Hifiverse

AI and Machine-Learning Policy

Hifiverse.io

Last updated: 6 May 2026 Version: 1.1

Our position in one sentence

Hifiverse does not use ML and generative AI to write articles, news, reviews, headlines, product descriptions, buying - guide entries or any other prose published under a Hifiverse byline. All editorial content is written and reviewed by named human audio experts.

Between June and October 2025 we tested an automated “AI Summary” feature on component - catalogue pages. That feature was permanently removed on a rolling basis through October–November 2025. By the end of November 2025, no AI - generated text - visible to users or hidden in markup - remained anywhere on the site, and none has been added since.

Why we are publishing this policy

In a period when much of the web is being filled with synthetic text - automatically generated reviews, automatically translated catalogue copy, automatically assembled “best of” lists - our readers, and the search engines that bring them to us, deserve clarity about how Hifiverse content is made.

This page is also a commitment. By publishing what we do, what we do not do, and what we tried in the past, we make it harder for ourselves to drift toward shortcuts later. If our practices change, we change this page first and document the change in the changelog. We do not change practices and update the policy quietly afterwards.

How Hifiverse content is actually created

Every review, news article, comparison, how - to guide, buying guide and feature on Hifiverse is written by a named human author from our editorial team or our roster of disclosed freelance contributors. Every piece goes through the same workflow:

  1. An editor proposes a topic in the weekly editorial meeting, based on real reader interest, what is genuinely new in the audio market, what we have tested recently and have something distinctive to say about, and the long - running coverage commitments. Topic selection is not algorithmic.
  2. The writer is assigned. The writer researches the subject in primary sources (manufacturer documentation, AES/JAES papers where relevant, the original engineer’s own words, owner - community forum threads, our own archive of past coverage, and conversations with engineers when they are willing to speak.
  3. The writer listens to the equipment in their own room, on their own reference system, for the minimum periods set out in the Editorial Policy (two weeks for components, four weeks for speakers, longer for cartridges and tube equipment that are still changing during the listening window). The writer takes their own notes, runs their own measurements where applicable, and photographs the equipment in their own listening room.
  4. The writer drafts the prose from scratch, by hand, in their own words.
  5. A second editor - a different person from the writer - checks the draft for factual accuracy, internal consistency, fairness, and adherence to house style. Long - form reviews and any “best of” or comparative piece receive a second pass from a senior editor before publication.
  6. The final version is published with the writer’s byline, a link to their public profile page, the publication date and the modification date.

Our authors are practising audiophiles and journalists who have built personal listening installations over years or decades, who have attended audio shows in person, and who have hands - on experience with the categories they cover. Their profiles on hifiverse.io/company/editors disclose the equipment they own, the rooms they review in (with a published photograph of each main listening room), the brands they have tested, the shows they have attended in the past two years. We believe this kind of accountability is the single most important signal of whether a piece of audio writing is worth your time, and we hold ourselves to it.

What we do not do

We do not use generative AI to write Hifiverse content. Concretely:

  • We do not use ChatGPT, Claude, Gemini, Mistral, Llama - derived assistants, Perplexity, You, Copilot, or any other large language model to draft article text, headlines, subheadings, captions, alt text, summaries, news leads, product descriptions, listing descriptions on the Marketplace authored by Hifiverse editorial, buying - guide entries, comparison - table prose, or newsletter copy.
  • We do not use AI to paraphrase or rewrite human writing - our own or anyone else’s - in order to produce additional articles. We do not “spin” content. We do not run an existing article through a model to make a “fresh” version of it.
  • We do not run human - written text through AI “humanisers” to disguise machine origin. The premise of that tooling - that you have something to disguise - disqualifies it from a publication that wants to be trusted.
  • We do not generate review verdicts with a model, even where the model is fine - tuned on audiophile content. A verdict reflects first - hand listening by a named reviewer in a disclosed room on a disclosed system; a model has never heard a pair of speakers.
  • We do not generate product photography with a model and present it as photography of a real unit. Where we publish AI - generated illustration (a diagram, a conceptual figure, an explanatory rendering) it is labelled as illustration in the caption.
  • We do not generate measurement data with a model and present it as measurement. Measurements are taken on real equipment under stated conditions with stated test gear, and the methodology is disclosed in the article.
  • We do not use AI to automatically generate articles based on trending search queries, regardless of how the resulting text is wrapped or labelled.

If you discover Hifiverse content that you believe was generated by a model in violation of this policy, please email info@hifiverse.io with subject line “AI / ML Policy” or “Corrections” and the URL. We read every report.

Where we do use software, narrowly

There are places software is unavoidable in modern publishing, and pretending otherwise would be dishonest. The list below is exhaustive of what we do.

Spell - checkers and grammar tools. The kind built into word processors, browsers, and editorial CMS plugins. These read text the human author has already written and suggest mechanical corrections. They do not generate prose.

