Everyone is talking about ChatGPT and similar generative artificial intelligence (AI) tools that have the potential to profoundly change our lives—and our businesses. Like other technological advances, generative AI has its enthusiasts and its naysayers. Most publishers find themselves somewhere in the middle, posing questions that have yet to be answered.

At Central Avenue Publishing, Michelle Halket worries that some publishers aren’t paying enough attention to the impact of generative AI. “Publishers should be concerned,” she says. “All of us in the creative world should be concerned. It’s going to change a lot. It is worth being knowledgeable about it and understanding how it works.”

Paul Nylander, book designer at Illustrada Design, points out how much we’ve yet to learn about this developing technological advance. “As with any automation or template type of tool, who is reviewing the results, and what are their qualifications?” he says. “In the quest to cut costs, how do we make sure we also aren’t cutting quality? Then there is the ownership/copyright question, both for the unspecified source material feeding the algorithm and the uniqueness and control of the results. And if anyone can create art, create ad copy, and create content, then what are we doing to differentiate ourselves? How [can] we be better than the robots?”

The AI Assistant

By using generative AI for “repetitive, low-level tasks,” Nylander says he is freeing up time to be more creative and provide more individualized service to his clients. “ChatGPT has been a godsend for creating first drafts for summaries, [generating] questions and topics for discussion, and writing book ad copy. Visualization tools such as NightCafé have made cover concepting, image compositing, and style alterations far more powerful.”

At Imbrifex Books, publisher Megan Edwards says she is using ChatGPT to help with copywriting. “As an example, we recently needed a new, shorter description of an existing book,” she says. “While it’s a task we’ve done hundreds of times and could easily do without the help of a bot, ChatGPT allowed us to do it in half the time by generating a condensed version of the long description we’d already created. The bot’s = output was not usable as generated, but it gave us a head start. It cut the time the task took by about half.”

Edwards has also used ChatGPT to write sales copy. “With thoughtful prompting, its output invariably provides some useful ideas,” she says. “The better the prompt, the better the result.” She has also tested ChatGPT for translations, treating it more like an assistant to a human translator than a stand-alone tool.

She has also experimented with entering text and asking for developmental editing suggestions. “The results have been mixed, largely because precise prompts affect the quality of the AI’s output,” she says.

Halket sees applications in both productivity and creativity. In addition to “some pretty decent marketing pitches” AI has created, she recently asked for help generating a 12-slide presentation on the publishing process. “As a person who’s not a creator, I have a hard time coming up with that initial idea,” she says. “But I’m very good at editing.” She took the “bones” AI gave her and fleshed them out.

One of the poets Halket publishes recently co-wrote a book with AI, available at Sounds True. “Forward-thinking human artists are recognizing the power of AI and exploring that,” she says.

For creators who want to explore the potential of generative AI, Kenney Myers at eBookFairs offers two free tools authors can try: a character-naming application and a cover analysis tool. “So far, the results are amazing,” he says. “We think these tools are very helpful to authors without disrupting the creative process.”

Morissa Schwartz of Dr. Rissy’s Writing and Marketing uses generative AI for help in brainstorming content and drafting outlines. “I work with a medical doctor who asked me to generate blog ideas through AI to promote his book,” she says. “It did surprisingly well.” Still, she notes that AI-generated content can be less than creative—or accurate. “It doesn't pick up on much humor or intricacies,” she says.

AI is good at summarizing, says Tara Alemany, co-owner of Emerald Lake Books. “With a properly worded prompt and the text of a book’s introduction or an author’s description, we’ve been able to have AI create some solid first drafts of back cover blurbs,” she says. “We’ve also created drafts of series descriptions by providing AI with the book descriptions of each book in the series.” Deeming the results to be “80% of the way there,” she finds it a viable option to “staring at a blank page, wondering how to get started.”

Tip of the Iceberg

At this point, Alemany says, publishers are seeing only a fraction of AI’s potential uses. At Emerald Lake, she is tapping Lumen5’s help to create nonfiction book trailers and ProWritingAid’s AI-enhanced editorial suggestions to working with text. She also uses ChatGPT, with “a dose of salt,” to summarize content, perform metadata and competitive research (using prompts created by KDSpy), and create social media and landing page content.

On the productivity side, she uses Magical to help with inbox management and is testing the Gmail assistant Superflows. She’s also investigating the potential of Motion’s AI scheduling tool. “It evaluates your calendar and to-do list and then lays out the best plan of attack to get more done in the time you have available,” she says. She’s also keeping an eye on AI resources like QuickWrite to help with authors’ generative work and Book Blaster to help authors create automated sales funnels. She also sees potential in HappyScribe’s video transcript software for expanding short-form content into long-form content, as well as tools like ContentFries, Repurpose.io, and Munch transforming long-form videos into short-form.

