AI Transparency Policy

Owner: Joschka Witte (admin@dadchecks.tech) — see the Impressum.
Last reviewed: 2026-11-12. Next review: 2027-05-12.

This page states how artificial intelligence is actually used on dadchecks.tech,
what is labelled, and which legal transparency duties apply. It is written to
match real use. Where a legal duty does not apply, this page does not claim
that it does.


1. Summary

AI is used for Research assistance, draft assistance, mechanical article validation, illustration, routine site maintenance
Who publishes AI drafts; the automated validation agents of the editorial pipeline check the draft and decide whether it may go live. The owner is accountable for what the site publishes, and is the provider named in the imprint.
AI is not used for Deciding conclusions, confirming first-person experience, publishing, or establishing facts on its own
Every article carries A visible AI-assistance notice, the byline “DadChecks Editorial”, and editorial responsibility
First-hand testing None. This site publishes research-based articles, not hands-on tests
AI-generated images Labelled as AI-generated illustrations, individually and in aggregate
Machine-readable marking Applied to the shop product images; not currently present on article illustrations — see § 4

2. How AI is used on this site

2.1 Research assistance

AI helps gather and summarise manufacturer documentation, datasheets, release
notes, and support articles. The owner reads the underlying source. A summary
produced by a model is a pointer to a source, never a substitute for it, and
a claim resting on an AI summary alone does not meet the evidence bar.

2.2 Draft assistance

AI helps structure drafts, tighten prose, and produce first passes at headings
and summaries. The automated validation agents of the editorial pipeline check every article before it can go live. The wording is written with AI assistance, not by a human from scratch. The owner is accountable for what the site publishes, and is the provider named in the imprint; routine research articles do not wait for a personal sign-off.

2.3 Article validation (the core of the process)

Every article passes through an automated validation pipeline before it can be
published. This is a genuine mechanical check, not a formality: it extracts
each factual claim from the article, classifies it, assigns it an evidence
level, and tests it against nine categories — technical feasibility,
real-world plausibility, mathematical correctness, hardware/software
compatibility, source quality, temporal validity, internal consistency,
cross-article consistency, and personal-experience verification.

It also classifies the article into one of five editorial types and enforces the
rule that a title promising hands-on testing must be backed by first-person
evidence. Claims that no rule recognises are reported as unresolved and
escalated to the owner rather than passed silently.

AI does not make the publishing decision. The pipeline produces a report; the automated validation agents of the editorial pipeline check the report, and publication requires that decision to come back clean.

2.4 Illustration

Header images, thumbnails, and shop product images are generated with an AI
image generator. They are illustrations of artwork and concepts. They are not
photographs of physical products, and they are labelled as such — see § 3.

They are also not photographs of anything the author owns. No product on this
site has been physically handled, opened or measured by the author.

2.5 Site maintenance

Routine upkeep — checking that published links still resolve, refreshing
captured prices with a date, and similar — uses automated tooling. These
processes do not alter published conclusions.


3. Image labelling

3.1 Article illustrations

AI-generated article and thumbnail images are illustrations. They are used
for visual design and concept, and they are not presented as photographs of the
author’s own hardware or of a product in use.

An illustration is never used in a way that implies the author physically owns
or has tested the product depicted, when that is not true.

Measured gap, open for the owner. As of 2026-11-12 the article header and
thumbnail images carry no visible AI label on the page. The six published
articles each render a hero image and a card thumbnail, and neither is marked
in the visible UI; the thumbnail is rendered as a CSS background with an
aria-label that repeats the article title. The only visible marking in the
article layer is the per-article AI notice in § 5, which sits in the article
body below the hero image. The shop (§ 3.2) is visibly labelled.

The fix is a visible “AI-generated illustration” marker on the hero image and
on card thumbnails, plus a descriptive alt text that says the image is an
illustration rather than a photograph. That is a theme change, so it is filed
as a follow-up for the owner rather than applied here.

3.2 Shop product images

Every product image in the shop carries a visible notice:

AI-generated product illustration, not a photo of a real item.

The shop page carries a banner to the same effect:

All product images on this page were generated with an AI image generator and
are marked accordingly. They are illustrations of the artwork on the product –
not photographs of a real item.

Product listings state this because a buyer seeing a print-on-demand item
should not mistake a generated rendering for a photograph of a physical
product.

3.3 What is never done

  • Real product photographs are not altered to misrepresent a product’s features
    or the conditions in which it was tested.
  • Screenshots reflect actual software behaviour, or are explicitly labelled as
    mockups.
  • No fake benchmark screenshots and no fabricated test photographs are
    generated.
  • No image is generated to imitate a real person, a real review, or a real
    measurement result.

3.4 Accessibility

Images carry descriptive alt text. Where an image is decorative and carries no
information, its alt text says so rather than describing the picture.

