How Bricks does its research

When you ask Bricks about a brick wall, your question is not handed to one AI in one go. A team of AI specialists works on it, each with a single job, and rules that are not AI check what they produce.

Why not just ask ChatGPT or Claude?

Bricks uses the same kind of AI you can talk to in a chat window. The difference is everything around it.

In a chat window	In Bricks
A chat window knows only what you type or paste into it. A detail you leave out is a detail it never considers.	Bricks reads the person straight from your family file: their events, parents, spouses and children, and your notes and sources if you choose.
It answers from memory, and memory is where invented archives and made-up web addresses come from.	The team looks things up instead, in reference tables: hundreds of real archives and record websites, when record-keeping began in many places, and where records were destroyed by fire or war.
It judges its own ideas.	A separate AI's whole job is to argue against every idea. The weak ones are marked down or thrown out.
Nothing tests its web links before you click them.	Every web address in the plan is opened and tested first.
You get a conversation, and turning it into a plan is up to you.	You get a numbered research plan — one record and one place per step, with what to do whether you find it or not — kept with the person so you can mark off each step as you go.

What the team does

Each has a single job.

The record reader. Reads what your family file holds about the person and their family and lays it out as a timeline, pointing out what is missing and what contradicts itself.
The historian. Works out what was happening in that place at that time: who kept records then, which of those records survive, and how names were written.
The interviewer. Writes down the questions only you can answer — a family story, an old letter or photograph, a name on a gravestone — that could change where to look, and puts them in your report.
The researcher. Proposes explanations for the brick wall and plans a search for each one: which record, where to find it, and what to do whether you find it or not.
The second opinion. A separate AI whose whole job is to argue against the researcher. It holds every idea to the standard professional genealogists use — has the search been thorough, is each fact backed by a source, do the records agree — and marks down or throws out the ideas that do not hold up.
The writer. Puts it together for you: a short answer first, then a case summary, and the research plan in the order to work through it.

Then, checks that are not AI

When the team has finished, fixed rules written by people check the report again — every time, the same way. Anything they are unsure of is marked in the report for you to confirm.

A reference to an archive box or manuscript that looks invented is marked for you to verify.
An archive the plan names that is not recognized is pointed out, so you can check it exists before you rely on it.
A step that sends you to the wrong archive for that area is flagged.
A place that cannot be found on a map is pointed out. Old places are often missing from modern maps, so this is a prompt to check, never proof of a mistake.

None of this makes it infallible. What you get is a set of well-reasoned places to look — a research plan, not proof. Look at the actual record with your own eyes before you add anything to your tree.

← Back to Bricks: what it does, what it costs, and how to get it