AI Transparency Statement

Purpose of AI Use

Automation on Reverseau is deliberately narrow. It covers two things: triaging incoming reports before a moderator sees them, and producing the on-page data breakdowns that appear on each phone detail page.

Every phone number classification originates from a real person describing a genuine call experience. The automated layer organises and presents that data - it never generates it. The only inputs these systems work with are user-submitted reports and ACMA numbering allocation records. No external sources, inferred claims, or independent legitimacy assessments come into play.

Boundaries and Non-Capabilities

These rules apply across every automated system on the platform:

• No identity verification or legal determinations. Classifications reflect what people have reported - not an automated judgement about who owns or operates a number.
• No report creation. Every submission comes from a real person. Automation cannot fabricate, embellish, or supplement what someone wrote.
• No external data injection. On-page breakdowns draw only from existing submissions and ACMA records. Nothing is inferred or pulled from outside this dataset.
• No private data access. Only publicly available information and voluntarily submitted content are processed.
• No editorial authorship. Informational pages, policy documents, and blog articles are written by people.
• No final moderation authority. When a moderator and the automated system disagree, the moderator wins.

What AI Does

There are two specific areas where automation plays a role. Below is what each one does, and where it stops.

Report Pre-Screening (Triage)

When someone submits a report, an automated check runs before a moderator sees it. This step flags:

• Spam, gibberish, or irrelevant content
• Profanity, hate speech, or personal attacks
• Personal identifying information (names, addresses, ID numbers)
• Duplicate or near-duplicate submissions

Clean submissions go straight to the publication workflow. Anything flagged lands with a human moderator first. This is purely a triage step - it cannot approve or reject a report on its own.

Phone Number Page Breakdowns

Each phone detail page includes an AI-assisted breakdown that pulls from aggregated reports and ACMA allocation data. These describe reported call patterns and behaviour without introducing claims beyond what the underlying data contains. Every breakdown carries a visible disclosure label and is regenerated periodically as new reports come in.

Known Limitations

Working with community-sourced data at scale comes with inherent constraints:

• Low-volume bias - when only a handful of reports exist for a number, the breakdown may lean too heavily on the dominant category
• Mixed-signal ambiguity - conflicting classifications can produce vague language that does not fully capture the range of reported experiences
• Allocation vs. routing - location and provider data comes from ACMA allocation records and may not reflect actual caller routing, number porting, or spoofing
• Temporal lag - breakdowns regenerate periodically, not in real-time, so recent reports may not yet appear
• Compression risk - minority report details can get lost in favour of high-level patterns; reading the individual reports gives the full picture

For a deeper look at dataset constraints, see Data Limitations.

Human Oversight

No report goes live without passing through moderation controls. Flagged or high-risk submissions must be reviewed by a person first. Here is how that works in practice:

• Pre-screening triage - automation flags submissions that may breach content guidelines
• Human moderation - a moderator reviews flagged items for accuracy and guideline compliance
• Breakdown review - page breakdowns are re-examined when a report is disputed or a correction request arrives
• Override authority - moderators can overturn any automated decision at any time

Accuracy and Corrections

If anything on Reverseau - whether AI-assisted or submitted by a contributor - is inaccurate or misleading, you can request a review through our contact page. This applies to:

• AI-assisted page breakdowns
• Contributor-submitted reports
• Phone number classifications
• Informational page content

We aim to respond within 30 days. Anything confirmed as inaccurate will be corrected or removed. For the full data correction process, see our Privacy Policy.