Book Blurbs: Plot Dump vs Hook-First (Prompt Comparison)
Tidepool Press prints 4,000 first-run copies of every translated crime novel it publishes, so one paragraph on the back cover has to work 4,000 times before a single review exists. The book in question is The Ice Room, a 320-page Norwegian novel in English translation, and the launch is 11 weeks away.
The first blurb the team signed off ran to 214 words, named nine characters, and revealed the second body. A bookseller at the pilot account read it and asked whether the ending was meant to be a surprise. That question is the whole failure in miniature: a plot summary can be perfectly accurate and still be useless as a sales tool.
Why an Accurate Description Sells Nothing
Ask a model for a blurb and it will assume you want a description of the book. Descriptions are easy to grade, easy to fact-check, and easy to write from a synopsis. Selling is none of those things. The model has no way to know that a back cover has to create a reason to open the book in the four seconds a shopper gives it.
So it defaults to coverage. Every plot beat gets a clause. Every named character gets an introduction. The result reads like the paragraph a publisher writes for internal catalogues, where the reader already knows they are buying the book.
- Symptom one: the copy explains the premise instead of applying pressure to it.
- Symptom two: names pile up because the model treats characters as required fields.
- Symptom three: the last line gestures at a twist, which removes the reason to read.
- Symptom four: word count drifts upward, because coverage has no natural end.
None of this is a model defect. A prompt that says “write a blurb about a detective and a murder” has told the model what the book contains and nothing about what the paragraph must do. The model fills that gap with summary, because summary is the safest interpretation of an unclear brief.
The Prompt That Produced a Synopsis
This is close to what Tidepool’s marketing lead typed, minus a sentence of context about the author’s previous sales.
Write a blurb for our crime novel The Ice Room by a Norwegian author.
It is about a detective who investigates a murder at an old mountain
hotel. Make it exciting and around 200 words.What came back: 214 words in one block. It opened with the hotel’s closure in 1998, introduced the victim, the hotelier’s widow, two local police officers, and a journalist, then closed with “a truth that will change everything”. Nine proper nouns in 214 words. The detective’s personal history took three sentences. The actual promise of the book, a man forced to reopen a case he closed badly twenty years earlier, appeared in the middle of paragraph two, where no shopper would reach it.
Two instructions caused most of the damage. “Around 200 words” gave the model room to include everything, and “make it exciting” is an adjective with no unit. Exciting to whom, measured how, at what point in the paragraph? With no answer available, the model did what models do: it raised the volume. More names, more weather, more menace.
The Prompt That Sold the Book
The rewrite keeps the same premise and changes what the paragraph is for. It also spends most of its words on structure rather than on tone, because structure is the part a model can actually follow.
You are writing back-cover copy for a translated literary crime novel.
Book: The Ice Room, 320 pages, Norwegian original, English translation.
Reader: a browser at a bookshop table who has never heard of the author.
Premise you may use: a body is found in the freezer room of a mountain
hotel that closed in 1998. The detective who signed the original report
is sent back to reopen it.
Write 95 to 115 words, in four short paragraphs of one or two sentences.
Paragraph 1: one concrete image from the first forty pages.
Paragraph 2: the pressure on the detective, stated as fact, not feeling.
Paragraph 3: what it costs him if he is wrong.
Paragraph 4: six words or fewer that name the stakes.
Rules: do not name any character, use the detective's job title only.
No spoilers past chapter six. No rhetorical questions. Ban the phrases
"will change everything", "unputdownable", "breathtaking", "secrets",
and "nothing is as it seems". Present tense. British spelling.
Return the blurb only. Then a second line with the word count.What came back: 108 words. The opening image was a freezer door standing open on a strip of melted ice. The second paragraph stated that the case file still had his signature on it. The third said a wrong answer this time would end a career rather than a case. The closing line ran five words. No names, no twist, no weather.
The team read it aloud in a launch meeting and nobody stumbled. That is a low bar and a revealing one: most generated blurbs cannot survive being read aloud, because they are built for the eye scanning a page of text rather than for a voice.
Four Lines, Four Jobs
The structure matters more than the adjectives. Each line does one thing, and each thing fails in a predictable way when it is missing.
| Line | Job | What breaks without it |
|---|---|---|
| One | Give the reader one image they can see | The blurb opens on abstraction and the eye slides off |
| Two | State the pressure as a fact | The stakes become mood, and mood does not compel |
| Three | Name the cost of being wrong | The detective becomes an observer instead of a participant |
| Four | Close on the stakes, not the solution | The ending gets implied and the reason to read disappears |
Notice what the table leaves out: genre labels, comparisons to other authors, and praise. Those belong in metadata and endorsement slots. Mixing them into the blurb is how a 110-word paragraph turns into 200 words of positioning.
The spoiler rule deserves its own note. “No spoilers past chapter six” is checkable against the manuscript. “Don’t spoil it” is not, because the model’s sense of what counts as a spoiler comes from marketing copy, where revealing the premise is normal. Give it a page boundary and the argument disappears.
If you want the model to match a house style rather than invent one, the mechanics of supplying examples are covered in our guide to one-shot versus few-shot examples. Two back-cover paragraphs from your own list teach shape faster than any adjective.
Rewriting Your Own Back Cover
This skeleton works for fiction and for narrative non-fiction with minor edits. Replace the brackets and delete any line you genuinely do not need, rather than leaving it blank by accident.
Role: back-cover copywriter for [imprint], specialising in [genre].
Book: [title], [page count], [original language], [translator credit if any].
Reader: [where they are standing and what they already know].
Premise you may use: [three sentences, no ending].
Length: [X] to [Y] words, [N] short paragraphs.
Paragraph jobs: [image] then [pressure] then [cost] then [closing line].
Spoiler boundary: nothing after [chapter or page].
Banned: character names, rhetorical questions, [list of tired phrases].
Voice: [tense], [spelling], [one sentence on register].
Output: blurb only, then a word count.Fill the spoiler boundary first. It is the line that changes the output most, and the one most people leave out because it feels obvious. It is not obvious to a model that has read thousands of back covers where the twist is teased.
Then set the paragraph jobs. Four jobs in four paragraphs gives the model a shape to fill rather than a mood to imitate, and it gives you a way to reject a draft in ten seconds: if paragraph three does not name a cost, the draft is wrong regardless of how good the sentences sound.
Finally, run one critique pass before you send the file to the designer. Ask a fresh session to mark each paragraph against its assigned job and to list every proper noun. That second pass catches the failures that survive a first read, especially reintroduced names and a closing line that quietly explains the plot. A reusable version of that loop is described in our self-critique prompt.
Tidepool now runs this skeleton for every title in the translated list, changing only the premise block and the image in paragraph one. The blurb stops being a writing task that starts from zero and becomes a slot-filling task with a review step. Eleven weeks is enough time for that, and so is eleven days.
Sources & Notes
The four-line structure, the spoiler boundary rule, and the skeleton are our own framing, reviewed on 2026-10-20. Tidepool Press, The Ice Room, the 320 pages, the 4,000-copy first run, and the 11-week schedule are illustrative examples, not a real imprint or title. Word counts for back covers vary by format and printer; check your own cover template before fixing a limit. Banned-phrase lists date quickly, so refresh them each season rather than treating any list as permanent. Language models also change between versions, so re-test the prompt on a title you have already published before using it on a new one.