Still, Alemany points out that in its current iterations, AI isn’t especially imaginative. “If you give it a prompt today and then come back next week and give it the same prompt, the output you get will be pretty darn similar,” she says.

She looks forward to AI’s expanding usefulness in workflow automations. “Up to this point, generative AI has relied on document libraries that it reads and interprets to formulate responses,” she says. By year’s end, she says AI will also be using application programming interface (API) access to connect systems together, a development that could greatly streamline publishing operations.

“Imagine making a sale at an author event and having the checkout process automatically create a sales receipt in your bookkeeping software, add a sales transaction to your royalty reporting software, insert or update a contact record in your CRM, and trigger a personalized nurture sequence that engages your new customer,” she says. “That’s all possible to do now, but you need an hour or two plus a certain level of technical aptitude to get it all working properly.”

Alemany likens such AI applications to other tools that make publishing easier. “If you’ve ever experienced the pain of using a homegrown system for royalty reporting and then made the transition to using a tool that was actually designed to do what you need it to, that’s the kind of relief I believe AI will eventually be providing to us,” she says. “No one tool is right for every publisher. But when you find the ones that are best suited to your business, it’s a real game-changer.”

Who Owns What?

At Wherever Books, author-publisher Terry Ulick has used DeepAI, MotionLeap, and similar apps for “simple” marketing and promotions tasks as well as for prototyping book covers and illustrations. Still, he notes the inherent limitations. “Most people using them do not understand that like mapping programs for navigation, they are not unique programs or databases, but rather opportunistic front ends to large AI databases and libraries,” he says.

For this reason, Ulick avoids uploading any of his own art to generate AI images, which in at least some terms of service arrangements would allow the AI provider to add the art to its library. “This is a big issue yet to be addressed,” he says. “To date, AI has lifted copyrighted works and public domain works to build their databases. There is no governance or respect for copyrighted works.”

Schwartz shares his concern about copyright infringement. “Collaboration between AI developers, publishers, and legal professionals is essential in finding a balance between innovation and intellectual property protection,” she says.

On the output end, publishers face ownership questions that can affect their submissions process. “I am most concerned by not knowing if AI has been used in the creation of a story or manuscript,” says Geoff Habiger, managing editor at Artemesia Publishing. “Right now, the technology is new enough that detecting AI-written stories is pretty easy because there are so many errors. But that is going to change as the technology rapidly improves.”

Alemany says it’s a matter of figuring out how much help is too much for copyright purposes. “Graphic designers have been using stock images to create composites for years,” she says. “Artists collect reference materials to inspire and guide them as they create something uniquely their own. With AI, we have the question and concern of where it is getting its visual references from, how many of them are being used, and how closely the new image represents the original references. But we’d have similar questions of a human artist as well.”

The US Copyright Office has issued preliminary guidance on AI that will continue to be refined, says Keith Kupferschmid, CEO of the Copyright Alliance. For more of his thoughts on the copyright question, see the sidebar, “Copyright Concerns.”

Questions of Quality

“Beep, beep, beeeeep,” says Nylander. “The garbage truck is backing up, about to dump a ton of trash at our feet.” As AI-created books begin to flood an already crowded market, he says publishers need to “embrace our inner artists” and figure out how to rise above the mediocrity.

Habinger shares this concern. “With generative AI, a person can write a novel with a few keywords and then push it out to the world in less time than it takes for even the fastest writers to write content. These people won't care if it’s good writing or not.”

Alemany expects a trajectory much like what has happened with self-publishing. “As time went on, author-publishers sought to educate themselves and embraced industry standards in their work, resulting in some fantastic author-published books,” she says. “As AI-generated works start hitting the market, we’ll go through the same growing pains.”
She expects reader reviews will call out low-quality AI-generated books. “The more intriguing question,” she says, “is what we’ll think when AI-generated works become more popular than those we spend years slaving to write and publish.”

Yet as Ulick points out, AI-generated work is inherently derivative. “Writing is abstraction,” he says. “Machines are absolute even when seeming to create styles. The very fact you have to provide a prompt to generate an image or text means the machine isn’t creating anything. It has no life experience or imagination to create the prompt.”

