AI Token Saver: Caveman Prompt Generator

What is Caveman Prompting?

Caveman prompting is a prompt instruction that tells an AI model to answer in compressed, fragment-style language: no pleasantries, no preamble, no recap, no filler words, just the facts in the fewest words. Large language models are verbose by default, and every word of that padding is an output token you pay for and read through. Output tokens usually cost several times more than input tokens on API pricing, so a short brevity instruction at the end of a prompt can cut response length substantially: Verbito's estimates are about 30% for Lite, 60% for Caveman and 75% for Ultra, depending on the model and task. Verbito's Token Saver adds this instruction for you, and it always protects code, commands, numbers, names and quotes from being compressed.

How to use the AI Token Saver: Caveman Prompt Generator

1. Describe your goal. Type what you want the AI to produce, in plain language.
2. Pick format, tone and Token Saver. Choose the output format and tone, and optionally a Token Saver level (Lite, Caveman or Ultra) so the AI answers in fewer tokens.
3. Generate and copy. Verbito builds a structured, scored prompt. Copy it into ChatGPT, Claude, Gemini or any other AI model.

Best practices

• Use Caveman or Ultra for answers you will scan, not publish: debugging help, explanations, research notes, checklists and quick decisions.
• Use Lite or Off for customer-facing copy such as emails, blog posts and ads, where full sentences and tone matter.
• Keep code, commands, figures and quotes verbatim. Verbito's directive already says so, so do not remove that line.
• Put the brevity instruction at the end of the prompt, as Verbito does, so it is the last thing the model reads before answering.
• Measure it: compare the token counts of the same question with Token Saver off and on in your AI tool or API dashboard.
• If an answer gets too terse to follow, step down one level rather than turning Token Saver off entirely.

Tips

• Lite keeps full sentences but removes filler: a safe default for everyday work.
• Caveman drops articles and transitions: the best balance of savings and readability for technical Q&A.
• Ultra uses symbols, abbreviations and one idea per line: best for notes, logs and experienced readers.
• Token Saver reduces output tokens. The directive itself adds roughly 100 input tokens, which is small next to the savings on any long answer.
• It is disabled for Midjourney, DALL-E, image and video prompts, because image generators do not produce text answers to shorten.
• Your prompt score includes a Token Efficiency dimension, so prompts that bound their output length score higher.

Examples

Normal AI answer (165 tokens)

"Great question! When it comes to database performance, there are a few important things to keep in mind. A sequential scan reads every row in the table, which can actually be quite efficient for small tables or when a large share of the rows match your filter. An index scan, on the other hand, uses the index to jump directly to the matching rows, which is usually faster when your filter is selective and only returns a small fraction of the table. In general, the query planner chooses between them based on its cost estimates, which depend on table statistics. If those statistics are out of date, the planner can make a poor choice, so running ANALYZE on the table is often a good first step. I hope this helps! Let me know if you'd like examples of how to read an EXPLAIN plan."

Caveman answer (57 tokens, −65%)

"Seq scan: reads every row. Cheap for small tables or when most rows match. Index scan: jumps to matching rows via index. Wins when filter selective (<~5–10% rows). Planner picks by estimated cost; stale stats → bad pick. Fix: ANALYZE."

Ultra answer (36 tokens, −78%)

"Seq scan = read all rows → small table / low selectivity. Index scan = jump via index → selective filter. Planner: cost estimate; stale stats → ANALYZE."

Where not to use it

Anything a customer or reader will see as finished writing: newsletters, landing pages, cover letters. Generate those with Token Saver off or on Lite.

Example generated prompts

Debugging help (Caveman)

Goal: Help me debug why my React useEffect runs twice in development (prompt score 91/100)

You are a senior React engineer. Diagnose why a useEffect callback runs twice in development.

Cover: React 18+ StrictMode double-invocation, how to confirm it is the cause, how to write effects that are safe to run twice (cleanup functions, abort controllers), and when a double run signals a real bug.

Output: numbered causes, most likely first, each with a one-line check and a minimal code fix.

## Response style — Token Saver (Caveman)
Respond in caveman style: short fragments, answer first. Drop articles, filler words, pleasantries, hedging, and transitions. No preamble, no recap. Keep every fact, technical term, and step — cut words, not meaning. Never compress code, commands, numbers, names, quotes, or anything meant to be copied verbatim. If brevity would make the answer ambiguous or an instruction unsafe, spell that part out in full.

Research summary (Ultra)

Goal: Summarize the trade-offs between RAG and fine-tuning for a support chatbot (prompt score 89/100)

You are an applied ML architect. Compare retrieval-augmented generation (RAG) and fine-tuning for a customer-support chatbot over a 2,000-article help center that changes weekly.

Compare on: freshness, cost to build and run, hallucination risk, latency, and maintenance effort. End with a recommendation and the one condition that would flip it.

## Response style — Token Saver (Ultra)
Maximum compression. Telegraphic fragments only; bullets over prose. Symbols and common abbreviations allowed (→, =, &, w/, e.g.). Zero preamble, zero recap, zero pleasantries. One idea per line. Keep every fact and step. Never compress code, commands, numbers, names, quotes, or anything meant to be copied verbatim. If brevity would make the answer ambiguous or an instruction unsafe, spell that part out in full.

Watch it in 30 seconds

• You're paying for ChatGPT's small talk: The same database question answered normally (165 output tokens) and in Caveman mode (57 tokens), and how Verbito adds the Token Saver instruction to a prompt.
• Caveman mode has three levels: Lite, Caveman and Ultra explained with examples, and what Token Saver never compresses: code, commands, numbers, names and quotes.
• Your prompt score now counts tokens: A real Verbito result: the same prompt scored 87 without Token Saver and 92 with Caveman mode, with token efficiency rising from 80 to 90.

Frequently asked questions

What is caveman prompting?

Caveman prompting is an instruction that makes an AI model answer in compressed fragments, with no filler, pleasantries, preamble or recap, so responses use far fewer output tokens while keeping the facts. Verbito's Token Saver adds this instruction to any prompt at three levels: Lite, Caveman and Ultra.

How many tokens does Token Saver save?

It depends on the model and the question. Verbito estimates about 30% fewer output tokens with Lite, about 60% with Caveman and about 75% with Ultra on typical explanatory answers. Savings are largest on questions that normally trigger long, padded replies. The directive itself adds roughly 100 input tokens.

Does caveman mode make AI answers worse?

It removes words, not information: the directive tells the model to keep every fact, technical term and step, and never to compress code, commands, numbers, names or quotes. It is a poor fit for customer-facing writing where tone matters, so use Lite or turn it off for emails, articles and ads.

Does it work with ChatGPT, Claude and Gemini?

Yes. Token Saver is a plain-language instruction inside the prompt, so it works with ChatGPT, Claude, Gemini, Grok, DeepSeek and any other text model. It is disabled for image and video generators such as Midjourney and DALL-E, which do not write text answers.

Is the AI Token Saver free?

Yes. Token Saver is free on every Verbito plan, including the free tier with no signup required.

How does Token Saver affect my prompt score?

Verbito scores prompts on seven dimensions: clarity, context, specificity, constraints, output format, reusability and token efficiency. Turning on Token Saver guarantees a high token-efficiency score because a hard limit on response verbosity is built into the prompt.