01
A plan before an answer
Every engagement starts by decomposing the research question into a structured plan. The plan makes the scope of the work visible before any of it is done, and gives the user something concrete to adjust.
Case study · 03LIVE
Research that starts with a plan, not a prompt.
AI Research AssistantAI · SaaS · Productivity
An AI research assistant that turns an open-ended question into a structured research plan, then works through retrieval workflows to produce organised, referenced findings instead of a single unverifiable answer.
Visit AI Research Assistant- Role
- Founder / Product / UX / AI implementation
- Product
- AI research and planning SaaS
- Users
- People doing structured, multi-source research
- Status
- Live
- Year
- 2026
The problem
One prompt rarely produces real research.
Asking a language model a complex question produces a fluent answer, but fluency is not the same as rigour. Real research means decomposing a question, checking multiple sources, tracking what has been established and knowing what is still open.
A single question-and-answer exchange does none of that. It hides the steps where quality is actually decided.
Single-prompt answering
- Question
- Answer
Structured research
- Question
- Research plan
- Source retrieval
- Findings
- Synthesis
- Open questions
The insight
The valuable part of research is the plan. The AI should produce that first.
The central product decision was to separate planning from execution. Before retrieving anything, the assistant turns the question into an explicit research plan — the sub-questions, the order of work and what a good answer needs to cover — so the rest of the workflow has structure to follow instead of improvising from a single prompt.
- Research question
- Research plan
- Retrieval workflows
- Findings
- Synthesis
The product
02
Retrieval as a workflow, not a lookup
Research steps run as retrieval workflows rather than one-off queries, so information gathering follows the plan and findings accumulate in an organised way instead of arriving as disconnected fragments.
03
Findings you can navigate
Results are organised and referenced rather than delivered as a single block of generated text, so the user can see what the research established and follow it back to its sources.
Product decisions
Decisions that shape the research experience.
D01
Plan-first, not answer-first
Making the research plan explicit keeps the work structured and lets the user steer the direction before effort is spent.
Trade-off
Slower to a first visible result than a single-prompt answer, in exchange for research that holds together.
D02
Workflows over one-shot generation
Breaking retrieval into steps means each part of the research is deliberate and inspectable rather than buried inside one generation.
Trade-off
More moving parts in the product, in exchange for a process the user can actually follow.
D03
Structure over fluency
Organised, referenced findings are more useful than a beautifully written answer whose basis cannot be checked.
Trade-off
The output reads more like research notes than an essay, which is the point.
How it works
More than a chat window with sources.
The interface is the surface. Underneath it, a question is first converted into a research plan, and that plan drives the retrieval workflows that gather and organise findings before anything is synthesised for the user.
The point of the architecture is that the AI's behaviour is directed by an explicit plan rather than left to improvise around a prompt.
- Research question
- Planning
- Retrieval workflows
- Organised findings
- Synthesis
The assistant works from
- An explicit research plan
- Retrieval workflows
- Accumulated findings
Outcome
From a way of working to a working product.
LIVEThe AI Research Assistant is live and publicly available, with the core experience running end to end: from an open-ended research question to a structured plan, retrieval and organised findings.
Lessons
01
Structure beats scale for research tasks.
An explicit plan and organised workflow contribute more to output quality than a larger or more capable model answering in one shot.
02
Users trust what they can inspect.
Showing the plan and the organised findings makes the research legible; a single polished answer asks for trust it has not earned.
03
Slowing down the first answer improves the final one.
Spending the first step on planning rather than answering changes the character of everything that follows, for the better.
Have a question worth researching properly?
The Research Assistant is live and available to try.
Visit AI Research Assistant