promptengineering.ai

Sample report

Exactly what you get, thirty seconds after pasting.

Marie, a marketing manager, pasted “Write a LinkedIn post about our new product launch.” This is her report — yours will look like this, about your prompt.

report № 1204 · 2026-08-25 · ran on fast tier

Plate Ⅰ · Diagnosis

41 / 100needs work
  • No audience defined.The model guesses “everyone” — and a post written for everyone is written for no one.
  • No voice, no constraints.An open invitation to boilerplate: “We're thrilled to announce…” is what fills the silence.
  • No example to match.One post whose tone you liked is the strongest lever you're not using.

Plate Ⅱ · Your prompt, rewritten

tuned per model — same intent, different craft
Specimen · optimized · chatgpt
You write LinkedIn posts for founders and heads of product at B2B
SaaS companies (50–500 employees).

Write a post announcing our product launch.
- Open with a tension or observation, never the announcement itself
- Voice: direct, concrete, lightly wry. First person singular.
- Hard bans: "thrilled", "excited to announce", "game-changer", emojis
- 120–180 words, short paragraphs, one question at the end

Here is a post whose tone to match:
"{paste a post you admired}"

ChatGPT drifts toward enthusiasm; the hard-ban list and the word budget are doing the real work here.

Plate Ⅲ · Proof, side by side

run live on gpt-4.1 nano · temp 0.7 · 2026-08-25
Specimen A · your promptas pasted
🚀 We're thrilled to announce the launch of
our new product! This game-changing solution
will revolutionize the way teams collaborate.

Our team has worked tirelessly to bring this
vision to life, and we couldn't be more
excited to share it with you all…
Specimen B · optimized★ the keeper
Most product launches die in the feed.
Ours almost did too.

Six months ago we watched a customer
rebuild our core feature in a spreadsheet —
because our version asked for eleven clicks
where she needed two…

Same model, same temperature. The prompt is the difference.

Plate Ⅳ · Which model, for this task

tested2026-08-25

Marketing copy — panel score, fast tier

Your optimized prompt, run on all three. Median of 5 marketing tasks, scored blind · scale /10

claude haiku 4.5
★ best today
7.8
gpt-4.1 nano
6.4
gemini flash lite
5.9

The verdict changed this release — it was GPT’s for four months. See the monthly battery data →

Keep this working

Run your own prompt.

The Workbench is free — three tests a day, no account, evidence included.

Open the Workbench →