Careful research is not one act; it is five kinds of work that build on each other. You frame a topic into a question worth answering. You gather material and work out what can be trusted. You analyse it into findings that are not sitting in any single source. You conclude, testing the conclusion before you rely on it. Then you report, turning the finding into something other people will read. Skip one and the others suffer: gathering without framing wastes effort on the wrong question; concluding without analysis is just the largest pile of evidence winning; concluding without reporting leaves the work where no one can use it.
Each step is something a good researcher does by habit and an AI does not do on its own. This page explains them. The prompts that carry them out are in manual prompts. Underneath the steps run a set of standards, each applying where it is relevant.
This step turns a topic into a question you can actually research, before any searching starts.
An AI will not question a badly framed topic; it answers what you give it. A researcher works differently. Handed a broad brief, they first ask what decision the research is meant to inform and what a useful answer would look like. Often the question worth answering turns out to sit just to one side of the one that was asked.
This step works best as a conversation at the start, because the reframing usually needs a decision from you, not just an answer from the AI.
This step collects material and works out what in it can be trusted. The two happen together, one item at a time, not in separate passes.
This is where an AI on its own falls furthest short: it reads the top few results and stops. The methods here keep it chasing details to sources that can stand behind them, and recording what has and has not been confirmed.
This step produces findings that are not sitting in any single source. It is what separates real analysis from a summary.
An AI does none of it on its own; asked to analyse, it gives you a summary with commentary attached. The main move is combination: putting two things side by side and reading what comes out.
This step tests the finding and puts it in a form you can trust.
The difference between an adequate conclusion and a sound one is mostly here: stating only what was actually confirmed, and actively looking for the evidence that would prove the conclusion wrong before relying on it.
This step turns finished research into something other people will actually read.
Concluding and reporting are different work. Concluding settles what is true; reporting composes it for a reader who was not in the research. The step produces three layers, a top post that earns attention, a thread that earns the click, and a document that holds the depth, each carrying only its own job.
The three layers are the same finding at three depths, not three different takes. What varies with the topic is how the content inside each layer is arranged.
Standards check the quality of each step, because every step builds on the ones before it. Each one applies where it is relevant: sources are ranked where they are gathered, effort is matched where claims are made, the record is kept throughout, and a late finding can send the work back.