BBotnexio
EnglishDeutsch

Buy now!

← All guides

AI Research Prompts: Research, Check Sources and Summarize Better (2026)

Deutsche Version

AI tools are fast at research and just as fast at sounding confident when they're wrong. Good prompts don't remove that risk, but they make the model show its work so you can check it. This guide covers three prompt types: research, source checking and summarizing.

1. Structure a research prompt

A research prompt works best with five parts:

  1. Role and goal – what you want to know and why.
  2. Scope – time range, region, type of sources.
  3. Output format – table, bullet list, JSON.
  4. Evidence rule – every claim needs a source, or must be marked as unverified.
  5. Unknowns – ask the model to list what it couldn't find.
Goal: Understand {{topic}} for {{purpose}}.
Scope: Sources from {{year_from}} onwards, region {{region}}.
Output: A table with columns claim | source URL | date | confidence.
Rules: No claim without a source. Mark anything unverified as "UNVERIFIED".
Finally: List open questions you could not answer.

2. Check sources in a second pass

Don't ask for research and verification in the same prompt. Run a separate pass on the output:

Here is a list of claims with sources. For each one:
1. Does the source actually say this? Quote the relevant sentence.
2. Is the source primary (original data, official page) or secondary?
3. Is the date recent enough for {{purpose}}?
Return: claim | supported yes/no/partly | quote | source type | note.

Then open the important links yourself. A model can describe a page it never read.

3. Summarize without losing nuance

Summarize the text below for {{audience}} in {{n}} bullet points.
Keep numbers, dates and names exactly as written.
Add a final line "Not covered:" with relevant points the text leaves out.

Asking for "Not covered" stops summaries from quietly sounding more complete than the source.

4. Use variables and keep prompts in files

Placeholders like {{topic}} make prompts reusable. Store them as Markdown for reading and as JSON for scripts, and fill the variables with a small script instead of editing by hand each time.

Shortcut: the Botnexio Research Prompt Pack

The Botnexio Research Prompt Pack – Markdown + JSON includes, according to its product page:

  • 4 prompts for research
  • 4 prompts for source checking
  • 4 prompts for summarizing
  • Formats: Markdown and JSON, plus a small script to fill in variables (and one to check that JSON and Markdown match)
  • Designed for common chat AIs such as ChatGPT or Claude; results depend on the model

Not included: software or AI access – you need your own AI tool. AI output can be wrong; always check sources yourself. Delivery: ZIP download after payment.

12,00 € – Sold by Aramac GmbH · Price incl. VAT
Personal, non-commercial use only, for consumers only – see the licence.

FAQ

Which AI tools are the prompts designed for? They are designed for common chat AIs such as ChatGPT or Claude; results depend on the model. You need your own AI tool – the pack contains no AI access.

Can AI do reliable research on its own? No. Treat output as a draft and check the key sources yourself.

Why separate research and source checking? A second pass with a different instruction catches claims the first pass didn't back up.

Why Markdown and JSON? Markdown is good for reading and copying; JSON is good for scripts and automation.

How many prompts are in the pack? 12: four each for research, source checking and summarizing.

Related guides

ChatGPT is a trademark of OpenAI; Claude is a trademark of Anthropic, PBC. The names are used descriptively only. This product is not affiliated with or endorsed by either company.