© Honorlock 2023
Preventing AI Content Tools in Higher Ed
2
AI is more sophisticated, human, and integrated into our
lives than ever before, including in higher education.
AI helps personalize learning, provide quicker feedback,
offer exibility, increase access, and automate tasks that
give instructors more time to teach.
But some forms of AI, like ChatGPT and other tools,
can also put academic integrity at risk. They can write
anything, they’re virtually undetectable, and they’re rapidly
advancing.
So, how can you stop them?
We’ll show you.
Note: This eBook uses ChatGPT as the primary example because it’s the most
sophisticated AI content tool. Other tools, like Google’s Bard chatbot and Microsoft’s
Bing AI chatbot, are trained on ChatGPT’s response data and machine learning
model. References to ChatGPT also include newer releases.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
5
What are AI content tools?
AI content tools, sometimes referred to as AI writing tools and AI chatbots, generate
content for you based on the writing instructions and information you give them. The
best tools were trained on billions of internet resources (websites, news, research,
etc.) and use machine learning to continually improve responses.
How does ChatGPT stand out?
While some AI writing tools are repetitive and awkward, ChatGPT stands out because
of its ability to write just like a human. Its ability to be “human” is because it was
trained on one of the largest text datasets and received expert feedback from
humans that helped it learn and rene its responses.
What can it write?
• Essays of any length
• Responses to test questions
• Jokes & poems
• Music & screenplays
• Resumes & cover letters
• Code and debugging code
Did you know ChatGPT
passed these exams?
• MBA & law school exams
• Bar Exam
• GRE & SAT
• Medical school nals
• 3-part U.S. Medical Licensing Exam
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
6
How do you use ChatGPT?
It’s pretty simple: you enter a prompt, and it generates a response based on the
information and instructions provided in the prompt.
Prompt examples
Prompts can be basic, complex, or anything in between. Detailed prompt instructions
generate tailored responses, which is difcult for plagiarism tools to detect (we’ll
show you in Part 2)
Add some detail
Length
Specify the length of the response with instructions like:
• In [number] words, explain…
• In [number] paragraphs, describe…
Format
Describe the response format with with instructions like:
• Write a short paragraph that explains the steps to…
• Make a bulleted/numbered list of the steps to…
• Use short, choppy sentences to explain...
Prompt
Response
Generated
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
7
Set the tone
Give ChatGPT context about how formal or informal the response should be, if
it should be funny or serious, etc. This information can signicantly change the
response you’ll receive, as shown in the examples below.
What to include, exclude, and limit
You can provide ChatGPT with specic words and phrases to use or avoid, limit the
frequency of certain words, and include extra guidelines, like using abbreviations.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
9
Do these AI content tools just plagiarize?
These tools do not plagiarize. For example, ChatGPT was trained WITH hundreds of
billions of internet resources, but it doesn’t plagiarize FROM those resources.
Sounds confusing, but here’s an analogy:
• You learn from articles, textbooks, research, instructors, lectures, etc.
• But you don’t just repeat the information word-for word, right?
• You understand the information and you’re able to make connections between all
of it to apply in different ways.
ChatGPT basically works the same way. It makes connections between all of the
information it learned, renes responses based on feedback from humans, and
generates responses based on the information it’s given. All of this makes ChatGPT
more human and extremely difcult for plagiarism checkers to detect.
Part 2 shows you how 10 plagiarism detection tools performed against ChatGPT
when we used basic, semi-detailed, and highly detailed prompts.
Try this!
Ask ChatGPT to write like a certain celebrity
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
11
Some plagiarism checkers are pretty
sophisticated. Can they catch ChatGPT?
This section shows you:
• A brief summary of how plagiarism detection works
• Real screenshots of ChatGPT responses to prompts with varying amounts of detail
• Response plagiarism scores from 10 different checkers
How do plagiarism checkers work?
They compare written text submissions to their databases of content. Some
plagiarism checkers can only nd text that exactly matches, others can identify
paraphrasing, and a few were designed specically to detect ChatGPT.
How accurate are they?
They’re accurate in some cases, like when text is copied and pasted, slightly
paraphrased, or when responses to basic prompts in an AI tool are used.
But we’ll show you how more detailed prompts make things more difcult for
plagiarism checkers.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
12
ChatGPT vs. 10 plagiarism checkers
We tested ChatGPT against 10 plagiarism checkers.
Some of the checkers we tested were newer and specically made for AI plagiarism
detection; others were general plagiarism checkers that are browser-based and free
to use; and one is a popular browser extension that has free and paid options.
How did we do this test?
• Gave ChatGPT three prompts with varying levels of instructions
• Copied and pasted responses to each prompt into all 10 plagiarism checkers
• Recorded and combined the plagiarism scores for each prompt, which indicate
the probability that the content was written by AI
• We used the median number as the overall plagiarism score because there were
many outlier scores, especially with two of the checkers (example below)
• This median plagiarism score is located at the bottom left of each ChatGPT
screenshot in this section.
The rst scored everything as 0% plagiarism. The second scored everything as
99.9% plagiarism - even the custom paragraph below, starting with, “We made this
paragraph to show you how this plagiarism detector falls short and because we
think Honorlock proctoring is pretty cool.”
