a price, a policy. Vibes are not claims, and an empty or refused answer produces zero claims by design.

Every claim was graded against ground truth assembled from Liquid Death's own published site, and the suspect ones were then individually verified against the live web before anything below got its label. One rule matters more than the rest: when the engines agree with each other but disagree with the brand's own site, the site wins. Engines fail in correlated ways, so consensus is not truth.

What failed the check

Every published claim graded wrong, stale, or invented, with the correction and the sources behind the verdict. The engines named a founder who does not exist, listed investors who never invested, described a product line that was never made, halved the valuation, and misdated the funding round. Here are the first five; the rest are in the full run.

WrongPerplexity, asked for the company facts

“The concept was born in **2008** at the Warped Tour”

The concept was born in 2009 at the Warped Tour, not 2008.

Checked against: en.wikipedia.org

OutdatedClaude with web search, asked for the company facts

“Liquid Death CEO and founder Mike Cessario has a brand valued at $700 million”

As of March 2024, Liquid Death reached a $1.4 billion valuation, double the $700 million valuation from October 2022.

Checked against: en.wikipedia.org · bloomberg.com

FabricatedClaude, asked for the company facts

“Mike Cutler remains the founder/CEO”

The founder and current CEO of Liquid Death is Mike Cessario, not Mike Cutler.

Checked against: en.wikipedia.org

WrongGrok, asked for the company facts

“Iced teas (Armored; launched 2023–2024)”

Liquid Death launched iced teas in 2023 with flavors such as Grim Leafer, Rest in Peach, and Armless Palmer (later renamed Dead Billionaire), but no iced tea line called 'Armored' exists.

Checked against: en.wikipedia.org

OutdatedGemini, asked what a buyer should know

“now sold in 113,000 retail locations across the U.S. and the U.K.”

Liquid Death is now available in over 133,000 retail locations, primarily across the U.S., as of late 2025.

Checked against: fueler.io · microventures.com · marcom.com

+ 16 more flagged claims checked in this run

Run it on your brand

Held at the floor

These read wrong, but the sources the verifier surfaced did not clear the quality bar, so they publish downgraded rather than as accusations.

+ 6 held at the no-source floor in the full run

Run it on your brand

The seven verdicts

Every claim lands in exactly one of these.

Correct: Confirmed currently true.
Wrong: Currently false, with a quality-bar source behind the correction.
Outdated: Was true once; a later change made it stale.
Fabricated: An invented specific with no evidentiary basis anywhere.
Vague: Too soft to carry a truth value.
Unverifiable: No source could confirm or deny it.
Unverified disagreement: Reads wrong, but no quality-bar source exists. The downgrade floor: never published as Wrong.

Everything else the engines said

A few of the claims that were not findings, each with its verbatim quote; the rest of the table is in the full run. One deliberate absence: no aggregate accuracy number appears on this page. Our own grader has not yet cleared the accuracy floor we hold it to, and the discipline that refuses to round that corner is the same discipline that graded these claims.

Correct (6)

Claims the run confirmed as currently true.

“Notable individual investors include actors Josh Brolin and comedians Whitney Cummings and Tom Segura”ChatGPT, asked for the company facts
“founded in 2018”ChatGPT, asked what a buyer should know
“2.2 out of 5 rating based on eight reviews”ChatGPT, asked what a buyer should know
“in June 2025, the company released a limited edition iced tea in collaboration with Ozzy Osbourne”ChatGPT, asked what a buyer should know
“Los Angeles-based beverage company founded in 2018”ChatGPT, asked what a buyer should know
“founded in 2018 by Mike Cessario”ChatGPT, asked what a buyer should know

+ 226 more claims checked in the full run

Run it on your brand

The sample is the ad. Yours is the product.

Everything above is one run against one famous brand. Run against your brand, this becomes a standing fact canon: the claims engines actually make about you, checked against your verified brand record once your brand is Connected, with corrections ready to publish.

Running it on your brand is paid, and this page will not pretend otherwise. There is no free tier of this tool and no crippled preview. You saw the whole depth of one real report instead.

Apply for early access

Checkout for this tool is not open on the site yet. Apply for early access and we will bring you in as it opens; pricing publishes with general access.

How it works

Canon, sampling, verification.

A fact canonFor a Connected brand, ground truth is the verified brand record: facts confirmed about your business, not scraped guesses. This public sample ran against the strongest substitute, the brand's own published site.
Engine samplingSix engines get the same buyer-shaped questions, twice each. Answers are captured verbatim, and every checkable claim is extracted with its exact quote. A quote that does not appear word for word in the answer is dropped, so no claim can be invented on the way in.
VerificationSuspect claims are individually verified against the live web. A Wrong verdict requires a quality-bar source; no source means the claim publishes as a disagreement, never as an accusation.

How the whole engine works

Questions

What people ask.

Where do the facts come from?: For a Connected brand, the canon is your verified brand record: facts confirmed about your business, kept current, with the record itself as ground truth. This public sample ran without one, so its ground truth was assembled from the brand's own published site plus live web verification, and every finding shows the sources it was checked against.
How current is a report?: Every report is a stamped run: the answers are captured live at run time and verified against the live web in the same run, and the date at the top is the date it happened. A report is a snapshot, re-run on demand, not a feed.
What if the engines say nothing about us?: That is a finding too. An empty or refused answer produces zero claims by design; the extractor never invents a claim just to have something to grade. Knowing the engines have nothing, or only something vague, is the honest starting line the rest of Jinn works from.
Why show a full sample instead of letting me run it once free?: Because a crippled run teaches you nothing and a blurred one insults you. We would rather show the entire depth of one real report and be plain that running it on your brand is paid. Gate the tool, never cap it.

Run it on your brand