79% of NSW government schools operate public Facebook Pages

Parents consented to sharing. Not to AI training.

Parents at NSW government schools sign a "Permission to Publish" form. It asks whether the school can share photos and work samples of their child on the school website, in newsletters, and on social media. Most parents say yes. They want to see their children celebrated.

That form does not mention AI training. It does not mention facial recognition databases. It does not mention deepfakes. It was written before any of those risks existed. Nobody has updated it for a world where Meta scrapes every public Facebook post to train its AI models, where Clearview AI harvests public photos for facial recognition, and where public photos are available to anyone building an AI training dataset.

The gap is not between parents and schools. It is between what the consent framework covers and what actually happens to publicly accessible photos in 2026.

1,713 schools. 596,069 students. One pattern.

This audit covers 2,160 of the 2,210 NSW government schools listed in the NSW Department of Education's public schools master dataset. For each school we searched Google for site:facebook.com "[School Name]", then categorised every result by match quality. In August 2026 we went further: for every page the search called a school's own, we checked the name of the account that actually posts on it. A page only counts below if that name is the school's.

This page reports the verified figure: 1,713 schools, 79% of the 2,160 searched. Another 223 pages appear in search results but showed no public posts to check, and 78 more are flagged because the posting account carries a different name. Both groups sit outside the headline figure until a manual review settles them. The primary stat understates rather than overstates the finding.

These are not relics. Of the 1,713 verified Pages, 1,613 (94%) had posted within 30 days of the verification pass on 13 August 2026. A further 45 had posted within six months. Only 55 were dormant.

The pattern holds for every school type. Schools for Specific Purposes are not spared.

81 Schools for Specific Purposes operate verified public Facebook Pages. These schools serve 4,611 students with disabilities and additional needs. Their photos are publicly accessible under the same consent framework that does not mention AI.

77% metro. 82% regional. 83% outer regional. The pattern is statewide.

This is not a city problem or a country problem. The pattern holds across every remoteness classification in NSW.

The consistency is the point. When 79% of schools follow the same pattern regardless of where they are, the cause is not individual decisions. It is the system working as designed.

Top 10 local government areas by student count

All 93 NSW state electorates. Not one is unaffected.

We matched each verified school to its current NSW state electorate, using the boundaries in force since the 2023 redistribution and the electorate recorded for each school in the NSW Public Schools Master Dataset. All 1,713 of the 1,713 verified schools matched.

Every one of the 93 seats in the NSW Legislative Assembly contains government schools operating public Facebook Pages. This matters because the policy is a state departmental instruction, so the accountable representatives are state members, not federal ones. The table shows the ten electorates with the most students at those schools.

Top 10 of 93 by student count. Londonderry, ninth on this list, is the electorate of the Deputy Premier and Minister for Education and Early Learning. Student counts are fractional FTE enrolment rounded per electorate, so they may not sum exactly to the statewide total of 596,069. Members and electorate office addresses from parliament.nsw.gov.au, retrieved 4 August 2026 and cross-checked against a second source.

These are the Pages that Meta confirmed it scrapes for AI training

A public Facebook Page is not just visible to parents. It is visible to everyone on the internet. Search engines index it. Automated systems that collect data at scale can reach it.

The evidence for what happens to content on public Facebook Pages is documented across this site:

• Meta AI training. Meta confirmed under oath to the Australian Senate that it scrapes every public post since 2007. That includes children's photos on school Pages. Australians get no opt-out.
• Facial recognition. Clearview AI scraped 50 billion+ photos from Facebook for facial recognition. The Australian Information Commissioner found it breached the Privacy Act. The database grew after the deletion order.
• Training datasets. Human Rights Watch found 362 identifiable Australian children in a single AI training dataset, from less than 0.0001% of the data. Some photos came from school uploads.
• Deepfakes. Students generated explicit AI images of 50+ schoolgirls at a Melbourne school from publicly accessible photos. Deepfake reports to the eSafety Commissioner are doubling year on year.

1,713 NSW government schools operate the kind of public Facebook Pages that these systems draw from. 596,069 students attend these schools. The consent forms their parents signed do not mention any of these uses.

Schools are not at fault. Nobody has updated the framework for AI.

NSW Department of Education policy PD-2011-0418 governs how schools use social media. The policy states that school Facebook accounts "must not restrict access or be set as 'private' or 'closed'." Schools are following the rules.

The department wrote the policy in 2011. Before Meta confirmed it uses public posts for AI training (2024). Before Clearview AI built a 50 billion photo facial recognition database from Facebook. Before 362 Australian children were found in an AI training dataset (2024). Before students used AI to generate explicit images of schoolgirls from their publicly accessible photos (2024).

