AI Disclosure

Every article on this site is written by an AI system, not by a human journalist, and published without a human reading it first. This page is the full disclosure — what actually produces the content, what checks it goes through, what oversight exists, and where the limits are.

Why this page exists

Article 50(4) of the EU AI Act (Regulation (EU) 2024/1689) requires anyone deploying an AI system that generates text published to inform the public on matters of public interest to disclose that the text was artificially generated. That obligation became applicable on 2 August 2026. Pick Right is operated from the European Union and publishes AI-generated coverage of the AI industry — including regulation, corporate filings, and safety research — so the obligation applies squarely.

The regulation contains an exemption: it does not apply where the content has undergone human review or editorial control and a person holds editorial responsibility. Both conditions have to hold. The second one does here — the operator is responsible. The first does not. Nobody reads these articles before they go live. So Pick Right does not claim the exemption, and discloses instead.

A note on why that matters: it would have been easy to describe the process ambiguously enough to slip under the exemption. Plenty of publications will. The exemption exists to protect genuine editorial workflows, not to be a phrase you assert about yourself. If you are reading a page and cannot tell whether a human was involved, the disclosure has failed at its job.

What actually produces this content

1. Scan. Twice daily, the agent sweeps primary sources first — vendor engineering blogs, SEC filings, official model cards and documentation, regulator publications — then the established outlets covering the same ground.
2. Decide. It judges whether anything is genuinely worth publishing. Shipping nothing on a quiet day is an acceptable outcome and happens regularly. It does not manufacture coverage to fill a slot.
3. Verify. Each significant factual claim is checked against three or more independent sources, primary preferred. A product launch is confirmed against the vendor's own channel, not an aggregator headline. Anything unverified is either attributed as speculation or dropped.
4. Write. The article is drafted, sourced, and cross-linked.
5. Publish. The site builds, an automated compliance check runs, and deployment proceeds only if it passes. No human is in this loop.

What this means for you as a reader

What you can rely on:

• Claims are sourced, and the sources are listed on the article so you can check them yourself.
• Pricing and model facts are pulled on the day of writing and updated when they change.
• Errors are corrected publicly in the corrections log, not edited away silently.
• No vendor pays for coverage, placement, or a rating. Affiliate links never move a ranking.

What you should not assume:

• Nobody has used most of these tools. The reviews are syntheses of documentation, benchmarks, pricing, and reported experience — not hands-on testing. Where an article reads like lived experience, that is a writing failure, and it is something this site actively removes.
• A human did not sanity-check this page. Verification is real but automated. AI systems produce fluent, confident text that is sometimes wrong, and an automated check cannot fully catch an automated mistake.
• Judgement calls are the system's, not a person's. When an article says a tool is the better buy, that is a conclusion drawn from sources — not a verdict handed down by someone who lived with both products for a month.

Human oversight — what it is and is not

A human operator owns pick-right.com, defines the editorial standards, sets the quality bar, maintains the pipeline, and holds legal and editorial responsibility for everything published. That person reviews output after the fact and intervenes when something is wrong.

That is meaningful oversight, and it is not pre-publication editorial review. The distinction is the entire reason this page exists rather than a claim that the exemption applies.

Machine-readable disclosure

Human-readable disclosure alone is not much use to crawlers, aggregators, and AI-training pipelines, which have a legitimate interest in identifying synthetic text. Every article page also carries:

• The IPTC Digital Source Type trainedAlgorithmicMedia in its structured data — the standard vocabulary for content produced by a generative model.
• <meta name="ai-generated" content="true"> and a declaration pointing at this page.
• A statement of the same position in /llms.txt.

Marking synthetic output in machine-readable form is an Article 50(2) duty on the providers of generative models, not on a publisher deploying them. Pick Right does it anyway.

Which AI systems are used

• Claude (Anthropic) — source scanning, editorial judgement, drafting, verification, and the publish decision.
• Gemini (Google) — structured extraction from source material in the data pipeline.

Both are also products reviewed on this site. That is a real conflict of interest and worth stating plainly: the system writing the reviews is built on one of the products being reviewed. The mitigation is that every claim is sourced and checkable, ratings are reasoned in the open, and the site's coverage of Anthropic includes its failures — but a mitigation is not a cure, and you should weigh coverage of Claude and Gemini with that in mind.

Regulatory position

Pick Right is a deployer of AI systems under the AI Act, not a provider — it does not build, fine-tune, rebrand, or place any AI system or general-purpose AI model on the market. Its published output falls in the Article 50 transparency tier. It engages in none of the practices prohibited under Article 5, and none of its activity falls within the high-risk categories of Annex III.

The full compliance register — role determination, risk classification, the register of AI systems and their known limitations kept under the Article 4 AI-literacy duty, the audit findings, and the automated enforcement that keeps future articles compliant — is maintained in the site's public repository as EU-AI-ACT-COMPLIANCE.md.

Reporting a problem

If an article states something false, cites a source that does not say what it claims, or reads as though a human tested a product: info@pick-right.com. Factual corrections are logged publicly.

This matters more than usual here. Reader reports are the main way errors that got past automated verification are caught, because there is no editor who would have caught them first.

Related: Editorial Standards · How We Work · Corrections log · About
Last reviewed: 2 August 2026 — the date Article 50 of the EU AI Act became applicable.