Writing Book Descriptions with Artificial Intelligence

The cover for a book titled Race Scoring: How to Determine the Winner of your Next Individual, Series, or Team Running Event. It features a colorful illustrated group of runners in a race and was written by Blaine Moore.

We have recently spent some time testing different A.I. models to see how well they stack up against one another for creating a rough draft of our book descriptions. It stands to reason that some tools will work better at the task than others, while different tools will charge varying amounts for the output they create.

For our controlled test, we used the same prompt in each tool to write our book description. (Well; for the most part. Some of the tools do not have the vision capability, so for ones where we could not upload the cover image, we did adjust the language slightly to account for that.)

We also uploaded the complete manuscript for our book to the models that have a large enough context window to ingest the entire book. The entire book is just over 13,000 words or just over 72,000 characters, and is 19,000-20,000 tokens depending on which model we are referencing.

Here was our prompt (which you can download at the bottom of this article):

Title: Race Scoring: How to Determine the Winner of Your Next Individual, Series, or Team Running Event
Author: Blaine Moore
Genre: Non-Fiction (Sports)
Target Audience: Race Organizers
Book Manuscript: RaceScoring.pdf (attached)
Cover Image: FinishLineCoverMockup-arty.jpg (attached)

Using the provided book manuscript and cover image, please generate a compelling and optimized book description that includes the following key elements:

  • Brief summary of the book's plot, characters, and central themes (approximately 100-150 words)
  • Unique selling points and key benefits for the target audience
  • Engaging and persuasive language to capture the reader's interest
  • Inclusion of relevant keywords for online discoverability
  • Alignment with the book's overall branding and messaging
  • A call to action or next steps for the reader (e.g., buy now, download, etc.)
  • The book description should be approximately 150-250 words in length and written in a clear, concise, and compelling style. Please ensure the description accurately represents the content and tone of the book manuscript.

Once I had generated a dozen book descriptions between 10 different models, I also used the Claude 3 Haiku model to analyze each description and provide feedback for each one. Here was the prompt I used, along with the criteria that it judged each one on (which you can download at the bottom of this article):

You are a book marketing expert, specializing in crafting and analyzing sales descriptions for books. Please use the following objective and subjective criteria to evaluate the sales descriptions for a book.

Objective Data:

  1. Length of the description (in words or characters)
  2. Inclusion of key information (e.g., book title, author, genre, target audience, unique selling points)
  3. Use of persuasive language and marketing techniques (e.g., emotional appeals, social proof, calls to action)
  4. Readability and clarity of the writing
  5. Formatting and visual elements (e.g., use of headings, bullet points, images)
  6. Keyword optimization (for online sales platforms)
  7. Consistency with the book's overall branding and messaging

Subjective Data:

  1. Engagement and interest level generated by the description
  2. Perceived trustworthiness and credibility of the description
  3. Alignment with the target audience's needs and desires
  4. Uniqueness and differentiation from similar books in the market
  5. Emotional impact and ability to create a connection with the reader
  6. Perceived value and benefits offered to the reader
  7. Overall impression and likelihood of the description to convert potential buyers

By evaluating each book description based on these objective and subjective criteria, we can gain valuable insights into how the descriptions are likely to perform in the market. This information can then be used to refine and optimize the sales descriptions to better attract and engage your target audience, ultimately leading to improved book sales.

Here are the criteria to individually critique each description:

Objective Data:

  1. Length of the description:
    • 1-50 words: 1 point
    • 51-100 words: 2 points
    • 101-150 words: 3 points
    • 151-200 words: 4 points
    • 201+ words: 5 points
  2. Inclusion of key information:
    • Missing 3 or more key elements: 1 point
    • Missing 1-2 key elements: 3 points
    • Includes all key elements: 5 points
  3. Use of persuasive language and marketing techniques:
    • Little to no use of persuasive techniques: 1 point
    • Some use of persuasive techniques: 3 points
    • Extensive use of persuasive techniques: 5 points
  4. Readability and clarity of the writing:
    • Difficult to read and understand: 1 point
    • Moderately clear and easy to read: 3 points
    • Highly readable and clear: 5 points
  5. Formatting and visual elements:
    • No formatting or visual elements: 1 point
    • Some basic formatting or visual elements: 3 points
    • Engaging and visually appealing formatting: 5 points
  6. Keyword optimization:
    • No apparent keyword optimization: 1 point
    • Some keyword optimization: 3 points
    • Excellent keyword optimization: 5 points
  7. Consistency with the book's overall branding and messaging:
    • Inconsistent with the book's branding: 1 point
    • Somewhat consistent with the book's branding: 3 points
    • Highly consistent with the book's branding: 5 points

Subjective Data:

  1. Engagement and interest level generated:
    • Low engagement and interest: 1 point
    • Moderate engagement and interest: 3 points
    • High engagement and interest: 5 points
  2. Perceived trustworthiness and credibility:
    • Low trustworthiness and credibility: 1 point
    • Moderate trustworthiness and credibility: 3 points
    • High trustworthiness and credibility: 5 points
  3. Alignment with the target audience's needs and desires:
    • Poor alignment with the target audience: 1 point
    • Moderate alignment with the target audience: 3 points
    • Excellent alignment with the target audience: 5 points
  4. Uniqueness and differentiation from similar books:
    • Lacks uniqueness and differentiation: 1 point
    • Somewhat unique and differentiated: 3 points
    • Highly unique and differentiated: 5 points
  5. Emotional impact and ability to create a connection:
    • Low emotional impact and connection: 1 point
    • Moderate emotional impact and connection: 3 points
    • High emotional impact and connection: 5 points
  6. Perceived value and benefits offered:
    • Low perceived value and benefits: 1 point
    • Moderate perceived value and benefits: 3 points
    • High perceived value and benefits: 5 points
  7. Overall impression and likelihood of conversion:
    • Low overall impression and likelihood of conversion: 1 point
    • Moderate overall impression and likelihood of conversion: 3 points
    • High overall impression and likelihood of conversion: 5 points

This scale can be used to systematically evaluate and compare the different book sales descriptions, providing a more objective and quantifiable assessment of their potential performance.

When I a book description, I would like you to provide an overall objective data score, an overall subjective data score, a summary of the strengths and areas for improvement in the description, and a breakdown of the specific scores for each criteria.

The subjective criteria wound up being for the most part a waste of time; except for the Gemini 1.5 Flash model (which was scored 3 points for each criteria), it gave a score of 4 across the board to every single criteria on all of the models.

In my opinion, none of these are ready to use “as is” without further editing, but some are certainly closer to usable than others. Here are each of the book descriptions, how much they cost to generate, and how they stack up compared to the others:

Gemini 1.5 Pro (Models #10 & #11)

Claude 3 Opus (Model #4)

Claude 3 Sonnet (Models #5 & #6)

Claude 3 Haiku (Model #7)

GPT-4 Turbo (Model #8)

Gemini 1.0 Pro (Model #12)

GPT-4o (Model #3)

Gemini 1.5 Flash (Model #2)

Claude Instant 1.2 (Model #9)

Gemini 1.5 Flash (Model #1)

If there's one thing that will always be true, it's that there's always new and (potentially) better models. Since writing this article, I have tested a few more models and included the results below so that you can see and compare the results, though I have not changed my original summary based on the first 12 models:

Claude 3.5 Sonnet (Model #13)

GPT-4o mini (Model #14)

Gemini 1.5 Pro Experimental 0801 (Model #15)

Now that we've seen what the generative AI systems have come up with, here is an example that I created using some of the text from the original 12 models as a book description:

Are you ready to take your race to the next level?

Are you tired of the same old race formats? Are you looking for new and creative ways to score your running events and keep your participants engaged? Look no further!

Whether you're organizing your first 5k or a seasoned pro looking for fresh ideas, Race Scoring by Blaine Moore is your comprehensive guide to scoring running races. Drawing from years of experience and countless races, this book provides detailed insights into various scoring methods for individual, series, and team events.

Inside you'll discover how to:

  • Score individual runners in a single race using gun time, net time, age/gender grading, handicapping and more.
  • Determine series winners across multiple races using cumulative time, average pace, and points-based scoring.
  • Set up team competitions with positional cross-country scoring, cumulative time, and relays.
  • Get creative with unique formats like last man standing, non-determinate courses, obstacle races, and virtual events.

Packed with real-world examples and pro tips, this book is a must-have resource for race directors, running clubs, and event organizers.

Whether you oversee a local 5K, manage a regional race series, or aspire to direct a major marathon, this book is your one-stop resource for scoring methods that are accurate, efficient and engaging for runners and organizers alike.

Reinvigorate your events, boost participant satisfaction, and cement your reputation as an innovator in the running community today by purchasing a copy of this book for yourself! Buy now or download to ensure your next event is a winning experience for all of your participants.

Blaine's Draft, 4.71 average score (69 total)

Running my description through the same criteria, it came back with an objective score of 4.86 (on par with models #10 and #11 where it improved on the formatting but fell a little short of those two on the keywords criteria.) However, it also scored 5s across the board for the subjective criteria, of which all of the other models only scored 4s (except for model #1 which scored 3s.) Overall, it scored 7 points better than the next two models.

Before selling the book, it would probably be worth finding some way of including more keywords in the description naturally, but overall I am pretty happy with how it came out. Overall, I spent a total of $0.84 generating all of the book descriptions, and another $0.12 analyzing them.

If you would like to use either of our prompts as a starting point, you can download them as text files here:

Sample Book Description Prompt
Non-Fiction Description Scoring Rubric Prompt

I hope that this case study was helpful for you. It will work most directly with other non-fiction titles, but a similar process could be used for non-fiction and children's books as well. Have you used A.I. to help you write your book descriptions? Why or why not? We'd love to hear about your experiences.

If you are interested, here are the individual scores for each model, direct from the Google Sheet where I was tracking that information:

240606