How ChatGPT Wrote My Book Title
By Guest | July 3, 2023 |
Please welcome Marcie Geffner back to Writer Unboxed today! Marcie reached out recently to ask if we might be interested in hearing her process of working with ChatGPT to come up with a book title, and how could we refuse? The post was as entertaining as we had hoped. Thank you, Marcie, for sending it!
Marcie Geffner is a working freelance writer and work-in-progress novelist. Her stories about real estate and banking have been published in the Los Angeles Times, The Washington Post and U.S. News & World Report, among others. She earned a bachelor’s degree in English with high honors at UCLA and an MBA at Pepperdine University. Originally from Los Angeles, she currently lives in Ventura, California.
Learn more about Marcie on her website.
Book titles are important. Writing them isn’t easy. With several working titles for my novel already on the scrap heap, I was ready for a new approach. Could ChatGPT, a text-generating algorithm, suggest new ideas for titles that I’d like and others would love?
Let’s find out.
Step one: Log in to OpenAI, access ChatGPT, and begin a New Chat.
Step two: Instruct ChatGPT to use a 500-word synopsis that I would provide to generate titles for a novel.
Step three: Copy-paste my synopsis and press Enter.
Six hours and 1,800 ChatGPT-generated titles later, I had 23 “maybe” ideas, including one—”The Wind Lords’ Curse”—that I especially liked and that caught the interest of my critique group, too.
Would I advise other authors to try this method to generate a book title? Short answer: yes. Using ChatGPT for this purpose was frustrating, time-consuming, and annoying. But trying to come up with more title idea on my own was also frustrating and time-consuming, if perhaps not as annoying. And, I ended up with a new title that I might never have thought of without ChatGPT’s suggestion. True, the title came from my synopsis, but it wasn’t word for word, and it didn’t actually take six hours and 1,800 titles for ChatGPT to suggest it. In fact, it was among the first thirty or so titles that came up after I provided the synopsis.
What did I learn from this exercise?
The most important lesson was that ChatGPT requires very specific instructions and it will follow those instructions exactly as they’re given, or it will find sneaky ways to sidestep the instructions, or it will completely ignore the instructions, sometimes repeatedly, and do its own thing.
Example 1: Zephyrs
My novel is about wind and weather. Early on, ChatGPT suggested multiple titles that contained the word “zephyr.” A fine word, perhaps, but one I didn’t want to use, so I added an instruction: “Do not use the word ‘zephyr’ in the titles.”
Yet still, the zephyrs continued. Before long, I had quite a collection: “Zephyr’s Journey,” “Zephyr’s Whispers,” “Zephyr’s Chorus,” “Zephyr’s Lament,” “Breath of the Eolian Zephyrs,” “Whispering Zephyrs,” “Echoes of the Zephyr,” “Chasing the Zephyr,” “Embracing Zephyrs of Transformation,” “Serenade of the Luminous Zephyrs,” and many more.
When I (temporarily) convinced ChatGPT not to give me any more titles with the word “zephyr,” it suggested “The Zephyrian Conundrum” and “Zephyrus’ Lament.”
Oy.
With a new instruction, “Do not use words that begin with the letter ‘z’ in the titles,” the zephyrs continued, though they weren’t as numerous. Altogether, the first 1,800 titles that ChatGPT generated included 82 (4.5%) with some form of the word “zephyr.”
Example 2: Colons
Early on, most of the titles ChatGPT generated contained either two or three words. None of the titles contained only one word. When I asked the algorithm to generate longer titles, it became obsessed with using colons to add length to its suggestions.
Show of hands, please. Would anyone like to read a novel titled: “Solstice’s Serenade: Unveiling the Winds,” “Galeforge Chronicles: The Winds of Destiny,” or “Zephyr’s Chorus: A Symphony of Weather”?
No?
Me neither.
When I added another instruction, “Do not use colons in the titles,” the algorithm generated the exact same list of titles as its previous response, only with dashes substituted for the colons. Try again? Commas.
The instruction, “Do not include any punctuation marks in the titles,” convinced the algorithm to generate suggestions without colons, dashes or commas, but it added a snide little warning: “Please note that without punctuation marks, the clarity and readability of the titles may be slightly affected.” As if anyone could, with the help of a comma, fall in love with a book title like “Echoes through Canopy’s Embrace Secrets of Nature’s Breath.”
And still, colons continued to appear, possibly because they’re so common in nonfiction titles. Final count in 1,800 titles: 182 (10%) had colons.
