Skip to main content
Home/Legal/AI Transparency
Last Updated: 2026-11-16

AI Transparency

About This Page

This page explains where AuditSocials uses AI, how AI-generated text is labelled and checked, and who is responsible for it. It is our disclosure under Article 50 of the EU AI Act (Regulation (EU) 2024/1689). AI drafts and checks content on AuditSocials; it does not make decisions about people.

AuditSocials is a product of Reaktör Teknoloji Tic. Ltd. Şti. (trading as Reactll). AuditSocials is a Signatory to the EU Code of Practice on Transparency of AI-generated Content (Section 2) — see the European Commission's list of signatories. Signing is voluntary and is not a certification or endorsement by the European Commission.

1. Policy Change Summaries

When a platform policy page changes, an AI model drafts the title, summary, severity, action steps and industry notes from the before/after text. The quotes the model relies on are machine-checked against the actual page text, and an editor reviews every change before it is published. These summaries are labelled AI · Summary · Editor-reviewed.

2. Regulator Signals

An AI model summarises public regulator notices and filings. A signal is published automatically only if it passes automated checks: the source is in our highest source tier, the platform involved appears verbatim in the source, and a separate AI check reviews the summary against the source. There is no human review before publication, so these are labelled AI · Summary · Auto-checked against source.

3. Compliance API, MCP & Free Checker

Content you submit is checked by deterministic rule matching first, then by an AI layer that looks for risks keywords cannot catch. Every finding states its origin: source rule or source ai. Findings from the AI layer contain AI-generated explanations and suggestions and are labelled AI · Generated in our free checker.

4. Dashboard Briefs

Logged-in users may see a short personalised brief generated by an AI model from the policy and enforcement data we have already published, filtered to the industries and platforms they selected. Briefs are labelled AI · Generated.

5. Articles, Guides and Glossary

Our blog articles, platform and industry guides, country pages, rejection guides and glossary are drafted with AI assistance and then edited by the AuditSocials team. We do not claim that every article has been individually fact-checked before publication, so we label this content openly instead: an AI · Drafted with AI assistance label next to the headline, and a note at the end of each article saying who is responsible for it. Content published before 2 August 2026 is labelled too.

6. One-Time Reports

Our one-time reports — the content audit, rule plan, rule map and data request — are generated by our AI system from AuditSocials data and the material you send us, and reviewed by our team before they are sent to you. Each report says so in its footer. Reports link every rule to its official source; they are not legal advice and do not guarantee how a platform will act.

7. Models We Use

  • Primary: OpenAI GPT-4o mini via the OpenAI API (since November 2026).
  • Alternate: Anthropic Claude models via the Anthropic API, which we may switch back to. Our one-time reports are prepared with Anthropic Claude.
  • Fallback: if the primary provider is unavailable, most AI features switch to an open-weight model (gpt-oss) served through NVIDIA's API.
  • We do not train our own AI models. We use general-purpose models from these providers.

All three providers are listed as sub-processors on our Security & Trust page.

8. How We Label AI-Generated Text

  • On web pages and in the dashboard: AI-generated text is shown with an "AI" label next to it, visible without any click, and the element holding it carries the machine-readable attribute data-ai-generated="true".
  • In the Compliance API and MCP: each finding has source: "rule" | "ai", and responses include aiGenerated and aiDisclosure. Policy-change and regulator-signal tools include the same two fields.
  • In emails and PDF reports: AI-generated text is tagged "AI" (or "[AI]" in PDFs) with a disclosure line, and PDF reports also state it in their document properties.

9. Editorial Responsibility

  • Responsible: Reaktör Teknoloji Tic. Ltd. Şti. (trading as Reactll), the company behind AuditSocials — AuditSocials editorial team.
  • Address: Serdar-ı Ekrem Sok. İş Yeri, No: 15/A, 34420 Beyoğlu/İstanbul, Türkiye. Istanbul Trade Registry no. 298746-5.
  • Contact: info@auditsocials.com

How review works. Policy changes are published only through an editor review queue: the editor reads each AI-drafted change against the platform's own policy page before publishing it. Automated checks (verbatim-quote matching and a separate AI check) support this review but do not replace it. Regulator signals are not human-reviewed and are labelled accordingly. Articles and guides are edited by the team and labelled as AI-assisted (section 5).

10. What We Don't Use AI For

  • No chatbot. You are not talking to an AI anywhere on this site — the order assistant on our solutions pages is a scripted form, not an AI.
  • No scoring, profiling or decisions about individual people.
  • No biometric identification, biometric categorisation or emotion recognition.

11. Limits, Reporting Errors and Mislabelling

AI can be wrong. Policy changes and regulator signals link to their source where one is available — check it before you rely on the summary. Nothing on AuditSocials is legal advice, and a clean compliance check does not guarantee a platform will approve your content. Found an error in AI-generated content, or content that is labelled wrongly (or not labelled)? Email info@auditsocials.com — we review every substantiated report and correct it without undue delay.