Schematic preview — the chain and its handoffs, drawn as a diagram rather than shown as prompt text.
01 What's inside
What's inside
01client-research-prompts.mdMarkdownThe library: prompts across five chained stages, each with its named variables, one filled-in example call and the expected output shape. (40 prompts)~62 KB
02input-checklist-and-output-rubric.mdMarkdownWhat you need in front of you before you start, and a nine-point rubric for judging the output before it reaches a client. (9 rubric points)~18 KB
03failure-modes-and-model-notes.mdMarkdownSeven named failure modes with the change to make in each, plus notes on what tends to differ between assistants. (7 failure modes)~16 KB
04prompt-library-notion-link.pdfNotionDuplicate link for the same library as a Notion database, one prompt per row, with stage and variable properties for filtering.~90 KB
4 files
02 Specification
Specification
File formats
Markdown, Notion
File size
~186 KB total
What you need
Requires your own account with an AI assistant, and a free Notion account if you want the database version. Outputs depend on a third-party AI service we do not control — you need your own access or subscription to it, and model behaviour and provider terms can change.
Compatibility
Plain Markdown files, readable in any text editor and pasteable into any assistant. No application, no installer, no API key.
Variables use double-brace placeholders and are listed at the top of each prompt, so a find-and-replace fills a whole stage at once.
Written without model-specific syntax. Steps that need a long context window are marked as such.
The Notion version is a duplicate link: one prompt per row, filterable by stage, on the free plan.
Contents are in English, and the prompts instruct in English; asking for output in another language works but is not something we have specified.
Download link on screen and by email as soon as your payment is confirmed. See how buying works.
03 Suited to / not suited to
Who this is for
Suited to
Consultants and agencies preparing for a first meeting with a company they do not know
Freelancers who want their discovery call to sound informed rather than exploratory
Anyone who already uses an assistant for research and gets shapeless answers back
Not suited to
Anyone expecting verified facts — an assistant produces plausible text, and the rubric exists because of that
People who want a one-click tool; this is a routine you run, and it takes an afternoon the first time
Note
Running these prompts requires your own account with an AI assistant, a separate service we have no connection to and which may charge you. Output varies between runs and between models, and no particular result is promised.
04 More previews
More previews
Schematic preview — the index of all 40 prompts, titles only.
Because the files are made available immediately at your request, the 14-day withdrawal right ends then. Your rights if a file is faulty are separate and last two years. See Refunds & Withdrawal.
Download link on screen and by email as soon as your payment is confirmed. See how buying works.
07 Version and changes
Version and changes
Version 1.0Initial release.2026-07-25
08 Questions about this product
Questions about this product
Which assistant do I need?
Any current general-purpose one. The prompts avoid model-specific syntax and the notes flag the two stages that want a long context window. The account is yours to arrange and may cost money; we have no connection to any provider.
Will it find facts about a company?
It will produce well-organised text that reads like facts. Whether it is true depends on the assistant, its access to current sources and the input you give it. The rubric is there precisely because that gap is real.
Can two prompts be run at once?
Within a stage, yes. Across stages, no — stage three expects a specific block from stage two, and running them out of order is the most common way the chain degrades.
How is this different from a list of prompts I could write myself?
You could write it. What is here is the sequencing, the variable naming, the documented handoffs, the rubric and the failure modes — roughly the part that takes weeks of use to arrive at.
Will it still work after a model update?
Prompts do drift as models change. These are written as plain instructions with explicit output shapes, which tends to age better than syntax tricks, but we do not promise future compatibility and there is no update service.