Using Generative A.I. Responsibly in Writing
The Challenge—and Opportunity
The rapidly-mutating world of Generative A.I. is changing professional practices and our daily lives in multiple ways, so much so that tracking changes or even bringing them to awareness challenges most of us. The constantly shifting landscape of A.I. in professional writing—in academic contexts, in journalism, and within organizations—calls for a principled approach to using A.I. chatbots so that editors and readers alike can count on transparency from writers: They need to be able to count on the fact that it is the writer’s own voice and agency underpinning articles, chapters, pitches, and grant proposals. Most importantly, a principled approach that honors the best of “human authorship” offers the potential for ensuring that A.I. tools are used with care and discretion.
We are very much in an emerging arena, with uncertain boundaries, for using Generative A.I. in writing. A balanced approach to using A.I. chatbots and other tools is necessary. The distinction between “A.I.-assisted” and “AI-generated” writing is not always clear. Some of this confusion surfaced in a recent report by Substack, Inc., itself, in which the platform conducted a survey of 2,000 Substack writers to elicit information about their practices, and the findings suggest a big divide. Respondents reported unease about A.I.-assisted practices, and the ongoing need for individual writers to make informed choices and retain their own agency.
This report found that the most common uses of A.I. among this group include:
Brainstorming/ideation
Writing assistance (proofreading, grammar, spelling)
Image creation (the most cited use)
Language translation
The list of tools most commonly used by Substack writers: ChatGPT: Claude, Grammerly; Gemini, and Grok.
A small number of respondents did report that they used an A.I. tool for content generation for 100 percent of their articles, but most writers surveyed reported their great concern about diminishing their own voice and agency by using A.I. tools too frequently, and are finding their way in using those tools as virtual “thought partners” or amplifiers of their own voices and creative energies, rather than as substitutes and “creative offloaders” for their voices and energies. Most of the respondents also expressed serious concerns about ethics—their own work being trained in LLM and not receiving credit, and the general concerns about AI-generated writing (which we often refer to as “AI slop”) swallowing up so much of the publishing and information landscape.
This Substack report tells us something, as a snapshot in time, of current or recent practices. Other statements about responsible A.I. use in scholarly or scientific publishing are available from major commercial or academic publishers, or smaller presses. Examples include explicit statements from Princeton University Press; and Taylor & Francis Publishing. Alternatively, some smaller and heterodox publishers, including Heresy Press and Pitchstone Books, accept only submissions of manuscripts marked by originality in authors’ voices, in fiction, memoir, poetry, and social scientific perspectives with unique authorial framings.
Even without specific A.I. guidelines for these publishers, they follow the standards articulated in the Authors’ Guild guidelines, AI Best Practices for Authors. The common threads across these highly varied publishers, both academic, commercial, and smaller niche presses, are an insistence on authorial integrity; preservation of copyright and intellectual property rights; and authors’ responsibility for ensuring attribution to other writers and creators.
The chaining of attribution into the enormously large corpora of LLM texts figures in a recent Scholarly Kitchen post by Todd Carpenter of NISO (National Information Standards Organization). He points out for researchers and authors at all levels the care to be taken in understanding distinctions among attribution, reference, citation, provenance, quote, and even source itself, when using A.I. tools in scholarly writing—and to grasp the importance of tracing claims made in articles back to their original sources. Building trust in Generative AI systems to allow such tracing is crucial for both scholarly and more informal writing often found in the Substack realm.
Some Use Cases for Independent Writers
With this brief snapshot of A.I. practices in both Substack publishing and editor-mediated or -controlled writing, we suggest that the uncertain and general guidelines about A.I. use found in the publishing world in general need to be sharpened for Heterodoxy in the Stacks. The principles for preserving human authorship are already found in the Substack itself under “Information for Writers.” We insist on transparency and clarity from all contributors, and descriptions of the specific tools, methods, prompts, and other A.I.-created information to help all readers understand how A.I. figured in the creation of a particular article.
Some specific use cases or examples may clarify what is appropriate for “AI-assisted writing” on Heterodoxy in the Stacks, and we offer them here:
“I am writing an article where I need to find obscure quotes from presentations given at conferences by LIS faculty. Is it appropriate to use Claude or ChatGPT to attempt to find those quotes?”