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
13
Prompt 1: Basic instructions
Basic prompts = generic responses that are easily detected
ChatGPT generated an accurate, but pretty generic response to a basic prompt,
“Dene macroeconomics.” As expected, the plagiarism checkers found that the
response was mostly generated by AI.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
15
Prompt 3: Specic instructions
More details make responses virtually undetectable
This prompt gives detailed instructions to generate a response within a certain word
count (50), to use an analogy, and not to use specic words and phrases.
The result? The tools scored the response as almost entirely human-written
If plagiarism checkers won’t really work, how do you stop ChatGPT?
Part 3 shows you practical instructional design tips and with technology that can
stop AI content writing tools from being used during assignments and exams.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
17
Ask for short-form responses tied to course materials instead
of long-form essays
Instead of assigning lengthy essays, ask students to review two case studies and
reply to specic questions requiring them to think critically and integrate their
opinions and reasoning. Doing so makes it harder to use AI content tools because
they won’t have enough context to respond effectively.
Review both case studies and respond to the following questions:
Case Study 1 - ABC Organization’s QA improvements with Six Sigma
1 Dene each component of Six Sigma’s DMAIC and how it was used by ABC Org. to
improve its QA process.
2 Do you believe using Six Sigma instead of Agile was the best choice for ABC Org.?
Explain why or why not, and provide three specic reasons to support it.
3 Provide one recommendation for ABC Org. to further improve the QA process.
Explain why it’s needed, the steps to implement it, and how to measure success.
Case Study 2 - How Agile increased product development speed for XYZ Company
1 Explain the six stages (steps) of Agile methodology by describing how XYZ
Company used them to increase product development speed.
2 Although development speed increased by 17% using Agile, provide an argument
with two reasons Six Sigma could also work for XYZ Company.
3 Provide one recommendation to help XYZ Company increase production speed
from 17% to 20%. Explain how your recommendation can increase production
speed and how to measure its success quantiably.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
18
Set a certain number of hours participants have to complete
the assignment once the begin
Instructors can create rules that specify a certain number of hours participants have
to complete the written assignment once they begin.
Make sure they aren’t asking Siri or Alexa for help
Voice assistants, like Siri and Alexa, can be used to navigate ChatGPT during the
assignment. Prevent these from being used with Smart Voice Detection that listens
for specic keywords or phrases, like “Hey Siri,” “Alexa,” and “OK Google.” It also
records and transcribes what’s said while the assignment/exam is in progress.
Prevent access to sites, applications, & browser extensions
Use a BrowserGuard to prevent students from accessing ChatGPT and other AI
writing tools, whether they’re browser-based, applications, or browser extensions.
BrowserGuard can also give students access to specic websites and applications
while blocking all others.
This is important because you may want to give students access to certain case
study PDFs, internet articles, or lectures in PowerPoint as a supplement while they’re
writing, but you want to make sure that they can’t access ChatGPT or other AI writing
tools.
Detect cell phone use & other devices
Honorlock’s online proctoring software can also detect when a test taker attempts to
access test bank content using a cell phone or other devices during the proctored
assessment activity.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
19
Disable keyboard shortcuts like copying & pasting
If a student has a general idea of what the written assignment will be about, they
could have ChatGPT write a response prior to starting the assignment/exam and
then copy and paste the response as their own.
But Honorlock’s BrowserGuard prevents students from using keyboard shortcuts, like
copying and pasting, and ags the behavior if a student attempts to use them.
to complete the written assignment once they begin.
Use the webcam to check the room & monitor behavior
Check the room for unauthorized resources (cell phones, notes, other people, etc.),
and monitor/record behavior throughout the assignment/exam.
This way, you’ll know that test takers aren’t accessing ChatGPT and other AI tools
from other devices or asking friends in the room to do it for them while they’re writing
their essay.
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
20
Honorlock’s online proctoring solution
can prevent the use of ChatGPT & other
AI content tools
What is Honorlock?
Honorlock’s online proctoring solution combines AI with human proctors. Our blended
proctoring solution can detect and prevent the use of AI content tools and other
forms of academic dishonesty during written assignments and online exams.
How can it prevent AI writing tools from being used?
During written assignments and online exams, Honorlock’s solution can prevent
the use of AI writing tools by:
• Restricting access to AI content writing websites, applications, and extensions
• Checking the room for unauthorized resources like cell phones and other devices
• Detecting words/phrases that activate voice assistants
• Recording desktop screens
• Preventing keyboard shortcuts like copying and pasting
© Honorlock 2023
Preventing AI Content Tools in Higher Ed
21
How does
Honorlock work?
Honorlock blends AI with human
proctors to prevent cheating while
improving the testing experience.
Honorlock’s online
proctoring features
From detecting cell phones and voices
during exams to preventing the use of
ChatGPT, Honorlock protects academic
integrity from all angles.
AI monitors behavior during
the assignment/exam
AI alerts a live proctor if it
detects suspicious behavior
The proctor can review the
behavior and intervene if
necessary
Admins can access in-
depth reports with time-
stamped recordings of
notable behavior
24/7/365 online proctoring
with live support for
students & faculty
Flat-rate cost per exam or
per test taker
Combines AI with
live proctors
Detects cell phone
use & other devices
Monitors & records
behavior
Locks the browser
Searches for leaked
test content
Detects phrases
that activate voice
assistants
Integrates with the
LMS & third-party
exam tools