Search engines and reference databases. Google, DuckDuckGo, manufacturer websites, AES/JAES archives, patent databases, audio - measurement repositories such as the various manufacturer and laboratory archives, owner forums. The output of that research is read, evaluated and synthesised by a human author, not piped into a model and condensed.

Machine translation as a research aid. When a relevant source is in a language a writer does not read fluently - a Japanese review of a vintage cartridge, a German engineering paper on driver design, a Russian forum thread on a Soviet - era amplifier restoration - the writer may use a translation tool to read the source. The resulting article is then rewritten by the human author from primary sources, with the original - language source cited and quoted in its original language where the precise wording matters.

Transcription of recorded interviews and listening sessions. Where a writer records an interview with an engineer, designer or industry figure (with consent), or a spoken listening - session log, a transcription tool may be used to produce the working transcript. The article is then written by hand from the transcript and the writer’s own notes.

Code, scripts and data analysis. Where we work with data - measurement files, the Hifiverse Price Index pipeline, log analysis, dataset preparation for editorial coverage of measurement trends - the editorial and engineering teams may use coding assistants. The resulting narrative in any published article is written by a human; the code behind a tool may be assisted, with human review.

Standard analytics. Server logs, Plausible Analytics, search - console data and similar. We use these to understand what topics readers care about and how they navigate the site. We do not use them to “automatically generate articles based on trending queries.” Editorial decisions are made by named editors in editorial meetings, on the basis described in the Editorial Policy.

Image processing. Standard photo - editing tools (cropping, exposure, white balance, perspective correction). We do not use generative tools to add or remove objects from a photograph of equipment under review. Where AI - assisted background removal is used for a product silhouette, the underlying photograph is a real photograph of the real unit.

Email and operational chat. Standard productivity tools used for internal communication. We do not auto - reply to reader messages with model - generated text; reader messages to info@hifiverse.io are read and answered by humans.

If a tool we use materially changes how a piece of editorial content is made - for example, if at some future point we evaluate model - assisted first - pass research summarisation as a productivity aid for fact - checkers - we will update this policy and the changelog before adopting it, not after.

What happened with the “AI Summary” feature in 2025

Between June and October 2025 we ran a five - month experiment on component - catalogue pages - for example, the page for an integrated amplifier such as the Audiolab M5xi or the Naim NAIT XS 3. On those pages, alongside the human - written component description, we displayed a small “Get AI Summary” button. When a reader clicked it, an overlay appeared containing an automatically generated summary of what owners and reviewers had written about that product on public audiophile forums.

The text inside that overlay was generated by a large language model from publicly available forum posts. It loaded only on click and was not part of the page’s indexed HTML, but it was visible to users who chose to open it. The component description above the overlay, and the rest of the catalogue, was always written by humans.

We launched the feature because we thought a quick distillation of “what people are saying” could be useful to a time - pressed reader who wanted a gut - check signal on a new component before reading a long - form review. In practice, three problems emerged.

Removal timeline. The feature was removed on a rolling basis through October and November 2025. By the end of November 2025, no AI Summary overlays remained on any page of hifiverse.io.

No AI - generated text - visible to users or hidden in markup - exists anywhere on hifiverse.io as of the date at the top of this page.

Disclosure of any future use of AI

We will not adopt a new ML - assisted editorial practice silently and we will not normalise it with marketing language.

Corrections, fact - checking and accountability

If you find something on Hifiverse that you believe is inaccurate, generated by a model in violation of this policy, or otherwise not up to the standard described here, please email info@hifiverse.io with subject line “Corrections” or “AI / ML Policy” and the URL. We read every message and respond within five working days.

Tipsters whose reports lead to a correction or to a policy change are credited (with permission) in the correction note or in the changelog below.

The full corrections process - what counts as a correction, what triggers republication of an article, where corrections are logged publicly - is described in the Editorial Policy section “Corrections and updates” and in section 11 of the Editorial Ethics & Independence Policy.

How to reach us

For questions about this policy, suspected violations, or proposals for narrow, well - bounded ML use that you think we should consider, write to info@hifiverse.io with one of the following subject lines:

  • “AI / ML Policy” - questions or concerns about this specific policy.
  • “Corrections” - to report a factual error or suspected AI - generated content on the site.
  • “Editorial” - general editorial correspondence.
  • “Editorial Ethics” - escalations about the editorial - commercial firewall.
  • “Data Licensing” - for AI - system operators wishing to negotiate a licence to use Hifiverse content for training purposes.

The full subject - line index for the single - mailbox model is in section 18 of the Hifiverse Terms of Service. If your enquiry doesn’t match any subject line, write “General” and we will route it. We do not reject messages because of missing or wrong subject lines.

Changelog

  • 6 May 2026 - version 1.1. Edited and expanded