Measured state as of 2026-11-12. The six article hero images use
alt="" — the correct markup for a decorative image, though an
“AI-generated illustration” marker (see § 3.1) would be the better treatment
because the image carries information about provenance even though it carries
no information about the article’s subject. The author avatar uses
alt="josh". Alt text on the theme’s featured images is otherwise
population-level, not article-specific.

Open item for the owner: give the hero and card images alt text that
states both the subject and the fact that the image is an AI-generated
illustration.


4. Machine-readable marking (C2PA / metadata)

Measured state as of 2026-11-12 — this differs by image class:

Image class Machine-readable marking
Shop product images (JPEG, e.g. DC-CAP-001.jpg) Yes — XMP metadata carrying an explicit “KI-generiertes Bild … AI-generated image, synthetically produced” description, the generator named (OpenAI gpt-image-2), the statement that it was not photographed by a human, and an ai-generated=true flag
Article header / thumbnail images (PNG and the theme’s WebP derivatives, e.g. raspberry-pi-5-homeserver-test-1200x630-1.png) No — measured: the PNG contains only IHDR / IDAT / IEND chunks, no iTXt/tEXt XMP chunk, no c2pa / ContentCredentials manifest. The WebP derivative contains only a VP8 chunk with no EXIF or XMP chunk
C2PA content credentials Not present on either class

So: the shop images are machine-readably marked and the article illustrations
are not. That is an honest gap, and it is recorded rather than papered over.

Why this is not a breach of Article 50(2). As set out in § 6.1, Article
50(2) binds providers of AI systems that generate synthetic content — the
model service that produced the image — not a publisher who uses that service.
dadchecks.tech does not provide an AI system, so it does not carry the
provider-side marking duty. The site is nonetheless applying the marking where
it can, and the visible labelling in § 3 applies to both classes regardless.

Open item for the owner (not applied here): extend the XMP ai-generated
metadata from the shop pipeline to the article header and thumbnail images, so
that both classes carry machine-readable marking. This is a change to the
image-generation and upload path, not to page copy, so it is filed as a
follow-up rather than applied as part of this card.

Alternative if the tooling ever supports it: C2PA content credentials
(c2pa manifest) would be the stronger form, since they are cryptographically
verifiable rather than descriptive. The current generator path does not emit
them.


5. The article notice

Every published article carries this notice at the top of the article body:

KI-Hinweis: Dieser Beitrag wurde von DadChecks Editorial KI-gestützt erstellt und von den automatisierten Prüfagenten der Redaktion geprüft. Keine Person hat den Beitrag selbst geschrieben oder selbst getestet. Es handelt sich um eine Recherche, nicht um einen Praxistest: Es gab keine eigene Testhardware, keine eigenen Messungen und keine eigene Kaufentscheidung.

(English: “AI notice: This article was produced with AI assistance by DadChecks Editorial and checked by the site’s automated validation agents. No individual person wrote or tested it. It is a research article, not a hands-on test: there was no test hardware, no own measurements and no own purchase decision.”)

The notice is currently in German because it sits alongside the German
affiliate disclosure and the German legal documents, and the site’s interface
language is English. Recommendation for the owner: switch the article notice
to English, or publish it bilingually, so that the notice is understood by the
site’s actual audience. This is recorded as an open item in the interface
language audit (§ 7), not silently changed here, because it changes published
content and therefore needs approval.


6. Legal framework — what applies, and what does not

This section was checked against the primary legal texts on 2026-11-12. It
states the duties that apply to this site and, just as importantly, the ones
that do not.

6.1 EU AI Act — Article 50 transparency obligations

Regulation (EU) 2024/1689 (the AI Act) applies from 2 August 2026
(Article 113), with the general application date now in force.

Article 50 sets out four distinct obligations. Which one applies depends on
the role and the content:

Art. 50 Duty Who it binds Applies to dadchecks.tech?
50(1) Inform people they are interacting with an AI system Providers of AI systems intended to interact directly with natural persons No. dadchecks.tech does not operate a chatbot, voice assistant, or any system a reader interacts with. This site has no chat interface, no AI assistant, and no AI-generated support reply.
50(2) Ensure outputs are marked in machine-readable format and detectable as artificially generated Providers of AI systems generating synthetic content No, as drafted. The obligation falls on the provider of the AI system — the model service used to generate the image — not on the publisher who uses that system. dadchecks.tech uses third-party AI tooling and does not itself provide an AI system to others. Where the tooling supports content credentials they are applied, but the measured state of this site’s images is set out in § 4 — and it is not uniform across image classes.
50(3) Inform people exposed to emotion recognition or biometric categorisation Deployers of emotion recognition or biometric categorisation systems No. dadchecks.tech does not deploy emotion recognition or biometric categorisation.
50(4) Disclose artificially generated or manipulated deepfakes; disclose AI-generated text published to inform the public on matters of public interest Deployers of systems generating/manipulating content Qualified — see below.