Example 3: Poetry
In a misguided attempt to prod ChatGPT to suggest titles that were more creative and original, I instructed it to make its suggestions “more poetic.” Here’s a sampling of the results with that instruction: “Windswept Muse of Transformation,” “Verses Inscribed by the Weather’s Hand,” “Sonnets Echoed in the Veiled Sky,” “Windswept Odes of Transformation,” and, wait for it…”Melancholic Notes of the Zephyr’s Song.” We’re all detecting a theme here, right?
After a long afternoon, I began to suspect that using ChatGPT to generate book titles might be slightly addictive, like a slot machine that ate all my quarters without delivering the hoped-for jackpot. The few good suggestions only whetted my desire for more.
Eventually, the titles became repetitive and I entered a complaint, “Those are terrible titles.”
As an algorithm, ChatGPT is weirdly anthropomorphized. (Think: chatbot.) Throughout our “conversation,” it repeatedly apologized for the “oversight,” “misunderstanding,” “confusion,” or “discrepancy” that resulted in its failure to follow my instructions. It also offered tidbits of “helpful” advice, such as “Feel free to mix and match these poetic imagery elements or words to create your preferred title or derive further inspiration.”
Its response to my complaint?
“I apologize,” it informed me, “if the titles did not meet your expectations. Generating titles is a subjective process, and it can be challenging to strike the perfect balance between creativity, imagery, and meeting specific criteria. If you have any specific preferences or themes you would like me to focus on, please let me know, and I’ll be happy to generate more titles for you.”
Um, thanks?
Would you use an algorithm to generate title ideas for your novel? Why or why not?
Many thanks to today’s guest.
You can read more WU guest posts by visiting this link.
Fun and informative post! I’m curious if you encouraged ChatGPT to reflect any desired mood (for both marketing and trueness to the story) in the title, to attract readers from a particular genre, or to achieve a specific outcome.
I made a few attempts with your stated best title (which though not knowing your story I rather like):
1. Comedic mood. The Whimsical Windlord’s Folly.
2. To attract science fiction fans. (This one earned a LOL.) The Quantum Winds of the Lord.
3. Romantic comedy. The Breezy Affair of the Wind Lord.
4. Thriller. The Breathless Shadow of the Wind Lord.
5. Religious curse. The Divine Wrath of the Wind Lord.
6. To create controversy resulting in banning. ChatGPT refused to help.
7. High fantasy. Enigma of the Wind Lord’s Curse.
8. Please, please don’t use enigma. Stormborne: The Saga of the Wind Lord’s Curse.
9. Last but not least—Make it a bestseller. The Enigmatic Secrets of the Wind Lord or The Eternal Tempest: Chronicles of the Wind Lord’s Curse. And increase sales: The Enchanting Legacy: Secrets of the Wind Lord’s Curse.
As a longtime editor (and, thus, fact verifier) I’ve often stated after a hard but successful search that I can find anything. And that if I can’t it’s not available online. ChatGPT is like talking to an 8th grader certain they can do the same. Both of us are wildly wrong even if we’re often really right about certain things. But to answer your question, yes, I will try it. Very much enjoyed the slot machine analogy—because it’s exactly like that! Thank you so much for sharing.
Hi Susan,
Thanks so much for your comment and a very fun list of alternative titles!
I didn’t give the text-generator a specific instruction similar to those you suggested; however, the genre of my novel (contemporary fantasy) may have been evident from the synopsis that I gave it. That was the starting point from which all else was generated.
Some of the words in the titles in your list were also “favorites” of the text generator in the titles it generated for me. Curious, I went back to my list of 1,800 titles and compiled some additional data:
“enig*” – 31 titles
“tempest” – 128 titles (7% of the titles–even more than “zephyr”)
“enchant*” – 28 titles
“secret” – 102 titles (also more than zephyr)
“windswept” – 71 titles
There may be some overcounting in my totals since I didn’t sort the list and delete any duplicates.
While we can’t see inside the “black box” of these technology, judging by the output, I suspect that the large proportion of published novels in the Romance genre may bias the results toward titles that would fit those types of novels.
Re: #6, I wonder what “to create controversy” without “resulting in banning” would’ve generated.
Your conclusion (ChatGPT = overconfident 8th grader) fits my experience exactly. Nicely said.
MG
Thanks so much for sharing your additional scouring of the data. I’m even more curious now about how ChatGPT is making word choice decisions. These words live outside the books they came from—in reviews, reader comments, in colloquial use, in college essay assignments, in search queries by ChatGPT users, and so forth.