With AI-generated content, Edwards has concerns about plagiarism. “Anyone who has asked ChatGPT to write a recipe for mac and cheese in the style of the King James Bible has enjoyed the startlingly accurate—and funny—results,” she says. “But let’s say someone succeeds in generating a book that appears to be authored by a prolific author like J. K. Rowling or George R. R. Martin. While original in the sense that it is not a replica of an existing work, such a book could obviously cause damage to the real author and publisher.”

And as Edwards notes, AI is only as accurate as the content it has been fed. “In its present stage of development, ChatGPT cannot be trusted to generate content free of bias, error, and outright misinformation,” she says. “We tested it with copy from a manuscript about fly-fishing. The results included plausible-sounding ‘facts’ about a dam rupture on a river in Texas. Knowing this was false, we challenged the bot about its statements. Its reply? ‘I may have been wrong about that.’”

Alemany notes that the risk of inaccuracy increases when AI is prompted to think for itself. “It scans its document library, made up of pre-2022 internet content, for answers from sources that may or may not be correct, and then responds with information that is highly likely to be inaccurate or incomplete,” she says.

Best Practices for a Changing Landscape

There’s little doubt that AI will change the publishing landscape—and change it quickly. “We are at the very beginning of a rapidly evolving scenario that’s going to challenge practices and assumptions we’ve considered solid for generations,” Edwards says. “Publishers must find practices they can live with in this fluid environment. Nimble response to change is the skill we’re trying to nurture.”

At Artemisia, Habinger has decided for now that he won’t accept AI-generated writing or art. In fact, he’d like to see the publishing industry “take a hard stance against generative AI.” But with AI integrated into more applications every day, such a line is becoming harder and harder to draw.

Ulick has added a contract clause for authors to affirm that all work is original and created without machine assistance. But he says he would consider licensing original work for machine learning at the right price and with a contract for attribution. He also wants to see publishers alert consumers when work is machine-generated.

At Imbrifex, Edwards makes sure all AI-generated material is fact-checked for accuracy and bias. “Generative AI is powerful and fast, but its output isn’t ready for prime time until a competent human has vetted it,” she says.

She is keeping an eye on the use of AI for content recommendations based on personal preferences. “If generative AI can predict accurately what will interest specific consumers, publishers can use it to improve results in PR, marketing, and advertising,” she says.

Similarly, Alemany is watching Amazon’s growth in the AI space. In recent job postings, the mega-retailer says it is looking for people who can “reimagine Search with an interactive conversational experience.”
“We certainly don’t have to evaluate every tool on the market,” she says. “But probing new AI tools with a curious and questioning mind may help us come up with ideas for our books and business that we’ve never thought about before.”

What Comes Next

on publishers will be much like the effect the internet had on newspapers. “We will see profound changes that will include the demise of some companies and professions and the birth of others,” she says. “We’ll see new laws and regulations emerge. We’ll see some metamorphosis, too, as publishers new and established incorporate new technologies in innovative ways.”

As Halket points out, AI will probably eliminate or greatly change some jobs, including copywriting and proofreading. And with the rapid pace of change, she suggests we add the word “yet” to our discussions of generative AI.

With eBookFair’s emphasis on technology, Myers is mostly optimistic about the future of AI and publishing. “It’s a bit like having access to the smartest person you know, 24/7. That could be very useful.” But he also acknowledges the challenges in policing AI-generated work and protecting the intellectual property rights of authors and publishers. “We are in for a long, drawn-out battle for sure,” he says.

Alemany expects AI will become more useful as authors train it using their own voices and as their prompts become more nuanced. At the same time, she notes that advances in AI-assisted audio narration and translation have the potential to eventually replace human work.

How readers will adjust to these market changes is anyone’s guess. “We already live in a world where there often seems to be too much noise and not enough downtime,” she says. “How will readers react to even more options to choose from? Or will we need to consider more personalized, customized content to attract their attention?”

Even with these concerns, Alemany predicts that business owners who make the best use AI will be “much better off a year from now” than those who ignore it. “The Age of AI has begun,” she says. “It’s just a matter of figuring out how AI can serve your unique circumstances.”

And while AI may play an ever-increasing role in content creation, marketing, and personalization, Schwartz thinks the human touch will remain “indispensable.” Ulick agrees. In fact, owing to AI’s derivative nature, he predicts the value of fully original material will increase.

“AI is craft,” he says. “Original work is art. Until a computer can feel the loss of a loved one or experience a traumatic event, it will only be able to use a human’s block of text or image of such things. I think that as true artists and authors learn that they go far beyond AI content, they will create even better works to distance themselves from AI.”

