ForkFree

Prompts / Research and analysis

Goal-First Document Review With Set Answer Shape

You get a focused review of a long document in exactly the format you asked for. It works because the material comes first, the goal and answer shape come last, and the model is told what you want rather than how to think.

Struggled in our testVerdict, quotations and ranking are fine, but several bullets exceed the 25-word limit. The third bullet also combines two risks plus an extra sentence, which strains "exactly three bullets".

The prompt

You are a careful contracts analyst helping a small business owner who is not a lawyer.

Task: Read the document below in full, then work towards the goal given after it.

Document:
{document_text}

Goal: {goal}
Answer shape: {answer_shape}

Steps:
1. Read the whole document before you form any view.
2. Decide for yourself how best to reach the goal; do not follow a fixed checklist.
3. Support each point with a short quotation from the document.
4. If the document does not contain enough to answer part of the goal, say so plainly instead of guessing.

Output:
Reply only in the answer shape given above, with no introduction or closing remarks.

Change the parts in curly braces. Use it when you need a quick decision from a long contract, report or policy and want the reply in a fixed, scannable format.

What happened when we ran it

We sent

You are a careful contracts analyst helping a small business owner who is not a lawyer.

Task: Read the document below in full, then work towards the goal given after it.

Document:
This agreement runs for 12 months from 1 March 2026 and renews automatically for further 12-month terms unless either party gives 90 days' written notice. The Supplier may raise prices once per term by up to 8% with 30 days' notice. Late payments incur interest at 4% above the Bank of England base rate. Either party may terminate immediately if the other commits a material breach that is not remedied within 14 days.

Goal: Decide whether I can safely sign this supplier contract for my bakery, and identify anything that could cost me money unexpectedly.
Answer shape: A verdict of 'sign', 'negotiate' or 'do not sign' on the first line, followed by exactly three bullet points, each under 25 words, ranked by financial risk.

Steps:
1. Read the whole document before you form any view.
2. Decide for yourself how best to reach the goal; do not follow a fixed checklist.
3. Support each point with a short quotation from the document.
4. If the document does not contain enough to answer part of the goal, say so plainly instead of guessing.

Output:
Reply only in the answer shape given above, with no introduction or closing remarks.

The model replied

Negotiate - Price rises: "raise prices once per term by up to 8% with 30 days' notice" lets costs jump each year; seek a cap tied to inflation and longer notice. - Auto-renewal: "renews automatically for further 12-month terms unless either party gives 90 days' written notice" can lock you in; diarise the notice date or shorten it. - Late interest and breach: "4% above the Bank of England base rate" and "not remedied within 14 days" are tight; ask for a longer cure period. Prices, volumes and minimum purchases are not stated.

Run on Claude Sonnet 5.5, 28 Oct 2026. Automatic check: 2 out of 5.

Mentioned for

Claude, ChatGPT, Gemini

Where it came from

Found on Web, shared by Emma Al (AI Portalen), Substack, 24 Oct 2026: ChatGPT, Claude, or Gemini: Which One Should You Use, and How Do You Prompt It?.

This is our own version of the idea, written and tested by us. The original may differ.

More research and analysis prompts

Research and analysis Worked in our test
Fact-Check an AI Answer Against Sources

You get a claim-by-claim audit of an AI answer, showing exactly which statements are backed by your source text and which were made up. It works because the model must quote evidence for every claim and is barred from using outside knowledge.

Found on GitHub, shared by comet-ml
Research and analysis Worked in our test
Audit Whether Citations Support Claims

You get a claim-by-claim table showing which statements are backed by the cited evidence and which are overstated or unsupported, with safer rewordings. It works because it makes the AI match each claim to a specific passage instead of trusting that a citation exists.

Found on GitHub, shared by Imbad0202