Article 50(4), second subparagraph is the provision that actually reaches
content like this site’s, and it carries an explicit exception:

“Deployers of an AI system that generates or manipulates text which is
published with the purpose of informing the public on matters of public
interest shall disclose that the text has been artificially generated or
manipulated. This obligation shall not apply where […] the AI-generated
content has undergone a process of human review or editorial control and
where a natural or legal person holds editorial responsibility for the
publication of the content.”

How this site relates to that rule. dadchecks.tech publishes product and
how-to articles. These are not “matters of public interest” in the sense the
Article means — they are editorial and commercial content, not public-interest
information such as news or public notices. On that reading, Article 50(4) does
not impose a disclosure duty on these articles at all.

But the site discloses anyway, and the basis for that is the editorial-control exception, not a general duty. Every article here is checked by the automated validation agents of the editorial pipeline and a named natural person (the owner) holds editorial responsibility for publication. So the site’s disclosure sits on the safe side of the exception, and the site does not need it to. The disclosure is a transparency
and consumer-protection commitment under German and EU unfair-commercial-practices
law (see § 6.3), not a claim that Article 50(4) compels it.

This distinction matters and is stated deliberately: this site does not claim
that the EU AI Act requires it to label every article. It says what it does, why,
and on what basis. Overstating a legal obligation is its own kind of
misleading.

6.2 AI Act — general-purpose AI model obligations

Article 53 obliges providers of general-purpose AI models (the model
developers) to, among other things, draw up and make publicly available a
summary of the training content, and to put in place a policy to comply with
Union copyright law. This is a duty on the model provider. It is not a
disclosure duty on a publisher, and this site does not represent that it is.

6.3 German law — unfair commercial practices (UWG)

Where content serves a commercial purpose — as it does for reviews, buying
guides, and affiliate links — § 5a UWG applies: withholding material
information, or providing it unclearly or in a misleading way, is unfair
competition. § 5a(4) UWG adds that a commercial purpose must be made
identifiable (kennlich machen) where it is not obvious from the circumstances.

The affiliate disclosure at the top of every article, the visible AI notice,
the visible AI labelling on shop images, and the editorial policy page itself
are the site’s response to that duty: the commercial context and the AI
involvement are disclosed clearly rather than left for the reader to infer.

The site also avoids a commercial purpose existing at all in some places:
Article 50(4)’s note that a provider receives no remuneration for a purely
editorial act is not applicable here, because the site’s articles are
published with affiliate links, so the commercial context is real and is
disclosed.

6.4 German law — provider obligations (§ 5 DDG)

§ 5 DDG requires the provider identification shown on the Impressum: name,
postal address, and information enabling rapid electronic contact. That is a
provider-imprint duty, not an AI-transparency duty. It is satisfied by the
Impressum and is not restated here.

§ 6 DDG requires, for commercial communications, that the commercial
character be clearly recognisable, and that the person on whose behalf the
communication is made be clearly identifiable. This is the legal basis for
making sponsorship and affiliate relationships identifiable, which the
Editorial Policy covers.

6.5 German law — privacy and consent (TDDDG)

Note on the citation: the provision that was formerly cited as § 25 TTDSG
now appears in the consolidated text as § 25 TDDDG
(Telekommunikation-Digitale-Dienste-Datenschutz-Gesetz). Both are checked
against the current consolidated text on gesetze-im-internet.de before use.

§ 25 TDDDG requires that storing information in, or accessing information
already stored in, a user’s end device requires the user’s consent based on
clear and comprehensive information, with the consent given in accordance with
Regulation (EU) 2016/679 (GDPR). Consent is not required where storage or
access serves solely to transmit a message over a public telecommunications
network, or where it is strictly necessary to provide a digital service the user
has expressly requested.

This is the basis for the site’s cookie-consent behaviour. The consent
mechanism itself is being reconciled with the privacy policy in a separate
change
, and this page does not assert that reconciliation as complete. What
is known: the privacy policy states that the site sets no cookies, while the
consent interface refers to cart and checkout cookies, and measurement on the
live site has since confirmed that WooCommerce order-attribution writes
tracking cookies before consent. The privacy policy, the consent interface and
the banner are being made consistent with each other and with actual
behaviour in that separate change.

The AI notice and image labels are static text and set no cookies of their
own.

6.6 United States — FTC Endorsement Guides

For content aimed at a US audience, the FTC’s Endorsement Guides (16 CFR Part
255) apply. These are administrative interpretations of § 5 of the FTC Act
(15 U.S.C. 45) — they are guidance, not a rule, but practices inconsistent with
them can lead to enforcement action.

