AI & Authenticity: Principles and Guidelines for Carleton Communicators

Photography and videography should accurately, authentically, and fairly represent the people, places, and scenes at Carleton. Our audience’s expectation that content is authentic must be honored, and AI tools may not be used in ways that materially misrepresent reality, deceive viewers, or undermine the integrity of College imagery.

We strive to capture photos and footage that have a candid, realistic feel and feature genuine interactions. Content should be compelling with an authentic vibe, reflective of the energy, mood, and activity that permeates campus. We prioritize a photojournalistic style, capturing moments as they naturally occur and avoiding overly staged content. 

When content is serving a journalistic, documentary, or archival purpose (such as event photos), no generative alteration is permitted, as we want to preserve the reality of what took place. When content is serving an editorial, educational, or design purpose, some generative alterations may be permitted with disclosure and approval, as we recognize that these uses may be elevated by creative or illustrative treatments that do not function as factual records.

We generally adhere to AP photography standards: Minor adjustments to content are acceptable, including cropping, toning, and color adjustments, but these must be limited to those minimally necessary for clear and accurate reproduction and that maintain the authentic nature of the photograph or footage. The description field of a photo in our digital asset management platform, Imagen, must note when AI tools were used to manipulate the photo, such as object removal or generative expand. 

Note: Images and videos created by the Division of Communications that are downloaded from Imagen may only be altered after consultation with the Division of Communications.  

Raw Photo

(Color captured by camera)

Proper Toning

(Color accurate to what was seen)

Improper Toning

 (Excessive, inaccurate saturation)

Requires your review: 

• Image or video enhancement tools.
  • Example: Reducing excessive digital noise in a low-light photograph, upscaling the resolution of a low-quality image, or using AI tools for color grading, noise reduction, and technical post-production in videos.
  • What to consider: Technical improvements can enhance usability of low-quality photos. However, the factual reality of the image or footage must be preserved.
• Generative expand or object removal tools.
  • Example: Extending the edge of the sky or a blank wall to fit a different layout format without adding new people, buildings, or campus features or removing temporary construction equipment from a photograph.
  • What to consider: These tools can improve design flexibility. They should never be used to generate content that could mislead viewers by altering the meaning or context of the image. In particular, documentary images (such as photos of events) must remain faithful records of reality. 
• AI-assisted culling or workflow tools.
  • Example: Using AI software to identify sharp images or sort images into categories while the photographer makes the final selections. 
  • What to consider: AI can make the photo editing process more efficient. Editorial judgment should remain with humans.
• Automated rough-cut assembly tools.
  • Example: Using AI to review interview transcripts and select potential sound bites for the editor to use in drafting the video. 
  • What to consider: AI can make the video editing process more efficient. Editorial judgment should remain with humans.
• AI-assisted accessibility tools.
  • Example: Using AI to draft alt text for website images.
  • What to consider: These tools can improve accessibility workflows, but human review is necessary to ensure everything is accurate. See the Accessibility on Social Media webpage for more alt text guidelines.
• Video captioning tools.
  • Example: Captioning a video tutorial that explains how to use a campus service.
  • What to consider: All videos are required to have captions to ensure accessibility, and AI can make the process more efficient and cost-effective than using Carleton’s preferred captioning service, Rev.com. However, the captions must be reviewed, as autocaptions often fail to meet digital accessibility requirements and misspell Carleton-specific words. 

Do not use:

• Do not create AI-generated portraits.
  • Example: Generating a synthetic headshot of a faculty member who was unavailable for a photo session.
  • Reasoning: AI-generated portraits can misrepresent individuals and undermine trust in institutional communications.
• Do not create deepfake or synthetic video or audio.
  • Example: Generating content that portrays a person saying or doing something they never did, or altering content that changes what they said, did, or experienced.
  • Reasoning: AI-generated videos can misrepresent individuals and undermine trust in institutional communications.
• Do not alter the historical accuracy of campus conditions or events.
  • Example: Generating additional crowd members at a campus event; creating snowfall that was not present; adding a person to a group photo.
  • Reasoning: Altering factual content compromises the documentary value of photography, and fabricated elements create a false representation of campus life.
• Do not modify personal characteristics.
  • Example: Changing a person’s body shape, age, skin tone, visible disability, or other defining attributes.
  • Reasoning: These alterations can be deceptive, discriminatory, and inconsistent with principles of authenticity, inclusion, and respectful representation.
• To simulate events or interactions that did not occur.
  • Example: Creating an image of students collaborating in a classroom that was never photographed or staging an AI-generated athletic competition.
  • Reasoning: Institutional photography and videography should document real experiences and activities. Generative AI is not a substitute for photography, and should not be used to recreate moments that were not photographed.

