How FutureNow Verifies Device AI Claims

How FutureNow verifies what your devices actually do

Phones, earbuds, TVs and smart-home hubs now ship with AI features that rarely appear on the box. The job here is to find out what they do, what data they touch, and how sure anyone can be about it. This page explains how I do that, including where the method falls short.

Last reviewed: October 3, 2026 · Written and maintained by Tom Morgan · Questions and corrections: contact form

Every factual claim about a device, a policy or a number should say how I know it. If I can’t tell you how I know, I don’t state it as fact.

This standard applies to this page’s date forward. Articles published earlier may not carry the labels yet.

The three evidence labels

Most coverage of consumer tech blends confirmed facts, company statements and educated guesses into the same confident sentence. I separate them.

VerifiedConfirmed from a primary document: a regulator’s filing, official product documentation, a published privacy policy or terms of service, or on-the-record confirmation from the company. When I checked behavior hands-on, the article names the device and software version. Either way, the document or test is named.

ReportedA credible outlet, researcher or company representative says it, but I have not independently confirmed it. It is attributed by name, with a note on what I could not check.

EstimatedMy own inference from incomplete data. I show the inputs, the assumption that carries the most weight, and the range, never a single falsely precise figure.

My five-step process

Start from the primary document. Before any commentary, I look for the terms of service, privacy policy, support pages, patent filings, regulatory submissions and release notes tied to the feature. Secondary coverage is a lead, not a source.
Use AI tools for the grunt work, and say so. I use tools such as Grok and Claude to gather and condense public sources, suggest structures and draft early versions. A draft is never treated as evidence. The same disclosure sits on the About page.
Check every fact, number and quote against its source. I open the original, not the summary of it. If a number has no traceable origin, it does not run.
Label every claim. Each material statement gets Verified, Reported or Estimated treatment, applied evenly, so a cautious hedge never appears on one claim and not on another of equal strength.
Edit for overreach. I reread headlines, summaries and social copy against the body and cut anything that claims more than the evidence shows. Where the evidence is thin, the headline is thinner.

What counts as a source

Source type	How I treat it
Regulator filings, official documentation, court records	Preferred. Cited by name and link where public.
Company statements and press materials	Treated as claims by an interested party. Attributed, and checked against documentation.
Independent security researchers and academic papers	Weighted by method disclosure and reproducibility. Peer review noted when it exists.
News reporting	Used as a pointer to primary material. Cited when it holds original reporting.
Forums, social posts, anonymous tips	Never sufficient alone. May prompt a closer look, never a Verified label.
Product reviews and affiliate pages	Not used as evidence for capability or privacy claims.

I do not invent statistics, composite users or testimonials.

Where this method fails

One person, with AI tools

FutureNow has no second editor and no testing lab. That means fewer products covered than a newsroom manages, and a real chance of errors a second pair of eyes would catch. It is the reason corrections are public.

Most checks are against documents, not devices

Unless an article names the device, software version and date of a hands-on check, assume none was done. Unless it says a company was asked and quotes the answer, assume it was not.

Software changes under us

A feature described in one version of the documentation may behave differently after the next update. Hands-on results carry a version and date for that reason.

I cannot see inside the cloud

I can document what a policy says and what a device is documented to send. I usually cannot verify what a company does with data after it leaves the device. Those statements stay Reported at best.

Regional differences

Features, defaults and legal protections vary by country. Unless a page says otherwise, a finding applies to the documentation and region I checked.

Corrections and disagreements

When I get something wrong, I fix it in the article, add a dated note at the top explaining what changed, and never edit silently. If you are a reader, researcher or manufacturer and believe I got something wrong, send the specific claim and your evidence through the contact form. Substantive disputes and my response are noted on the article when they bear on the reader’s understanding.

Independence and money

FutureNow is an independent site written by Tom Morgan. No manufacturer pays for coverage, supplies review units or sees an article before it is published, and nothing here is sponsored content.

If an affiliate link, sponsorship or free product ever enters the picture, it will be disclosed on the article where it appears and recorded in this section with the date. Product reviews and affiliate pages are never used as evidence for capability or privacy claims, as the source table above shows.