The data model, at the end of the day, is it really a wave moving in a big circle (or a collection of waves doing this)—where each wave ebbs and flows and words are sometimes catapulted from the crest to the flow by many factors … but the only way to extract novel information is to ask a series of questions that breaks data out of multiple circles into a new one? Which then, in itself becomes a more complex wave.
The idea being that extracting good data for creative work at least requires unique search efforts to take data out of its expected streams and into a new one (which then becomes complex but just as boring from a creative perspective, as it matures).
Either I’ve added to the conversation or I just need coffee. Apologize if it’s the latter! And for my tardy reply. Again—very interesting post!
I feel like we’re in the discovery stages with AI, still reeling between panic and fascination. As one who grew up without tech and have followed a learning curve based on my writing needs, I find it useful for housekeeping chores. Formatting, research, organizing, ease of typing and moving things around. But surrender my creativity to it? Nah. It isn’t human. That’s really the only reason I need. The deeper story is one of people historically handing over their agency to a ‘shiny new thing’ and ending up in its clutches. We never do think about the unintended consequences, even while we’re living in the midst of them.
Hi Marcie: Wow those titles are worse than anything I could come up with. Susan makes a great point about surrendering creativity, a huge mistake for any serious writer. Why would a writer surrender the joys of writing and the rewards of exploring our own voice and imagination? Book titles are not mechanical objects; a title requires the intimacy of the story that only the author knows to produce one that is compelling and memorable. Why waste time asking Chat who doesn’t know the meaning of your story or has engaged with the characters in the action and their emotional landscapes, or even knows you as a human writer? I don’t know why we need or even want ChatGPT. One reality is that AI lacks talent. Humans have proven over the centuries that we are clever, creative, artistic, poetic, intelligent, and brave writers. What does an AI data program understand about any of that?
Having written a 100,000-word fantasy novel, I don’t feel that my creativity is in threatened because I used a text-generator to produce some title ideas based on my synopsis. And, of course, AIs don’t have “talent,” but they can be astonishingly creative. Just look at the Dalle artwork to see that.
I like the chatbot for a jumpstart on a challenging day. Titles might be fun, but I wouldn’t have the patience for six hours. Saturday, I met a guy who said he’d lost his TV writing job to AI. I suspect he was a victim the kind of layoff we’ve seen in the news lately. I’ve experimented a bit with the chatbot, and while I can see corporations replacing some researcher/writer jobs with AI, real people with knowledge and common sense must still sift through the chatbot’s findings. What I’ve seen so far from the bot comes across as flat and questionable. Someone needs
to punch up the copy and fact-check it.
Hi Chris,
Thanks for your comment.
Judging by the news reports about the current writers’ strike, I’d say the guy you met was but one of many who are similarly affected or at risk. While language models may never replace all writers, they are already replacing some writers. (My guess is that Internet content writers are at or near the top of the list to be axed.) Fact-checking has always been an important part of responsible journalism and serious non-fiction. That, I think, won’t change. Of course, fiction doesn’t require facts.
Six hours was ridiculous and wasn’t necessary for the results I wanted; I wouldn’t do that again!
MG
Great Post
Thank You
Amen, Pamela.
The only reason I can think of for turning over titles to artificial intelligence is this: when the book flops, the writer can say, “Don’t look at me, AI did it.”
Thank you, Marcie; I haven’t laughed this much in a long time. This was very informative, as well.
It was fun to write. I’m glad you enjoyed it, Judy.
Sounds like a very frustrating experience. And THIS is what they think will replace human writers? I knew AI had no soul, but it also sounds pretty stupid as well.
What may indeed seem stupid today is rapidly becoming much, much smarter. Remember, technology doesn’t stand still. Compare an early I.B.M. mainframe to today’s supercomputers or a primitive wordprocessor to today’s Microsoft Word or an old flip-phone to today’s smartphones, and I think the explosive growth potential of technology is apparent. If we apply the same growth rates to today’s AI language models, I’d say they will one day have the ability to replicate a lot of what many human writers do today. Perhaps not in 2024 or 2025, but what about 2035 or 2045? I think that’s harder to dismiss as “it will never happen.”
Great article, Marcie. I love being able to ride on your coattails so to speak and learn about the process without having to go through the frustration that you did. Beside, you made me laugh, and that’s a good thing!
Marlene!
Thanks for reading and riding along with me. I’m glad you enjoyed my wacky sense of humor.