AI is a tool, not a creator. These guidelines do not exist to restrict creativity or slow us down. They exist so that we can maintain unique ideas, quality design, and reflect Carleton’s visual identity in adherence to our brand platform.

Requires your review:

• AI tools for generating early-stage inspiration and visual references for project mood boards.
  • Example: A designer is developing a tree-themed Reunion logo and using AI to quickly provide several references of tree illustrations across a stylistic spectrum. They format these images into a mood board to gauge which level of detail resonates before beginning to illustrate the logo itself. 
  • What to consider: AI here is functioning as a faster, more targeted version of a Google image search and does not create the idea or produce the design wholesale. In this case, the reference illustrations should still be evaluated against Carleton’s brand platform and should be labeled as AI-generated or otherwise marked to make clear they are not intended for use in the final product. 
• Stylizing photographs or imagery using color and texture filters or collage.
  • Example: Editing a photo of a student to apply a duotone treatment to break up full-color photos, help photos feel less posed, and introduce the brand’s secondary color palette.
  • What to consider: The brand platform’s approved filters are designed to provide variety without distracting from the subject, and to still adhere to the Carleton brand. All use of filters and collage should still be evaluated against the brand platform for accuracy. 
• AI-assisted design quality assurance for accessibility.
  • Example: After completing an event flyer, a designer runs it through an AI-powered accessibility checker that flags a low-contrast ratio between the gold headline text and the cream background, and suggests it won’t meet WCAG AA standards for readers with low vision.
  • What to consider: Accessibility compliance involves a lot of rule-based checking (contrast ratios, font sizes, alt text), and AI can potentially flag major issues faster than a manual review. However, AI tools can miss contextual accessibility issues, like whether alt text is meaningfully descriptive, so human judgment is still required. 
• Generative AI for creating abstract, illustrative, or conceptual imagery that does not depict real people or places.
  • Example: A designer needs a quick background texture for a series of social media graphics. Rather than spending time in Photoshop manually building a layered gradient texture, they use an AI image tool to generate a soft, abstract gradient in Carleton Blue and Late Night. The output is brought into their design file as a background layer behind typography.
  • What to consider: A gradient texture is purely compositional and carries no likeness, no identifiable style signature, and no narrative content that could misrepresent the institution or infringe upon copyrighted illustrative work. The texture should still be evaluated against Carleton’s brand color values and the file should be documented internally as AI-assisted in case the origin ever becomes relevant. 
• AI tools for assistance designing web elements.
  • Example: Assistance updating styles for a web block,iterating new blocks for development, or troubleshooting or finding errors where something is broken. 
  • What to consider: When updating styles in Carleton’s overall web theme, when done in a vacuum (without knowledge of global styles), updated individual style sheets may not inherit other set styles or may be overly bloated.
• AI-generated photorealistic images used as placeholders in designs, clearly labeled internally and replaced before publication.
  • Example: A designer mocks up a new piece using AI-generated photos of students studying in a library to show the photographer or client how the layout will feel. This should be accompanied by a visible “PLACEHOLDER / AI-GENERATED” watermark on each image while the real photography shoot is scheduled.
  • What to consider: Placeholders serve a purpose in speeding up the design review process, and photorealistic images could be more representative of the final product than gray boxes. Strict labeling is required because AI-generated people misrepresent Carleton’s actual community, and an unlabeled placeholder could accidentally make it into publication.

