A prompt template is a reusable instruction with clearly marked blanks for the parts that change—the reader, the notes, the chapter goal, the examples—so you do not rewrite the brief from scratch each time. OpenAI’s prompt engineering guide recommends managing production prompts in a structured way, with typed inputs or clearly defined dynamic values such as task options and source material, and it organizes longer instructions into identity, instructions, examples and context. That is a template in all but name: a stable frame with slots for the work that changes. Google’s prompt design strategies add that prompt design is iterative, and that you should experiment and refine based on the results you actually see. A template makes that iteration possible, because you can change one slot without rewriting the rest.
For eBook and workbook projects, Interactive eBook Builder is a $59 one-time Notion AI template with Spec Builder, Drafter, Polisher and SEO BOT already set up, so the standing brief can live in those stages while you only swap the chapter notes. Notion AI access and usage are separate from the purchase. The examples below are teaching prompts written for this article, not paid Builder prompts and not recorded model outputs.
Why a template beats a pile of one-off prompts
A one-off prompt can be excellent and still be a dead end. You get a good section, you cannot remember what you asked, and the next chapter starts from a blank box. A template solves a practical problem: the book’s rules stay still while the chapter’s material moves.
It also makes comparison fair. If chapter 3 feels worse than chapter 2, you can ask whether the notes were thinner, the job was vaguer, or the template itself needs a change. Without a template, you are comparing two completely different briefs and guessing.
OpenAI also notes that keeping reusable content in a stable position can help with prompt caching in the API, which is a cost and latency concern for developers. Even if you only work in a chat box, the same habit helps you: put the standing rules in a block you rarely touch, and put the new notes in a block you always replace.
What belongs in the standing part?
The standing part should be true for every chapter, or for every page of a given type. A useful default is:
- Identity: who the model is helping, and the voice of the book.
- Instructions: what to do, what to flag, what never to invent.
- Format: the heading pattern, length range and exercise shape.
- A rule for uncertainty: “Need from me” instead of a made-up fact.
- Optional short examples, if you have approved ones.
Anthropic’s prompting best practices recommend being explicit about the output you want, and giving a role when it helps focus tone. Those pieces belong in the standing section when they apply to the whole book. Chapter-specific facts do not.
Leave out anything that will be wrong on the next page: today’s outline, a single quotation, a temporary reminder to “make chapter 4 shorter.” Those go in the variable section.
What belongs in the blanks?
The blanks are the work that changes. Name them in the same way every time so you do not smuggle a new kind of instruction into the notes by accident.
| Slot | What you paste in | What you should not hide there |
|---|---|---|
| Reader | Who this page is for, if it ever differs from the book default | A new voice rule that belongs in the standing section |
| Job | What the reader should be able to do after this page | A new format for the whole book |
| Notes | Facts, examples and limits for this chapter only | Standing policy, or a second entire chapter “for context” with no label |
| Examples | One or two approved samples, when this page type needs them | Unedited old drafts that would fail your own rules |
| Request | The deliverable you want now: outline, draft, exercise, edit | A new standing rule phrased as a one-off request |
If you find yourself writing a new policy inside Notes, move it up. If you find yourself pasting the same policy into Request, move that up too. The template only works when each slot has one job.
A one-off prompt, then the template version
A one-off brief for a single chapter might look like this:
Write a 300-word section for new freelance designers about stating a
revision limit in the kickoff email. Use a calm tone. Don't invent legal
rules. End with a one-line email sentence the reader can copy and adapt.
Notes: the reader says yes to extra revisions because they are afraid of
seeming difficult. I have not chosen a default number of revisions.
That can work once. As a template, the same brief becomes a frame with labels:
Identity
You are helping the author of a short eBook for new freelance designers.
Use plain, calm language. Stay inside the notes. If a fact, number or
rule is missing, write "Need from me" instead of inventing it. Do not add
legal, medical or income claims.
Format
- Teaching sections: 250 to 350 words, then one short example or script
the reader can reuse.
- Workbook pages: a one-sentence instruction, fill-in lines, and one
backup question.
Reader
{{READER}}
Job
{{JOB}}
Notes
{{NOTES}}
Request
{{REQUEST}}
Filled in for the same chapter, the variables might be:
Reader
A new freelance designer who says yes to extra revisions.
Job
Help them put a revision limit in the kickoff email.
Notes
- The fear is seeming difficult, not a lack of email skill.