Yes, but acknowledgement of the tool used and the prompt developed need to included with the article
“I am collaborating with two faculty on a research project where our research project will result in a co-authored article by the three of us. One colleague has used ChatGPT, another has used Claude, and I haven’t used AI at all. How should we acknowledge the different uses or non-uses?”
This will require consulting with the editor to develop a disclosure statement, and the implications for the findings identified in the article. Best to have agreed on a common AI tool to use and how it will be used
A writer for Heterodoxy in the Stacks uses CoPilot for brainstorming and outlining purposes for an article. Should she acknowledge this limited use?
Yes; all uses of AI for this substack should be acknowledged specifically
Two writers for Heterodoxy in the Stacks co-author an article on pre-censorship in the publishing industry and use Claude to write over 60 percent of the article (i.e., the content). If they include a disclosure statement accompanying the article acknowledging the percentage of the article generated by A.I., is this acceptable?
No; an explicit acknowledgement of this large proportionality that’s AI-generated doesn’t meet this substack’s call for articles following the human authorship requirement
“I’m concerned that many of the summaries produced for my prompts in ChatGPT aren’t acknowledging unique scholarly contributions of some faculty I know of in communication studies. Should I use these summaries anyway?”
The commingling and summarization of scholarly contributions produced by LLMs often makes specific attribution very problematic. You should use your own knowledge of the topic or discipline to verify what needs to be acknowledged; an AI tool cannot be held responsible for ethics in attribution of scholarly contributions. As author, you are responsible
A writer decides to summarize a longer work–for example several chapters from their own book–using a chatbot, in order to adapt it as a HITS essay, and then edits the results
Yes, but acknowledgement of the tool used and the prompt developed need to included with the article
A writer for Heterodoxy in the Stacks published an article on changes in publishing practices, using Claude to do the following:
ideation/brainstorming; outlining (using the outline presented initially by the chatbot itself); using both paraphrased text and direct quotes from the AI summaries; and with three AI-generated images inserted into the text. The writer included a disclosure statement acknowledging only the use of Claude to “identify sources” and the “general issues” in the field. Is this acceptable?
No, definitely not. There are several issues even with this generic disclosure statement: the multiple forms of AI use; the degree to which the writer used the chatbot throughout the writing process, leaving readers uncertain about his own contributions in a piece that looks like “patchwork” writing; and the generic nature of the disclosure statement itself
These examples may or may not capture experiences of writers to various venues such as Heterodoxy in the Stacks, but they do indicate the quandaries many writers face, the choices they must make, and the temptations to forgo responsible uses of these tools. The examples also point out the possibilities for developing nuanced uses of A.I. tools in order to retain one’s own voice and agency in writing—or even amplify them. Questions writers should ask include:
What is my purpose in writing this piece, and the audience? (standard rhetorical framing questions)
What do I know about my own writing “voice” that I want to preserve even if I use tools that draw on LLMs?
What specific tools will best match my purposes? (Ithaka S + R has produced an updated inventory of A.I. tools and their best uses at: Generative AI Product Tracker - Ithaka S+R
Do I gain efficiency in using an A.I. chatbot and if so, what are the specific gains I’m making? In speed, accuracy, other benefits?
Am I missing or overlooking important contexts (methods, schools of thought, disciplinary perspectives, group identities) by using an A.I. tool?
If I write multiple articles using an A.I. chatbot, how do I guard against my own motivated reasoning in developing prompts? (Self-questioning; steel manning contrary evidence with a colleague; searching for examples of “adversarial collaborations” among disagreeing colleagues on a controversial topic);
If I write multiple articles using an A.I. chatbot, what habits of mind do I need to cultivate over time to maintain my own agency? (e.g., curiosity; intellectual humility; open-mindedness)
We encourage all future contributors to Heterodoxy in the Stacks to become familiar with our “Information for Writers” statement on A.I., and also to become more familiar with the changing A.I. practices at large in the world of independent writing. We are all learning together about this rapidly mutating world of AI in every professional arena, and learning some of the boundaries for sustaining human authorship and agency will benefit us all.
Addendum
For an illuminating discussion of one editor’s perspective on responsible uses of AI in scientific and scholarly writing—broadly applicable to all AI-assisted writing-- consider listening to the brief recording of “Integrating AI into research: Practical insights with @Avi Staiman, between Sara Falon and Avi Staiman of Wiley Publishing. ~13 minutes.