Homework submitted five minutes after it was assigned. Essays that read nothing like the student who handed them in. Teachers now grade in the shadow of tools built to do the work for their students — and the tools multiply weekly. Here is how districts draw a defensible line: block the shortcut machines, keep genuine learning aids open, and enforce that line on every school device, in every location.
Most academic-integrity incidents trace back to one of four tool types. Each behaves differently, markets itself differently, and needs its own place in your policy.
Type a prompt, receive a finished five-paragraph essay. These sites exist for one purpose and advertise it openly — some even promise "undetectable" output as a headline feature.
Students paste AI-generated or copied text and get back a reworded version engineered to slip past plagiarism scanners and AI detectors. This family exists specifically to defeat your detection layer.
Photo-scan a worksheet and the answer appears, worked steps included. Math, chemistry and physics assignments can be completed without the student reading a single problem.
Paste the assignment spec, copy the program. For intro computer science classes, where the whole point is writing the loop yourself, these tools erase the learning outcome entirely.
The instinct after the first wave of suspicious essays is to block anything with "AI" in the name. That instinct backfires. Teachers are experimenting with AI-assisted lesson planning, some curricula now require students to critique model output, and a district that flatly bans the technology tends to lose the trust of its own staff within a semester.
The workable question is narrower: which tools remove the productive struggle that an assignment was designed to create? An essay mill does. A grammar checker mostly does not. A conversational chatbot can go either way, depending on the grade level and the assignment — which is why blanket answers fail and category-level answers work.
One policy language, applied by grade band and course group — the same mechanism described in our school web filtering overview.
Strong policies share a shape: strict defaults where the integrity risk is unambiguous, deliberate openings where instruction needs them, and a paper trail for both. The categories do the heavy lifting; exceptions handle the rest.
Notice what this structure buys you when a parent or board member challenges a decision. You are not defending a judgment call about one website — you are pointing to a written rule that applies the same way to every student and every domain in the category.
Individual sites are disposable; the operators behind them register replacements constantly. Only continuous, categorized discovery keeps a block current.
Districts that skip straight to blocking end up reversing themselves publicly. This sequence produces a policy your teachers helped write and will therefore defend.
Pull two weeks of DNS or proxy logs and match them against AI-tool categories. Every district that does this finds tools nobody on staff had heard of — that list, not a vendor brochure, is your starting point.
Sort what you found into the four families above plus "legitimate aids." Resist debating individual sites; if a domain sits in the paraphraser category, the category — not the brand name — is what your policy addresses.
Younger grades get the strictest posture because no elementary assignment needs a code solver. High school warrants more nuance, where some AI use is becoming part of the curriculum itself.
When a teacher requests access to a specific tool, grant it to a course group with an expiry date rather than unblocking district-wide. The exception process is where trust with teaching staff is built or lost.
Tools change, curricula change, and last semester's exception may no longer have an owner. A one-hour review each term keeps the policy honest and gives you a dated record for auditors and the board.
A block policy is only credible if the allow side is just as intentional. These are the tool types that usually survive review — each for a reason you can state out loud at a board meeting.
Tools that mark up a student's own draft — flagging comma splices, passive voice, unclear phrasing — leave the thinking with the writer. The student still produced the sentences; the tool behaves like a patient copy editor.
Text-to-speech, dictation and translation supports are how some students access the curriculum at all. Blocking them in the name of integrity punishes exactly the learners who depend on them most.
A vetted conversational model used in class — to critique an argument, generate practice questions, or be fact-checked by students — is becoming legitimate curriculum. The keys are teacher presence, a defined subcategory, and an expiry date.
Staff accounts can carry a different policy than student accounts on the same network. Lesson-plan generators and rubric builders stay open for educators without opening anything for the students they teach.
Articles, courses and documentation about how models work are reference material, not shortcut machines. Students should be able to study the technology that is reshaping their world — and the blocklist's tool-domain focus means they can.
When a department wants to trial something new, scope it to one course group for one term and watch the logs. A contained pilot generates real evidence for the next policy review instead of another opinion war.
Homework is done at home. A filter that releases the device at the school gate protects the hours when a teacher is watching and abandons the hours when the essay mill is most tempting. Policy has to ride on the Chromebook itself — enforced by the device's managed configuration, not by which network it happens to join.
On a 1:1 campus, the school-issued device is where nearly all written work happens — which means it is also where nearly all AI-assisted shortcutting happens. If your block only applies behind the district firewall, it expires at 3 p.m. every day.
Managed Chromebooks and laptops can carry the filtering policy wherever they connect, so the essay-writer category that is blocked in the library is equally blocked on home Wi-Fi. The mechanics are covered in depth on our Chromebook & 1:1 device filtering page.
Some districts block, some scan submitted work, and some do neither and hope. The honest comparison looks like this.
| Outcome | Filtering only | Detection only | Filtering + detection + honor code |
|---|---|---|---|
| Stops the shortcut before it happens | On school devices | Acts after submission | Prevention first, verification second |
| Catches work done on personal devices | Out of reach | Reviews the artifact | Detection covers the gap |
| Survives "humanizer" laundering | Paraphrasers are blocked | What humanizers defeat | Blocking launderers protects detector |
| Teaches why integrity matters | A wall teaches nothing | Feels like a trap | Policy plus instruction, stated up front |
| Defensible in a dispute | Logged, uniform rule | Contested evidence | Multiple independent signals |
Filtering makes the dishonest path inconvenient on school infrastructure, silently, before any work is submitted and before any accusation exists. What it cannot do is judge a finished essay that arrived from a personal laptop.
Detection tools work from the opposite end — but their scores are probabilistic, they misfire on some honest writing, and the entire "humanizer" industry exists to erode them. Detection alone puts teachers in the uncomfortable position of prosecuting students with contested evidence. It works far better as a confirmation signal than as the whole strategy.
The third leg is instruction: an honor code that names AI misuse specifically, assignments redesigned so the process is visible (drafts, in-class writing, oral defenses), and honest classroom conversation about what these tools do to learning. Filtering removes easy temptation, detection verifies, pedagogy builds the internal reason not to cheat.
Many of these tools are consumer services with no student-data agreement. Every prompt a student pastes in — sometimes containing names, school details or personal struggles — leaves your governance entirely. Blocking ungoverned tools is a privacy control as much as an integrity one, a point that also supports the internet-safety-policy obligations described in our guide to CIPA.
Paraphrasing tools now advertise directly against detection software: paste flagged text, receive a version that scores as human. Each detector update is met with a wave of new rewriter domains within days. You cannot win that race with a static list — which is precisely why the paraphraser category in our blocklist is rebuilt daily from roughly 300,000 newly registered domains. Blocking the launderers is what keeps your detection investment meaningful.
Category-level reports show which AI-tool families students attempted to reach, from which grade bands, and when — a spike in essay-writer attempts the week before term papers are due tells a principal something worth acting on. That is a staff-meeting agenda item, not a discipline case.
When an integrity question does arise, a record showing the school device never touched a generation tool during the writing window is evidence in the student's favor. Districts that frame the filter this way — as protection for honest work rather than surveillance — find that students and parents accept it far more readily. These reports come from the same engine that powers all of our web filtering for schools, priced together on our pricing page.
We will show you which AI cheating tools your students already visit, what a grade-banded block-and-allow policy looks like for your district, and how it enforces itself on take-home devices.