Research & Summarisation Assistant
The prompt system that turns any chatbot into a disciplined research analyst. You get a four-stage prompt chain — research, source-verification, summarise, cite — plus output templates and a baked-in anti-hallucination guardrail that forces the model to state uncertainty instead of inventing sources. Paste it into your own ChatGPT, Claude or Perplexity account and run it on your own key. You own the prompts outright; we host nothing, store none of your research, and see none of your queries.
This is a set of prompts, not a piece of software. For A$59 one-off you get a Markdown download holding a four-stage chain — research, verify the sources, summarise, cite — which you paste into your own ChatGPT, Claude or Perplexity account and run there. Nothing installs. It exists because a general-purpose model asked to research something answers with total confidence, references and all, and some of those references may not exist.
The chain arrives tailored rather than generic: the research prompt is written around the kind of questions your work actually throws up, and the output templates come in the shape you nominate — Brief, Table or Memo. That tailoring takes about ten minutes, after which the pack lands on your order page. Three free AI revisions cover the case where the wording doesn't sit right against your subject.
Why verification runs as its own pass
The usual approach is one prompt doing all four jobs at once. The chain splits them: a research prompt that gathers, a source-verification sub-prompt that pressure-tests every claim before it lands, a summarise prompt that compresses without dropping the caveats, and a cite prompt that attaches attribution at the end rather than improvising it along the way. Separating verification from gathering is the part that does the work. A model asked to find and confirm in the same breath is inclined to confirm whatever it has just found.
Sitting behind that pass is the honesty guardrail — a block of prompt instruction requiring the model to state uncertainty rather than fill a gap. Where it cannot stand behind a source, it is told to say so instead of producing a plausible-looking reference. A plain-English run guide sets out which prompt to paste when, on each of the three platforms.
What this is not: a fact-checking system. It lowers the odds of a fabricated citation and builds in a step most people skip, but nothing in the pack connects to a database, opens a URL or confirms anything on your behalf. Every source the model hands back still needs your own eyes on it. Treat the result as a well-organised first pass, not a verified record.
Where the pack lives after delivery
On your machine and in your own chat window. Once the download arrives, nothing runs on our infrastructure — included is a data-safety note on using your own key and your own account, and the practical effect is that we host nothing, store none of your research and see none of your queries.
You own the prompts outright, so edit them or extend the chain as the work demands. Your only ongoing cost is whatever your chosen chatbot already costs you. The A$59 is the sole payment to ECTD, with no subscription attached.
What you get
- The 4-stage research prompt chain (research → verify → summarise → cite), paste-ready
- A source-verification sub-prompt that pressure-tests every claim before it lands
- Output-format templates for your chosen format (Brief / Table / Memo)
- The anti-hallucination honesty guardrail — "state uncertainty, never invent sources"
- A plain-English run guide: which prompt to paste when, on ChatGPT / Claude / Perplexity
- A "paste your OWN key here" data-safety note so your research stays yours
- 3 free AI revisions included
Turnaround: ~10 minutes. Delivery: Prompt-System Pack — Markdown download (paste into ChatGPT/Claude/Perplexity).
Frequently asked questions
I already ask the chatbot to cite its sources. What does the chain add?
An instruction to cite sources tends to produce citations rather than accuracy — the model completes the pattern it was asked for. The chain separates the steps so verification happens as its own pass with its own sub-prompt, before summarising and before any citation is written. The guardrail then instructs the model to flag what it cannot support rather than paper over it.
Can I point it at legal, medical or financial questions?
You can point it at any subject, but the output is AI-generated general information and is not advice. In regulated areas, use it to organise a first pass and to surface what needs checking, then have a qualified professional review anything you intend to rely on. The guardrail reduces invented sources; it does not make the model an authority.
Which account do I need, and will a free tier cope?
An account on at least one of ChatGPT, Claude or Perplexity, run on your own account or key — which of the three you pick is your call, and the run guide covers all of them. The prompts are plain text, so they will paste into a free tier, though free tiers generally cap message volume and context length and a long research session is where you would notice. Whether your session can reach live web results depends on the model and mode you run it in; the pack itself has no connection to anything.
Included in Creator Content Pack — save with a bundle.