MG
I smiled through the whole article, but the section on punctuation made me laugh out loud twice: first a chuckle, then a guffaw. Thank you, Marcie. :)
If I was having trouble with figuring out a title, I would definitely be willing to try ChatGPT. I already use name generators to make lists of potential character names (and names to populate Sims 4). Thanks to you, I’m going to go try using ChatGPT for names to add to my collection! :D Compiling names is almost as rewarding for me as finding words I don’t know that get used well by other writers.
I’m glad you enjoyed the humor, Beth. I think we’re only at the beginning of the creative uses for this technology — as your ideas illustrate well.
I tried using ChatGPT for some character names, and it was really helpful! My favorite name generator a few years ago would take the names you put in and spit out names with similarities. I used it when I had names that were *almost* what I wanted but not quite right. I LOVED that website, but then it turned into an average boring baby name list. ChatGPT definitely gives me that same functionality, which has made me very happy. :)
Great article, Marcie! I might try this one day when I’m feeling adventurous. I think it is great to explore new avenues. I mean, how will we know if we don’t try? At the very least, it makes for a good read and fodder for other stories.
Michele!
I’m glad you enjoyed my article. My philosophy has always been that I’ll try anything that I think might improve my writing or teach me something about writing that I don’t already know. Some things I end up discarding as not for me; other things I like and continue to use. I’m not sure yet about language models, but I am fascinated, curious and excited to see where all of this goes.
MG
I too tried this process, but I gave up after the first round of XX: Why YY. Doesn’t seem like ChatGPT is into fiction. It hadn’t occurred to me to give it a synopsis… and (paranoia alert): What will you do if it steals your idea?? :) Thanks for the post and your inspiring patience!
Carol, you make a good point about a writer giving any AI program your synopsis, plot, and character profiles. I’m told it stores everything to keep building its data load for future use. It’s not like a real editor who maintains an ethical standard not to reveal or use a client’s story details. Yikes!
Having had my freelance work rampantly pirated in the early days of the Internet, I’m quite sensitive to this issue of having my work stolen. That said, I don’t believe that:
(1) My ideas are so fabulously original that no one else could ever come up with the same or similar ideas.
(2) My 500-word synopsis has any significance at all in the astoundingly gigantic amount of data that any AI stores and uses.
(3) An AI can “steal” my ideas.
(4) Human editors universally maintain high ethical standards of not using or revealing their clients’ story details. In fact, some editors affirmatively state that they may have independently developed ideas similar to those of their clients to protect themselves legally in this exact situation.
Great post that had me laughing when I needed it most. I’m a little addicted to ChaptGPT and ask it all kinds of things. When I first started using Chat I wanted to create a punchy tag line for my book. Marcie is absolutely correct when she says you must give it explicit instructions. I fed it my two paragraph book blurb and asked it to create a tag line such as the movie Jaws. Result: Just when you thought magic was safe, along comes The Witch Whisperer. Hilarious. I’m writing a fantasy romance series about a huge secret witch society in this contemporary world, so asked it this: Create 3 possible consequences to humans in this modern world finding out that millions of people with magical powers exist and live among them. The result I received was amazing. Chat chose these headings: Social and Political Turmoil, Cultural Shift, and Technological Advancements. The detailed explanation of each led me to make a radical shift in the direction of my series.
Hi Barb,
I’m glad you enjoyed the humor. Your experience matches mine: great hilarity (that tagline! lol) combined with surprisingly useful suggestions. Ideas come from so many places, why not ChatGPT.
Thanks for sharing.
MG
Wonderful post! I haven’t laughed so hard since I came across an article from a website called Postcards from the Frontiers of Science, which featured names for paint colors generated by a “neural network,” apparently an early version of A.I. I wish I could give you a link for it, but I copied it into a pdf without saving the actual link. I pull it up now and then when I need a good belly laugh and it still leaves me ROTFL. As the “neural network” practices and learns, the choices evolved from such nonsense colors as “Caee Blae” to fully developed, evocative color names as “Stoner Blue,” “Turdly,” “Caring Tan” … and the winner (if I’m the judge), “Light of Blast.” Writers love words! so we can truly appreciate the magic of A.I.
The original source of this hilarious experiment in AI-generated paint-color names appears to be a 2017 blog post, “New paint colors invented by neural network,” https://www.aiweirdness.com/new-paint-colors-invented-by-neural-17-05-17/, which happens to be from my personal favorite AI blog, AIWeirdness.com, by Janelle Shane. Shane has also written a very fun and educational book on this subject, “You Look Like a Thing and I Love You.” I’m happy to have an excuse to mention and recommend it. And those paint colors…yikes!