Prompt Template Builder

How to use the Prompt Template Builder

1. Pick an example or paste your own prompt. Wrap each part that changes in double braces, such as {{audience}}.
2. Add a default after a vertical bar, as in {{word_limit|150}}, or leave it empty ({{notes|}}) when the input is optional.
3. Fill in the fields that appear. Each name gets one field, however often it repeats, and starts with its default.
4. Read the preview. Typed values are highlighted, defaults dotted, and anything without a value or default stays visible as {{name}} and is listed in a warning.
5. Copy the filled prompt into your chat window or application, or copy the template to keep the placeholders.
6. To keep a template, name it, describe the task, record the models you tested it on, its owner, and the date you last verified it, then save it to the library.

How it works

A variable is a name between double braces: letters, digits, spaces, underscores, and hyphens, case sensitive, with outer spaces ignored. Other double-brace text, such as {{user.email}} or an unclosed pair, is left as typed and flagged under the editor, so a typo cannot pass for a variable. Substitution is single-pass: a value containing {{other}} is inserted as plain text.

The examples share one layout: a role, each input in its own tag, a task, rules, and an explicit output format. Anthropic’s prompting guide says wrapping each type of content in its own tag reduces misinterpretation, and OpenAI’s guide says XML tags can mark where a supporting document begins and ends.

Two limits. Delimiters are not a security boundary: a value containing a block’s own closing tag, such as </notes>, ends the block early, so the preview warns about it, and OWASP’s 2025 entry on prompt injection says it is unclear whether fool-proof prevention exists. And tokens are a rule of thumb: OpenAI puts one token at about four characters of English text, while Anthropic says its models from Claude Opus 4.7 onward produce roughly 30 percent more tokens for the same text than earlier ones, depending on the content. Use your provider’s token counter before relying on a limit.

Variables and defaults

Give every variable one of three roles: required ({{audience}}), which stays visible until filled; defaulted ({{tone|warm}}), where an empty field uses the default; or optional ({{extra|}}), where an empty field inserts nothing. A blank required variable that quietly becomes “Rewrite this for .” can lead to a confident answer to the wrong question, which is why unfilled placeholders stay visible. Variables named today or dateget a “Use today’s date” button.

• Name a variable for what it holds: meeting_notes, not input.
• Keep each input in its own block, with a rule that it is data to process, not instructions to follow.
• Test with invented data first. Nothing typed here leaves your browser, but whatever you paste into a chat tool reaches that provider.

Worked templates and a fixture-testing routine are in how to write reusable prompt templates.

Checking templates after a model update

Providers retire and replace models. Anthropic says it gives customers with active deployments at least 60 days’ notice before retiring a publicly released model, and OpenAI’s prompt guide recommends pinning production applications to specific model snapshots and building evaluations that let you monitor performance as you iterate or change model versions. A template that worked last quarter is a claim with a date on it, which is why each entry records target models and a last-verified date, and why the list flags entries not verified in more than 90 days (our rule of thumb). Formatting alone can move results: Sclar and colleagues reported accuracy differences of up to 76 points between formatting variants of the same prompt on LLaMA-2-13B in few-shot settings, so re-run your test inputs after editing whitespace or delimiters too.

Sharing a library with a team

Export writes one prompt-templates.json, sorted by name with a fixed field order so changes show as small diffs. Commit it to a repository or attach it to a wiki page. Teammates import it, and entries merge by ID. Ownership and review are covered in how to build a prompt library your team will use. Keep the master copy somewhere you control: OpenAI’s deprecations page lists reusable prompt objects and the v1/prompts API for shutdown on November 30, 2026.

Using templates for study prompts

A study prompt is a template with {{topic}}, {{notes}}, and {{question_count|5}}. The prompts in how to learn with AI without fooling yourself, such as quizzing yourself from your own material or asking for hints instead of answers, work well as saved templates.

What it does not do

It does not run, score, or test a prompt. There is no model behind it, so you judge the output in the tool you paste into. It has no conditions, loops, or nested templates: a variable is replaced by text, nothing more. It does not sync between browsers or devices, and the only history it keeps is the last-verified date and your notes, so export before you clear site data or switch computers.

Privacy

The tool’s own code makes no network requests and calls no AI model, so nothing you type is sent to us. Your draft and library live in this browser’s local storage on this device, under keys starting with lhp.; values typed into variable fields are not saved. Clearing site data deletes the library, so export it if it matters. The privacy policy covers the site-wide scripts this page also loads.

Frequently asked questions

Does it send my prompt to an AI model?

No. It fills in text and shows it. You copy the result into whichever tool you use.

Why does a placeholder still show in the preview?

That variable has no value and no default. It stays visible on purpose, so an empty field cannot quietly produce a vague prompt.

How do I move my library to another computer?

Choose “Export all as JSON” here, then import the file there. Nothing syncs on its own, because nothing leaves your browser.

Will one template behave the same on every model?

The structure carries over; the results may not. Test each model you use and record the verified ones in the target-models field.

Related guides

• AI Meeting Summaries at Work: Record, Prompt, Check, RetainA workflow for team AI note-takers: when not to record, what to tell people, a summary prompt that quotes the transcript, a 5-minute check, and retention rules.
• Using AI to Write Internal Documentation Without Making It WorseUse the four Diátaxis types to decide what each internal doc is for, give AI only jobs it can source, then review and assign an owner so pages stay true.
• How to Learn With AI Without Fooling YourselfAI explanations feel like learning even when they aren't. See what 2025–2026 trials found, plus seven copyable study prompts and a weekly loop to follow.
• How to Write Reusable Prompt Templates, With Five to CopyBuild prompt templates that behave the same way twice: six parts, safe variables, fixture tests, versioning, and five complete templates you can paste.

Sources

• Prompting best practices — Anthropic, accessed October 2026. XML tags.
• Token counting — Anthropic, accessed October 2026. Counts are estimates; newer tokenizer.
• Model deprecations — Anthropic, accessed October 2026. Retirement notice.
• Prompt engineering — OpenAI, accessed October 2026. Delimiters, snapshots, tests.
• Deprecations — OpenAI, accessed October 2026. Reusable prompts shutdown.
• Key concepts — OpenAI, accessed October 2026. Tokens.
• Quantifying Language Models’ Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting — Sclar, Choi, Tsvetkov, and Suhr, 2023 (ICLR 2024). Formatting sensitivity.
• LLM01:2025 Prompt Injection — OWASP Gen AI Security Project, 2025. Prompt injection.