Citing AI Tools

Why Cite AI?

The focus of citing your work is not on whether or not you used AI, but on the transparency and quality of your AI use. Using AI tools to edit, clarify, or revise your own original work for your assignment is permitted at Quantic and Valar, but you must acknowledge that use.

The business value of using AI is not in tricking people into thinking you didn't use it, it's in articulating the details of your practice so that it can be leveraged across your team, department, or company to create a meaningful competitive advantage. Being able to share what you did and how you did it is useful, hiding it isn't.

Our Institutional Standards

At Quantic and Valar, complying with our citation standard requires a two section AI Disclosure statement, detailed below. The entire disclosure will be between 1-3 pages, will appear between the cover page and table of contents, and does not count toward your overall word count/length.

If you have questions, email library@quantic.edu.

AI Disclosure Requirements

Section 1: AI Strategy - this describes your overall approach, how you thought about incorporating AI tools into your process. this will be 1-3 paragraphs. See example below. Even for those who did not use AI at all, an AI Disclosure statement is still required (see non-AI use example below).
Section 2: Scope and Citations - this gives your reader detail about how you used AI, how it helped, and what you learned, and provides transparency with links to AI engagements.

NOTE: failure to properly disclose and cite your AI use may result in delays in grading and/or loss of points on your assignments.

Completing the Scope and Citations Section

Identify the different scopes, processes, and evaluation of your AI tool use with the Action / Evaluation / Protocol framework. You will likely have multiple scopes of work, and each scope will have the following points:

Scope: Be granular and specific. (research and source discovery, citation cleanup, etc)
Tool/Model: eg. ChatGPT 5-mini (OpenAI)
Action (What we did): how the tools were used. This develops the scope of work listed above, like 'ideation and brainstorming', and gives your reader detail about your workflow.
Evaluation (How it worked): this is your assessment of the output. What did the tools did well, what did they do poorly, what was effective or ineffective? This is where you articulate your own judgement about the value of the ai output.
Protocol (What we'd do next time): Each time you use an AI tool, you're learning how to get better. Translate your individual experience into systemic thinking: think of yourself as a director setting policy or giving guidance to a team of consultants and articulate what you would do in the future.
Reference Link: Accessible link to conversation. (https://g.co/gemini/share/847bd681fcba) Note: If your conversation links are not public, then you can copy and paste all of your prompts into an addendum document and link to that. This is obviously more work, but is an alternative when the links are not available.

See examples of scope and reference below.

Note that the link to the actual AI chat is important - it’s the difference in linking to “The New York Times” and linking to a specific article published by the New York Times. Linking to the entire archive of the newspaper is not a useful citation, linking to a specific article is.

Example

Here is an example of the AI Disclosure section that you can edit and revise. This is not an exhaustive list. There are many ways you can use AI tools and articulating what you used them for is part of this effort.

AI Disclosure:

The following sections describe how we used AI tools in our process. We take full responsibility for the content, conclusions, and any errors in this document.

AI Strategy

We wanted to use AI to respond to our ideas, helping us narrow things down and get to a starting point for deeper strategic thinking. We found we were much faster at rejecting poorly thought out, unformed ideas, and were able to quickly identify areas that held promise.

This saved substantial time, and allowed us to focus on stress testing and refining ideas to make them stronger, rather than trying to bolster bad ideas to make them better. Overall, our approach was designed to get us to a more robust, well developed report.

Scope and AI Citations:

Research and Source Discovery

Tool/Model: ChatGPT 5-mini (OpenAI)
Action (What we did): We used ai to aid in contextual research and source discovery. We started by asking who collects information on the functional foods industry in Canada, and then asked for a list of business newspapers or journals that regularly report on the topic.
Evaluation (How it worked): We discovered that quite a few 'articles' did not actually exist. This taught us to treat AI as a starting point for source discovery rather than an end point. It was far better at pointing us toward organizations and news outlets than at retrieving actual documents.
Protocol (What we'd do next time): if we were setting policy for a team, or had to do this again, we would advise the group to start by reading a few articles early in the process, to build a base of knowledge that would make identifying good and bad information easier.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Tool/Model: Gemini 1.5 Pro / Thinking Mode (Google)
Action (What we did): We used ai to help us narrow down our idea from a broad area (sustainability) to our much more specific topic - agrisolar arrays. We asked gemini to prompt us with a series of questions about what we were interested in and why, and ask us follow up questions about specific ideas.
Evaluation (How it worked): It was incredibly helpful in eliminating broad areas, helping us hone in on what we cared the most about, and fine tune our "big idea" into something manageable.
Protocol (What we'd do next time): We'd ask the ai to be a bit more critical, acting as a venture strategist or lead shareholder, and ask us questions about constraints, targeted capital expenditure, operational details, and other elements that would expose weaknesses early on. That would have allowed us to more quickly see the flaws, address holes in the plan, and develop our idea more fully faster.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Core Argument and Strategy