The result: 79% of schools follow the same pattern. Public Pages, not private Groups. This is not individual decisions. It is what the system produces.

Approximately 100 NSW schools already use private Groups

These schools communicate with their communities through private, member-approved channels. Parents request to join. Content is visible only to members. Search engines do not index it. Third-party scrapers cannot access it.

The transition from a public Page to a private Group is free. It takes less than an hour. No technical barrier prevents any school from making the same change.

Write to your MP →

How this was done, so anyone can verify it

Data source

NSW government school master dataset from Data.NSW. Published by the NSW Department of Education. 2,210 schools total. 2,160 processed (50 excluded by batch sizing, randomly distributed across school types, does not affect proportional findings). Dataset dated 15 April 2026.

Search method (phase 1, April 2026)

For each school we searched Google for site:facebook.com "[School Name]" using the Apify Google Search Results Scraper, in 44 batches of 50 schools. Top 5 Google results returned per query. Facebook page URLs extracted from results and classified as Page or Group.

QA methodology (phase 1)

Every result was programmatically categorised by:

1. Name overlap scoring. Distinctive words from the school name checked against the Facebook page title. Common words excluded (public, school, high, nsw). Zero overlap = false positive. Under 50% = weak match.
2. P&C detection. Title containing "P&C", "P and C", "Parents and Citizens", or "P & C".
3. Alumni detection. Title containing year ranges, "alumni", "old boys/girls", "reunion", "past students".
4. URL pattern analysis. facebook.com/p/ and numeric IDs flagged as auto-generated (Facebook creates these from location data; school may not manage them).

A stratified random sample of 30 results was manually reviewed against Google search results. 90% confirmed correct matches. 7% P&C or community pages. 3% false positives.

Verification method (phase 2, August 2026)

Checking a search result's title is not the same thing as checking a page. A result title can carry a school's name while the page belongs to someone else who mentioned the school. Two such cases surfaced in a manual spot-check, so we re-verified every one of the 2,017 pages the search had categorised as a school's own: we retrieved each page's two most recent public posts and compared the name of the account that published them against the school's name. That checks who runs the page, not what a search snippet says.

• Verified (1,713 schools). The posting account's name matches the school's. These are the only pages in the headline figure.
• Flagged (78 schools). The posting account carries a completely different name. Spot-checks found most are pages that merely mention the school, such as a local MP, a community group, or a news page. At least one is a real school page still running under a pre-merger name. Every one is under manual review and asserted neither way until then.
• Ambiguous (3 schools). Partial name overlap. Under manual review.
• No public posts (223 schools). The page exists in search results but returned no posts to check, so ownership is unconfirmed and the school sits outside the headline figure.

The 143 pages phase 1 had already excluded as P&C, alumni, false positive or weak matches were not re-checked. That categorisation stands.

Limitations

• Google's index may miss some Facebook pages. These figures are a floor estimate.
• The 223 pages with no public posts could not be verified either way and sit outside the headline figure. Some will be genuine school pages, which understates the finding.
• The 78 flagged pages are under manual review. Where one proves to be a real school page under a former name, the headline figure rises.
• Post sampling shows how recently a page posts, not what it posts. Sharing school news and events, which inherently involves student content, remains the stated purpose of most school Pages.
• 50 schools (of 2,210) were not processed due to batch sizing.

What this audit did and did not access

Phase 1, the search (April 2026): Google's public index only. No Facebook page was visited. The search query site:facebook.com returns only pages that are publicly indexed, confirming public accessibility by definition.

Phase 2, the verification (August 2026): the two most recent public posts on each page the search had called a school's own. What we read: the name of the account that published them, the post dates, and the caption text. That is what it takes to confirm who runs a page and whether it is active, and it is all we took.

Constant across both phases, no exceptions: no photo of any child was viewed, downloaded or stored. No student's name, image or personal data was accessed at any point. What this research verifies is who owns each account and how often it posts. Never content about any child.

This is a deliberate constraint. The research demonstrates that these Pages are publicly accessible without anyone needing to open a single photo. The same Google search that produced these results is available to anyone, including the automated systems documented in the evidence pages.

The audit is fully reproducible

The school dataset is publicly available from Data.NSW. The methodology above sets out every step. Anyone can run the same Google searches, and anyone can open any page on the list and read for themselves who posts on it.

This audit will be re-run quarterly to track whether schools move from public Pages to private Groups over time.