  • § 255.0 — purpose and definitions. The Guides address the application of
    § 5 of the FTC Act to endorsements and testimonials in advertising, and set
    out the general principles the Commission uses to evaluate them. An
    “endorsement” is any advertising, marketing, or promotional message that
    consumers are likely to believe reflects the opinions, beliefs, findings, or
    experiences of a party other than the sponsoring advertiser.
  • § 255.0(f) — “clear and conspicuous”. A disclosure is “clear and
    conspicuous” when it is difficult to miss (easily noticeable) and easily
    understandable by ordinary consumers. A visual disclosure should stand out
    from accompanying text or other visual elements by its size, contrast, and
    location. And: “In any communication using an interactive electronic medium,
    such as social media or the internet, the disclosure should be unavoidable.”

    The disclosure must not be contradicted or mitigated by anything else in the
    communication.
  • § 255.1(a) — general considerations. Endorsements must reflect the
    honest opinions, findings, beliefs, or experience of the endorser, and may
    not convey any express or implied representation that would be deceptive if
    made directly by the advertiser.
  • § 255.5(a) — disclosure of material connections. When a connection
    between the endorser and the seller might materially affect the weight or
    credibility of the endorsement, and the connection is not reasonably expected
    by the audience, it must be disclosed clearly and conspicuously. Material
    connections include monetary payment and the provision of free or discounted
    products. A disclosure need not give complete details, but it must clearly
    communicate the nature of the connection sufficiently for consumers to
    evaluate its significance.
  • § 255.5(b), Example 11 — the affiliate-blog case. This is directly on
    point for a site like this one: a blogger who writes independently but
    includes affiliate links “should clearly and conspicuously disclose the
    compensation”
    , because knowledge of that compensation could affect the
    weight or credibility visitors give the reviews.
  • § 255.0(g), Example 9 — what is not adequate. A disclosure on the
    profile page of a social media account, or one that requires clicking a link
    labelled “more” to see, is not clear and conspicuous. Nor is one in small
    low-contrast text that competes with other elements, or one shown for only a
    few seconds.

How this site meets that standard. The affiliate disclosure appears at the
beginning of the article body, before the first affiliate link, in a block
visually separated from the body text and marked with role="note". The AI
notice follows directly below it, before the article body. Affiliate links are
marked in the markup with rel="nofollow noopener sponsored". The disclosure
is not behind a link, not on a separate page, and not at the end of the article
— the three placements the Guides identify as inadequate.

Two measured gaps against this standard are open. They are listed in
audit/interface-language-audit.md and audit/site-defects.md:

  1. On the 8BitDo controller article, the first affiliate link appears in the
    opening paragraph before the affiliate disclosure block. The disclosure
    follows it rather than preceding it.
  2. On the Raspberry Pi 5 and home-lab rack articles, the disclosure promises
    that links marked with * earn a commission, but none of the affiliate
    links in those two articles carry an asterisk
    . The disclosure and the
    links contradict each other.

Both are content changes to published articles and therefore need owner
approval. The affiliate disclosure wording in those articles is also German;
the English FTC-facing wording is provided in the disclosure-format document
and is ready to use.

The article notice is currently German (“KI-Hinweis: Dieser Beitrag wurde von DadChecks Editorial KI-gestützt erstellt und von den automatisierten Prüfagenten der Redaktion geprüft. Keine Person hat den Beitrag selbst geschrieben oder selbst getestet.”). For a US
audience an English notice is the safer form. Recommendation: switch the
article notice to English, or publish it bilingually.
That changes published
content, so it is filed as an open item rather than applied here.

6.7 What this policy does not claim

  • It does not claim the EU AI Act requires an AI notice on product articles.
    The basis is voluntary disclosure plus unfair-competition law (§ 6.3).
  • It does not claim Article 50(1) applies — there is no reader-facing AI
    system on this site.
  • It does not claim Article 50(2) binds this site — it binds providers of the
    AI systems used, and this site does not provide an AI system.
  • It does not claim that any AI Act obligation creates a labelling duty for
    the shop. The shop labels its images as a transparency commitment.
  • It does not cite a norm from memory. Every citation in this section was
    checked against the primary source on 2026-11-12; see the verification log.

7. Editorial responsibility

Joschka Witte (admin@dadchecks.tech) is the natural person who holds
editorial responsibility for the content published on dadchecks.tech, as
identified in the Impressum. AI is a tool in that process; the responsibility
is not delegated to a tool.

Corrections to AI-assisted content follow the same route as any other
correction: log the claim, fix it, update the shared fact registry, and publish
the correction where it was material. See Editorial Policy § 6.


8. Related