No Going Back

The AI genie is out of the bottle. There’s no going back. But as Nylander points out, market upsets like AI can be opportunities for society to determine what really matters. That goes for publishing, too.
The industry is likely in for a bumpy ride, Edwards says. Yet she remains optimistic, pointing out that while the internet caused major disruptions, two generations have now grown up with the web as a fact of life.

Before long, we’ll likely be saying the same about AI. “As resistant and fearful as humans can be of change, they are also amazingly able to make the most of it,” she says.

Copyright Concerns

We asked Keith Kupferschmid, CEO of the Copyright Alliance, for his thoughts on questions of copyright and generative artificial intelligence (AI). Here are his responses as of May 2023:

Which copyright issues related to generative AI most concern the Copyright Alliance at present? How is the Alliance addressing these issues?

The generative AI issue that is most concerning to Copyright Alliance members at this time is the unauthorized ingestion of copyright protected materials into AI systems. We are working on a number of measures to address the issue, including gathering information on what various AI tools ingest for training purposes and learning how datasets are compiled and subsequently ingested into AI systems. We are talking with other stakeholders, especially AI developers, about abiding by copyright law. We are also working to educate AI stakeholders, lawmakers, and the public on the value of copyrighted materials being used by AI tool developers and the copyright law implications surrounding their ingestion. Foremost among these efforts is explaining that use of copyrighted materials for training

AI systems involves copying of expressive (and protected) elements of a work, that training AI systems on copyrighted material does not categorically qualify as fair use, and that copyright owners already offer licenses for use of their works to use as AI training materials.

What measures can authors and publishers take to prevent data laundering by AI developers?

Licensing models for AI ingestion already exist for many types of copyrighted works, including literary works and works of visual art, which AI developers recognize as valuable training material for large language models (LLMs) and image generators. While some AI companies are transparent in that they only use licensed material to train their systems, others have used works scraped from the internet without authorization and ignored available licenses. That said, recent comments by leading AI companies have acknowledged the copyright issues surrounding unauthorized ingestion and promised to work to respect the law and properly compensate creators and copyright owners. Time will tell whether these companies are true to their word. In the meantime, authors and publishers should try to determine whether works they own are available online and whether they have been scraped for AI ingestion. While it may be difficult to tell whether a certain work has been ingested for training purposes, including copyright management information (CMI) in the form of metadata or watermarks is a good first step to help track any unauthorized use.

Additionally, if a copyrighted work is being licensed for AI training purposes, it is good practice to ensure the license has clear terms that can help prevent unauthorized downstream uses that could result in data laundering. Another action that authors and publishers can take is to have an AI license and to make clear that they offer such licenses—and note when and where they are willing to do so. And where they are unwilling to let AI developers train on their copyrighted works, they should make that clear wherever appropriate, such as on the copyrighted work and website and other materials associated with the works.

What sorts of transparency best practices does the Alliance recommend publishers adopt?

An important place where transparency comes into play for publishers is with the creation of their works. Where works are created using AI in whole or part, that should be indicated on the work itself.

How do you expect the question of authorship to continue to be sorted out following the US Copyright Office’s recent Copyright Registration Guidance directive on using material in AI-assisted work?

The US Copyright Office’s guidance document is a good first step, but the AI copyrightability and authorship issue are complex, and there is still much that needs to be examined and re-examined. The Office’s initial guidance caused concerns from publishers and many others in the creative community. Since then, the Office has explained that the guidance provided is just the first step and that it will continue to hone the information as more is learned about AI and how the creative community is using it. It is important to understand that we are at the very nascent stages of generative AI. As we all better understand how AI is being used, we will have a clearer picture on where and how to draw the line between non-copyrightable and copyrightable creations.

In its position paper, the Alliance calls for education in ethical and lawful practices related to artificial intelligence. To what extent is this education happening? How might it be facilitated on a broader basis?

The Copyright Alliance is working to educate AI stakeholders, lawmakers, and the public on the value of copyrighted materials being used by AI tool developers and the copyright law implications surrounding their ingestion. Foremost among these efforts is explaining that use of copyrighted materials for training AI involves copying of expressive—and protected—elements of a work, that training AI systems on copyrighted material does not categorically qualify as fair use, and that copyright owners already offer licenses for use of their works as AI training material.

What is the best way for publishers to stay informed about questions involving AI and copyright?

We have a page on our website that is devoted to AI and copyright at copyrightalliance.org/trending-topics/ artificial-intelligence-copyright. We are continually updating the page, so we encourage people who want to learn more to check out the page and then come back often. We also have an AI and copyright alert that you may subscribe to for the latest news as it happens: copyrightalliance.org/get-involved/ai-copyright-alert.

---