This site makes hundreds of specific claims — prices, minute allowances, battery figures, server locations, owner complaints. A policy that can't survive inspection is decoration, so this one is operational: it defines what each claim rests on, how errors get fixed, how often numbers are re-verified, and where AI tooling fits in the pipeline. It's the constitution our review methodology operates under.

Standard 1

Evidence tiers, defined

Every factual claim about a device carries one of three labels. No exceptions, no unlabeled claims, no invented statistics:

[SPEC]
Manufacturer specification — product pages, pricing pages, official support docs
[REVIEW]
Third-party review — independent publications' hands-on reporting
[OWNER]
Owner reports — Reddit, Trustpilot, Amazon Vine, community threads

The tiers rank reliability: a manufacturer's own spec sheet wins on what the product is, independent reviews win on what it's like, and owner reports win on what happens over time. When tiers disagree — and they do; see the NotePin S battery specs claimed three different ways by three retailers — we publish the conflict and identify the authoritative source (PLAUD official, in that case; the flagged discrepancies are visible in our comparison tool) rather than averaging it away. Items we could not corroborate are flagged inline: the soundcore Work's reported offline model, Pocket's free-minute contradiction, Omi's pricing discrepancy.

Standard 2

Sourcing standards

A claim publishes only when it satisfies the tier system, with these mechanics:

  • Manufacturer claims get corroborated when possible. Specs and prices come from the maker's pages, but we prioritize facts that independent coverage or owner reports confirm — and we say when corroboration is absent.
  • Prices and terms are dated. Everything is “as of August 2026" because this market reprices constantly; we publish ranges where sources disagree rather than false precision.
  • Owner sentiment is quoted verbatim and attributed. Real phrasings from real threads (“Subscription too expensive”) with community attribution — never laundered into anonymous “users report.”
  • Negative findings publish with positive ones. Trust notes, conflicting specs, and complaint patterns appear on the same page as the pros — a device's weaknesses are part of its spec sheet.
  • Legal topics stay educational. Consent-law content frames authoritative resources (Justia, RCFP, Digital Media Law Project) and never becomes advice. See the guide's own disclaimers.

Standard 3

Corrections policy

Fast, visible corrections are the brand. The commitment:

  1. 01Report it: contact form (topic: “Correction”) or hello@airecorderguide.com.

    Include the page and the source that contradicts us — a link beats a memory. Manufacturers may use the same channel; claims get the same verification either way.

  2. 02We verify against sources — within 5 business days for confirmed errors.

    A verified error is fixed within five business days of confirmation, with a note recording what changed. Substantive corrections get a visible note; typos get fixed silently, like everywhere else.

  3. 03Disagreements get adjudicated in public.

    If we decline a “correction” — including one from a manufacturer — we'll say why, citing the same evidence standards. Being wrong in public is survivable; being quietly edited is not.

Standard 4

Update cadence

Stale numbers are errors wearing a date stamp. The maintenance rhythm:

  • Quarterly re-verification of every tracked price, subscription tier, and free-minute allowance against manufacturer pages.
  • Sooner for major market events — a discontinuation, price change, new tier, or safety/privacy incident gets covered and reflected in rankings as it happens (the Limitless shutdown is the worked example).
  • Every page displays its fact-check date in the header; if you spot one older than a quarter, that's a correction-priority report.

Standard 5

AI-assisted drafting disclosure

We use AI tools in our workflow — for drafting, research organization, and maintaining consistency across a large database of fast-moving facts. Here is exactly what that means and doesn't:

What AI tools do

  • · Draft prose from our structured, sourced research notes
  • · Organize device data, spec tables, and comparison matrices
  • · Keep terminology consistent across 40+ pages

What they never do

  • · Originate facts — no claim exists without a source in our research archive
  • · Replace review — every claim is human-editorial-reviewed against sources before publication
  • · Generate fake quotes, statistics, or reviews — owner quotes are verbatim from real threads

The irony isn't lost on us: a site covering AI hardware, disclosing AI assistance. That's the point. The technology is fine; unsourced claims are the failure mode, whatever drafts them.

Standard 6

Conflicts of interest and independence

Current commercial state: no manufacturer pays for rankings, sponsors placements, or has input on editorial scoring. AI Recorder Guide does participate in Amazon Associates, so marked Amazon links can earn commissions. The full commercial picture, current at any moment, is the affiliate disclosure.

The structural guarantee for when that changes: editorial decisions and any commercial relationships live behind a wall that this page defines. Rankings follow the published methodology; revenue can't reorder arithmetic. If a future relationship ever touches a covered product, we'll disclose it on the affected pages and here — and you can hold us to that in writing, via contact, under the same five-day corrections commitment as everything else.

FAQ

Frequently asked questions

They're the evidence tiers behind every claim: [SPEC] means the manufacturer's own product, pricing, or support pages; [REVIEW] means independent third-party coverage like CNET, ZDNet, or Wirecutter; [OWNER] means owner reports from Reddit, Trustpilot, and Amazon threads. When tiers disagree, we publish the conflict and identify the more authoritative source rather than hiding it.
Verified errors are fixed within five business days of confirmation, with a note recording what changed. The report channels are the contact form (topic 'Correction') or email — and yes, manufacturers can use them too; their claims get the same verification as everyone else's.
AI tools draft prose from our structured, sourced research notes and keep a 40+ page site consistent — but no claim exists without a source in our research archive, and every claim is human-editorial-reviewed against those sources before publication. The full disclosure, including what AI never does, is Standard 5 above.
Every tracked price, subscription tier, and free-minute allowance is re-verified against manufacturer pages at least quarterly, and major market events trigger sooner updates. Each page displays its fact-check date in the header.
They can challenge it through the same corrections channel as readers, with sources. If they're right, it changes within five business days with a note. If we decline, we say why, citing the same evidence standards — being quietly edited is not an outcome we offer.

Research sources

Site standards, effective August 2026 · Research archive: /research/devices.md and /research/demand.md (sourced searches) · Typed device database: /src/data/products.ts (verified Aug 2026)

Prices and subscription terms change often. We verified everything above in August 2026; confirm current terms on the manufacturer's page before buying.