- No default number of revisions has been chosen; the reader will fill it in.
- The email sentence should leave a blank for the number.
Request
Draft the teaching section, then the one-line email sentence. Do not
write the full workbook page yet.
The filled version is not much longer than the one-off prompt. The difference is that tomorrow’s chapter can reuse the frame. You can also keep an “outline” request, a “draft” request and an “edit” request as three small templates that share the same identity and format blocks.
How should you label the blanks?
Plain labels work in a chat box: Reader, Job, Notes, Request. If you want tighter boundaries, use the delimiter style vendors recommend. OpenAI suggests Markdown headings and XML tags to mark sections. Anthropic recommends consistent, descriptive XML tags. Google’s Gemini 3 guidance says to pick XML-style tags or Markdown headings and use one format consistently. For a deeper treatment, see how to use delimiters in prompts.
Whichever style you choose, be consistent. If Notes is a heading in one chapter and a tag in the next, you have made the template harder for both you and the model to parse. And if you use placeholders such as {{NOTES}}, replace them all before you run the prompt. A leftover placeholder is a hole the model will try to fill.
Where should each part go?
Placement matters more once the notes get long. Anthropic advises putting long documents near the top, with the query last, and reports that queries at the end can improve response quality by up to 30 percent in tests, especially with complex, multi-document inputs. Google’s Gemini 3 guidance says the same for large amounts of context: supply the context first, and put the specific instructions or questions at the very end. OpenAI, for API caching, suggests keeping stable, repeated content at the beginning of the prompt.
Those tips can be combined without contradiction if you think in two layers:
- Standing template text that you reuse often can stay at the top.
- Long chapter notes or source documents also go before the question.
- The request for this turn goes last.
If a tool only gives you one box, the order can be: standing rules, then notes, then the request. If a tool has a system or instructions field, put the standing rules there and still end the user message with the request.
How do you test and revise a template?
Do not roll a template across a book after one successful chapter. Google’s iteration advice applies: try the same frame on a second, slightly different input. A teaching chapter and a workbook chapter is a good pair. If both meet the requirements, keep the template. If only one does, look at the slot that changed.
Revise the standing section when the same miss appears more than once: invented numbers, a heading pattern that slips, a workbook page that keeps turning into an essay. Revise the request slot when the miss is local to that deliverable. If you change both at once, you will not know which change helped.
Save the template, the filled variables and the inspected result together. A template you cannot find is just another one-off prompt.
How this maps to a book-building system
Spec Builder, Drafter and Polisher in Interactive eBook Builder are three reusable instruction sets for three jobs: plan the chapter, draft it, then refine it. You still supply the seed, the audience and the notes, and you still review the work. After you set up SEO BOT, it can publish supporting articles to the Auto-Blogger blog, each linking back to your eBook, site and brand. The product is a working template system, not a substitute for reading the output. You can keep the supplied setup or adapt the audience, voice, format and activities.
If you are starting from a chat tool, you can get most of the benefit with one standing brief and a few labeled blanks. You do not need to invent a complicated framework before you have a template that survives two chapters. When you want a checklist of parts, our guide to prompt engineering frameworks compares the common ones.
FAQ
Is a prompt template the same as a system prompt?
A system prompt is often the standing part of a template. The template also includes the labeled blanks you fill in for each request. You can store the standing part in a system or instructions field when you have one, and keep the blanks in the user message.
How many templates do I need for one eBook?
Often two or three: one for outlines, one for drafts, and one for exercises or edits. They can share the same identity and limits. A new template for every chapter usually means the variables were not designed clearly enough.
Should I put examples in the template or in the blanks?
Put examples in the standing section when every page of that type should follow them. Put them in a blank when only some requests need a sample, or when the sample will change. Approved, short examples are more useful than a pasted chapter that would fail your own rules.
What if my template keeps getting longer?
Move one-off comments out. Merge duplicate rules. If the standing section is now a second manuscript, you are asking the model to obey a policy you can no longer review. Our guide to how long a prompt should be can help you trim without losing the decisions that matter.
Before you hit send
- Have you separated standing rules from the variables for this chapter?
- Are the blanks labeled in the same way you will use next time?
- Did you replace every placeholder with real notes or a real request?
- Is the request at the end, after the notes the model must use?
- Have you tested this template on more than one chapter or page type?
- Are you saving the filled template with the inspected result, not only the output?