Tool/Model: ChatGPT 5-mini (OpenAI)
Action (What we did): We did not use AI to generate any portion of our strategy or core argument.
Evaluation (How it worked): Early in our process we experimented with this, and found that the output flattened our ideas rather than sharpened them, using extremely confident text that we did not fully back. It was generic, and provided a document that literally anyone could get with just a prompt. Writing the core of this project ourselves forced us to find out whether we actually understood what we were arguing.
Protocol (What we'd do next time): We'd skip the ai to start, and begin with a rough draft of our own ideas from the beginning, so that our work would be unique and add value. If we were commissioning this from a team, we'd start them off with an overview of multiple ai-generated solutions as a way of establishing a baseline of unacceptable obvious, generic work.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Editing and Proofreading

Tool/Model: Claude 3.5 Sonnet (Anthropic)
Action (What we did): ai was used as the final stop in our process, to clean up our writing and grammar. We asked it to review our work from three different viewpoints, that of an investor, a rival company, and a potential C-level hire. We also used Claude to create Chicago Author Date compliant references, and double checked those with Gemini.
Evaluation (How it worked): We found it to be extremely helpful in making our writing sharper, eliminating unnecessary language, and making our team speak with a cohesive voice. Each of these personas lead to insightful edits.
Protocol (What we'd do next time): We would run the multi-persona review pass much earlier, after our first draft. That could have shown us where we were repetitive, saved us some duplicate work, and given us a cleaner second draft to work from.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Risk Analysis

Tool/Model: Claude 3.5 Sonnet (Anthropic)
Action (What we did): We asked ai to evaluate our operational plan using a pre-mortem failure heuristic: “Assume it is three years from implementation and this strategy has failed completely. Give us 4-6 distinct reasons that might have caused the failure.”
Evaluation (How it worked): This really helped us see some plausible regulatory hurdles that we had not anticipated, as well as highlighting some things that might be beyond our control (timing) but that could be mitigated through a more cautious approach. The starting five failure scenarios claude gave us led us to identifying and additional two on our own.
Protocol (What we'd do next time): We'd integrate this prompt before starting the risk mitigation planning, which would be particularly important for budgeting. This is the kind of thing more senior consultants could do independently based on their experience and expertise, and this workflow could help junior teams level up faster.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Project Management and Team Workflow

Tool / Model: ChatGPT-4o (OpenAI)

Action (What we did): We used AI as a project management assistant to help us go from a broad research strategy to a structured 12-week timeline. We gave it our milestone dates, individual team member roles, and constraints, and asked it to build a draft schedule with task dependencies and buffer days.
Evaluation (How it worked): The initial schedule was visually helpful, but far too optimistic about parallel processing and time required to complete tasks. For example, it assumed two team members could conduct research while simultaneously drafting the financial model. However, using the ChatGPT to map out potential bottlenecks and dependencies helped our team establish realistic weekly deliverables and internal review checkpoints.
Protocol (What we'd do next time): The draft was a good place to start, but we should have assigned someone in the role of project manager, and used the draft as a conversational guide for an all hands meeting to get real buy-in on responsibilities, deliverable dates, and checkpoints, then fine tuned it with ai for cleanup.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Capstone Compliance

Tool / Model: Claude 3.5 Sonnet (Anthropic)
Action (What we did): Prior to turning it in, we pasted the capstone rubric alongside our final draft. We wanted Claude to review it against the stated guidelines and audit us against the grading rubric, highlighting anything we missed.
Evaluation (How it worked): It did a great job at flagging a few things we missed, for example, that our competitive analysis didn't acknowledge substitutions. This gave us a chance to flesh out that section and turn in a better project.
Protocol (What we'd do next time): Rather than running this compliance check at the end, when we just wanted to be done, we should have done this much sooner. In a real world scenario, running a report alongside the client brief early and often would make the work much more straightforward.
Reference Link: [Insert accessible link to conversation - NOT just chatgpt.org]

Not using AI?

If you did NOT use AI tools at all (other than for spelling), you would use a version of the disclosure statement like this:

AI Disclosure for Non AI use:

This was produced with no generative artificial intelligence (AI) or large language model (LLM) input. LLMs were not used in the organizing, structuring, drafting, writing, editing, or revising of this work, nor for idea generation, research, or the creation of visual artifacts (tables, graphs, images). This work is entirely the original product of the authors.

Note: whether to use AI tools or not is always your choice. But keep in mind that using AI tools is not about fooling people into thinking you didn't, it's about showing that you can use them in a smart way that adds value.