Do not use:

• Do not create a design wholesale. If you’re not sure where to begin a design, consider referencing the Carleton brand platform or checking out Canva templates.
  • Example: Under deadline pressure, a staff member prompts ChatGPT to “design a brochure for Carleton’s open house” and uses the output as the final deliverable without significant modification.
  • Reasoning: AI design tools are trained on vast libraries of existing work, which creates copyright problems when outputs closely reproduce protected designs. Generative AI may be able to reference Carleton’s public-facing brand platform, but because AI also pulls from wider training, the resulting design could still use an incorrect brand voice, color system, typography, or positioning and must be reviewed. When content is recognizable as AI-generated, it can also make communications feel impersonal, rushed, or generic, reducing audience trust and perceived professionalism. 
• Do not generate designs that closely imitate the style of a specific artist or designer without attribution or permission.
  • Example: A designer prompts an AI tool to create illustrations “in the style of [a well-known living illustrator whose work Carleton has licensed before]” and uses the outputs in a publication without contacting that illustrator or noting the AI generation.
  • Reasoning: An artist’s visual style is their livelihood and professional identity. Copying that creates a legal gray area that is still being actively litigated. If Carleton admires a particular aesthetic, the right path is to hire the artist or find a licensed approach.

In Carleton communications materials, copy plays a key role in providing information and elevating our key messages while maintaining Carleton’s distinct voice. The use — or not — of AI tools should serve to match our high standards for writing and editing. These guidelines are meant to help writers and editors tune their workflows to this standard. Reference the brand platform to learn how to write like a Carl.

Requires your review:

• Completing parts of the pre-drafting process, such as brainstorming, coming up with initial interview questions, finding out what has been previously published on a given topic, referencing general facts/information, and getting pointers for where to visit for further research.
  • Example: Asking generative AI to come up with a list of titles for a major campus facilities project, to use as a source of inspiration for the eventual final name, as well as asking what titles other colleges have given to similar facilities projects on their campuses. Or, asking AI how many current fundraising campaigns use a certain phrasing in their campaign names.
  • What to consider: This serves as guidance for ideation, not final creation or review, which ensures a human will still be involved in the creative content to ensure consistency, quality, and correct facts. 
  • Note: Some studies have shown that consistently using generative AI for brainstorming can negatively impact creativity and critical thinking over time. If you decide to use generative AI for this purpose, consider utilizing techniques that still engage your brain first. 
• Synthesizing large amounts of submitted text into a shorter summary.
  • Example: Uploading a collection of disparate Alumni Award nomination letters for an alum into generative AI and having it write a single alumni bio that is under a certain amount of characters.
  • What to consider: Offloading this repetitive, time-consuming task to AI allows the writer to focus their time and energy on more wide-reaching, impactful stories and other important tasks. Any AI-generated text should still be reviewed, edited, and fact-checked by a person.
• Basic copyediting for spelling, grammar, and other small-scale edits.
  • Example: Using Grammarly to flag incorrect grammar or misspellings.
  • What to consider: This takes advantage of AI’s predictive capabilities, since English grammar and spelling have rules that the AI can follow. Grammar and spelling are also not related to the actual content of the communications material or its creative design, which should come from a human to ensure consistency and quality.
• Making editorial style decisions.
  • Example: Deciding what style of capitalization to use for a lengthy book title.
  • What to consider: Generative AI may be able to reference Carleton’s public-facing Editorial Style Guide, but because AI databases also pull from wider training and do not have access to the paywalled Chicago Manual of Style that Carleton’s guide is based upon, any resulting editorial style recommendations from AI are not fully reliable. Refer to the Division of Communications’ official written guide for review and final decisions.

Do not use:

• Do not use to write part of or a full story/article, especially if the published piece has a byline, to preserve the assumption that Carleton’s content is human-generated.
  • Example: Publishing a Carleton News story covering an academic research showcase, under the name of a staff member or student writer.
  • Reasoning: When a story has a byline that names a specific person, readers trust that that person solely wrote the piece; using AI would betray that trust. Also, AI-generated writing tends to follow specific patterns that can make communications feel impersonal, rushed, and generic, which weakens the Carleton brand and can reduce reader trust.
• Do not use for final fact-checking.
  • Example: Making sure a story about a group of alumni has all the correct class years, majors, minors, current job titles, etc.
  • Reasoning: AI predicts plausible text, it does not retrieve real facts from trusted sources. The only way to ensure total accuracy is to have a human fact-check the content.
• Do not use for official translations.
  • Example: Translating an English commencement speech into Spanish for the official online program.
  • Reasoning: AI tools are not yet fully accurate when translating languages, especially if any regional idioms or unique turns of phrase are used. Instead, hire a human translator service to ensure quality.