Research

Research on school and library web filtering

We study how filtering policies behave across the whole web, not on a handful of hand-picked sites. Our papers and open datasets are free to read, cite and reuse.

Research paper · October 2026

Where Over-Blocking Comes From: Simulating CIPA School Filtering Policies on 103 Million Categorised Domains

Samo Plibersek, Alpha Quantum · 12 pages · DOI 10.2139/ssrn.7638439

Schools and libraries that receive E-Rate support must filter the internet under the Children's Internet Protection Act (CIPA). For twenty years people have worried that filters also block useful pages on education, health and civic life. Until now the evidence came from small samples. This paper measures the question across 103,367,510 categorised domains: it applies five typical filtering policies, from the legal minimum to a strict district policy, and counts what each one blocks.

Read the paper (PDF) On SSRN Datasets How to cite

Key findings

0.92%of all domains are blocked by the CIPA core alone, the adult content the law actually names.
4.6%of domains are blocked by a typical district policy of 17 categories.
19.1%of domains are blocked once twelve more "instructional time" categories are added.
26×more education, health, government, news, science and civic sites are caught by the strictest policy than by the typical one.
1.24 millionof the 16.3 million domains with such a label are blocked by the strictest policy, 7.6% of them.
0.048%of those protected-label domains are touched by the CIPA core. The legal minimum barely causes over-blocking.

The strict policy also blocks 2.2% of .edu domains, 1.7% of US K-12 school domains and 4.2% of government domains. The CIPA core blocks none of them.

Not every category costs the same. The adult and circumvention categories (VPNs, proxies, anonymisers) are among the cheapest to block: they catch very few useful sites. Podcasts, forums, hate and discrimination, blogs, community sites and AI tools are the most expensive: between 25% and 60% of the sites they add to the block list carry an education, health, government, news, science or civic label.

The paper also shows how much the measuring stick matters. Adult content makes up 23% of the 1,000 most visited sites in the world, but at most 0.1% of the sites that other websites link to most. Claims about how common adult content is depend on which of the two you count.

The conclusion in one line: over-blocking comes far less from what CIPA requires than from what districts choose to add, and those choices can be tested before they reach students and patrons.

What it means for your filtering policy

The findings turn into a few practical rules for a school, a district or a library deciding what to block.

  • Start from the CIPA core. Blocking the content the law names, plus VPN and proxy sites that get around the filter, gives the most protection for the fewest wrongly blocked pages.
  • Add categories one at a time, and know their cost. Each extra category should have a reason your board could explain. Forums, blogs, podcasts and community sites block many useful pages along with the ones you want to stop.
  • Separate "harmful" from "distracting". Games, streaming and social media during lessons are a classroom question, not a CIPA question. Treat them as a policy for certain groups or times, not as a permanent block for everyone.
  • Decide AI tools by type. Blocking the whole AI tools category is one of the costliest choices in the study. Setting rules per type keeps study tools open while closing the ones you do not want.
  • Keep a fast way to correct mistakes. A short allow list, a staff unblock for adult research and a simple request process fix the pages any category rule will get wrong.
  • Check your own policy. Our free CIPA over-blocking audit lets you pick your blocked categories, check the sites your teachers use and print the result.

For the legal side, see What is CIPA?, the CIPA compliance checklist and adult unblocking in libraries. Terms used on this page are explained in our glossary.

Open datasets

All aggregate tables behind the paper are released as an open dataset under the CC BY 4.0 licence: category prevalence by popularity band, the full 57 by 57 co-occurrence matrix, the policy simulations by rule, domain class and popularity, the cost-ordered blocking curve, sensitive-category context, and country and suffix shares. The dataset contains no domain-level labels.

Dataset · IEEE DataPort · CC BY 4.0

K-12 Web Filtering Policy Outcomes: Exact Block and Over-Block Counts for 103 Million Domains

Samo Plibersek · published 2 November 2026 · DOI 10.21227/95k9-h035

The same tables, described for network and security administrators: exact counts of blocked domains and of blocked education, health and civic domains for five policies and three multi-label rules, by domain class and link popularity, with the cost-ordered blocking curve that shows the price of each category. A codebook, a citation file and checksums are included.

Open the dataset on IEEE DataPort Cite the dataset
Dataset · OSF · CC BY 4.0

K-12 Web Filtering Policy Outcomes: Exact Block and Over-Block Counts for 103 Million Domains

Samo Plibersek · published 2 November 2026 · DOI 10.17605/OSF.IO/4ZPT3

The same tables as an open OSF project. Its project wiki explains the law, the five policies, the results and every file, and is a good starting point for readers new to the data.

Open the OSF project Read the project wiki

How to cite

The paper

Plibersek, S. (2026). Where Over-Blocking Comes From: Simulating CIPA School Filtering Policies on 103 Million Categorised Domains. SSRN. https://doi.org/10.2139/ssrn.7638439
@misc{plibersek2026overblocking, author = {Plibersek, Samo}, title = {Where Over-Blocking Comes From: Simulating {CIPA} School Filtering Policies on 103 Million Categorised Domains}, publisher = {SSRN}, year = {2026}, doi = {10.2139/ssrn.7638439}, url = {https://ssrn.com/abstract=7638439} }

The dataset

Plibersek, S. (2026). CIPA School Web Filtering: Policy Simulations over 103 Million Categorised Domains [Data set]. figshare. https://doi.org/10.6084/m9.figshare.34336218
@misc{plibersek2026cipadataset, author = {Plibersek, Samo}, title = {{CIPA} School Web Filtering: Policy Simulations over 103 Million Categorised Domains}, publisher = {figshare}, year = {2026}, doi = {10.6084/m9.figshare.34336218}, note = {Dataset, CC BY 4.0} }
Plibersek, S. (2026). K-12 Web Filtering Policy Outcomes: Exact Block and Over-Block Counts for 103 Million Domains [Data set]. IEEE DataPort. https://doi.org/10.21227/95k9-h035
@misc{plibersek2026k12dataset, author = {Plibersek, Samo}, title = {K-12 Web Filtering Policy Outcomes: Exact Block and Over-Block Counts for 103 Million Domains}, publisher = {IEEE DataPort}, year = {2026}, doi = {10.21227/95k9-h035}, note = {Dataset, CC BY 4.0} }
Plibersek, S. (2026). K-12 Web Filtering Policy Outcomes: Exact Block and Over-Block Counts for 103 Million Domains [Data set]. OSF. https://doi.org/10.17605/OSF.IO/4ZPT3

Questions about the research, or a request for a table we have not published: [email protected].

Filtering that blocks what it should

Our hosted CIPA filter starts from the CIPA core and lets you add each category on purpose, school by school or branch by branch.

How the hosted filter works Build your plan