Building Your Own Prompt Templates
Stop re-explaining recurring work every time
① Write a fresh prompt every time and you'll drop or vary the context each time, which makes the output format shift and adds review time. A template removes that repeated cost.
② A good template separates the fixed frame from the slots you fill in. The instructions you've already validated (role, format, constraints) stay put — only the variable part gets swapped.
③ The most common way a template breaks is being written too specifically for one case. Build it around stability, verifiability, and minimal editing, and it holds up over time.
① Why not just write it fresh each time — the cost of repetition
The seven principles in prompt engineering basics (role, context, format, examples, constraints, verification, iteration) were about getting one prompt right. The trouble is refilling all seven from scratch every time for work you repeat weekly.
Two things leak out when you do that. First, you describe the context slightly differently each time, and that variance breaks consistency in the result. Second, if the output format shifts subtly each time, you have to figure out what to check this time, all over again, on every review. A template removes that recurring cost by reusing a frame you already got right once.
② The structure of a template — separate the fixed frame from the fill-in slots
What separates a one-off prompt from a template isn't length — it's whether it was designed to be reused.
| Type | Example |
|---|---|
| One-off prompt | "Summarize Nvidia's earnings release, focused on revenue and guidance, as a table" — you rewrite this whole sentence, or something close to it, from memory, every time |
| Template | "Read [company]'s [quarter] earnings release. Put revenue, guidance, and margin changes into a table, and give a one-line verdict on whether it was a surprise versus expectations. Show, as a quote, the sentence you based that verdict on." — swap only the [bracketed] slots and reuse it for the next quarter or another company |
The key is the [bracketed] variable. The rest of the instructions — role, output format, verification requirement — are already validated, so you leave them alone and swap in only the value that changes. Instead of reassembling the seven principles every time, you keep reusing a frame you assembled once.
③ Three conditions for a good template
| Condition | What it means |
|---|---|
| Stability | Avoid wording tuned to one case. Instead of "like this time when Nvidia raised guidance," generalize to "whether guidance was raised, lowered, or held versus last quarter." |
| Verifiability | Don't just take the output — require the reasoning with it. Baking in one line like "quote the sentence this judgment was based on" leaves you a thread to check later whether it was right. |
| Minimal editing | Mark the slots you fill each time clearly with [brackets], so the rest of the sentence never needs to be touched. If it's unclear what to edit, the template has lost its point. |
The seven principles said giving an example has the biggest impact — but locking a specific example into a template actually hurts stability, because a new input that differs in kind from that example pulls the result toward the example instead. If you need an example, it's better to leave a slot in the template where you drop in a fresh, situation-specific one each time.
④ In practice — three templates for recurring work
Turned into templates, the work an overseas investor in US stocks repeats often looks like this. Swap only the [bracketed] slots each time.
| Use case | Template |
|---|---|
| Checking an earnings release | "Here is [company]'s [quarter] earnings release. Put revenue, EPS, and guidance into a table, and mark whether each beat, met, or missed the consensus estimate. Quote the sentence each judgment was based on, and don't add any figure not in the source." |
| Extracting the core of an analyst report | "Here is a broker report on [ticker]. In five lines, summarize whether the price target went up, down, or held versus before, and whether the driver was a change in the earnings estimate or a change in the multiple. Don't add interpretation not in the source." |
| Pulling your watchlist out of a daily brief | "Here is today's full US market brief. Pull out only what relates to my watchlist — [ticker 1, ticker 2, ticker 3] — and give one line each on which data point or event they were mentioned for. If a name isn't mentioned, write 'not mentioned.'" |
What framework to read an earnings release with is in the Big Tech earnings guide; how a price target and its multiple are built is in how to read an analyst report. A template is only a tool that saves you from retyping that framework by hand every time — the judgment criteria themselves still have to come from the source article.
⑤ Where templates break
Even a working template stops fitting over time. Three common causes.
- The input format changed — if a brief's layout or a report's format shifts, an instruction like "put it in a table" may no longer match the new structure.
- It was overfit to one case — if you tuned the wording around the one example you had when you first built it, it behaves oddly on inputs of a different character.
- Your criteria changed but the template didn't — for example, your watchlist changed but you never updated [ticker 1, ticker 2, ticker 3] in the template.
The fix is simple. When a template misbehaves, suspect the template before the new input, and keep a dated note of what you changed. Don't aim to build the perfect template on the first try — it's more realistic to build it assuming you'll fix it when it stops working.
⑥ Storing and managing them
The hard part of a template isn't making it — it's finding it again to reuse it. Two approaches.
- A service's built-in feature — many AI services let you save frequently used instructions, or auto-load a setting (like custom instructions) into every conversation. The upside is not having to paste it in each time; the downside is you can't carry it over if you switch services.
- Managing it yourself — keep a "template collection" in a notes app or a single document and paste from there. It isn't locked to one service and works with any AI. If you revise often, noting the last-edited date in the filename or at the top makes it easy to tell later whether it's the current version.
If you share a template with a team or reuse it across many conversations, check once whether it has company information or someone else's information baked in — the risks covered in AI and your privacy apply just as much to templates.
⑦ A checklist before building a new template
| Question | Why it matters |
|---|---|
| How many times will you likely repeat this task? | For once or twice, building the template can cost more effort than it saves |
| Have you separated what varies from what stays fixed? | If the line is blurry, you'll keep having to rewrite the whole sentence instead of just the [brackets] |
| Does it include a line that demands reasoning? | Without one, you have to verify the output from scratch every time |
| Any wording left that only fits one specific case? | If so, it will behave unexpectedly on a different input |
A template is a fixed set of instructions you've already validated, with only the variable values swapped in. Don't lock in a fixed example, do require the reasoning, and mark the slots clearly — and it holds up over time. When it stops working, suspect the template before the input.
※ Written as of September 2026, describing general principles. How saving and recalling instructions works — and what it's called — varies by AI service, and the sample templates are meant to be adapted rather than copied verbatim. When handling documents that feed an investment decision, verify figures in the source rather than in a summary.
※ This guide is provided for general educational purposes and simplifies technical details for readability.
New guides, when they land
We publish AI literacy guides twice a week. Subscribe and the next one comes to you — free, unsubscribe anytime.