Methodology
Authority built for AI answers.
Recala is an Authority OS: a per-brand engine that turns a company's knowledge, live research, grounded writing, and automated checks into a public body of work that builds over time. You can run it yourself from your own account, or ask our team to run it for you.
A model can only repeat what is already on the web. We start somewhere else: with what your business already knows, from your site, your documents, and your people. The aim is original, dated, sourced findings an answer engine can check and quote. What we publish is public and citable; the material behind it stays yours.
From rankings to citations
Search is moving into AI answers, and fewer readers click through to a source. Being found increasingly means being quoted in those answers, so we write pages an answer engine can check and quote. No one can guarantee a citation; we build for the parts you control.
The World of Knowledge
Every brand has a private universe of proof: product data, customer stories, internal research, founder opinions, case data, category vocabulary, and documents. Recala structures that material into a source-tagged graph that every article can stand on.
The Recala Loop
- 01
Ingest brand sites, product data, CMS history, PDFs, docs, voice, and author identity.
- 02
Structure that material into a queryable World of Knowledge with chunks, embeddings, entities, and source tags.
- 03
Read the live search battlefield before writing: SERPs, competitor sections, semantic gaps, and source quality.
- 04
Write something new from the brand's own evidence, in the brand's voice.
- 05
Run automated checks on claims, links, citations, schema, and originality, and record what they find with the article.
- 06
Feed every published article back into the knowledge graph, so the next article builds on it.
Research before writing
Before a draft begins, we check the competing pages, the questions readers are actually asking, the quality of the sources in play, and the brand's own knowledge. The research decides what the article needs to cover.
New information, not a rewrite
Before writing, we map what competing pages already say and where they stop. The article is built to add what they're missing, drawn from what the brand can show with its own evidence.
It sounds like your brand
We write in your voice, with your opinions, your audience, and your evidence. Each piece is published under a named author on your side, not ours.
Every article is checked
Figures, citations, internal links, schema, and originality go through automated checks on every article, and what they find is recorded with it. Statements about your company without a source are taken out while the article is written.
What we are building toward
Principles behind the engine
These are the rules we are building Recala to follow. Working from your own data is not part of the monthly plan yet: today's articles are built from your site, the documents you share, your people, and public sources.
Findings come from your own data
The goal is to start with the numbers your business already has and compute an original finding from them: a figure for a specific segment, measured against a public benchmark, with the date it was true. That is the kind of original statistic a rewritten blog post never has.
We compute the number, then write the sentence
Figures should be calculated in a separate step, before any writing happens, so the language model only phrases what was already computed instead of inventing a number. That rule targets the most common way AI content goes wrong.
Every number is checked against its source
Each figure in an article is matched back to the source or data that supports it, and what the check finds is shown with the draft. A figure that doesn't check out is corrected or taken out; we would rather drop a number than publish one we can't stand behind.
Your findings are easy to verify
Each finding should be published with its method, segment, sample size, and date, so a reader, or an answer engine, can see exactly how it was reached. What stays private is the raw data itself.
Your data is read through one narrow door
When we work from your data, the plan is to read it through a single, read-only connection that returns only summaries, never individual records, and only when a group is large enough that no one customer can be identified. Your raw records stay with you.
We take a position, and we keep it current
We don't write "it depends." Each piece says who an option is right for, on evidence, and where it falls short. Every finding carries the date it was measured. Updating older articles as the numbers change is on our roadmap; the plan today covers new articles.
How we hold ourselves to this
We measure our own work the same way we'd measure anyone's: does an article say something new, does every number check out, and is it useful to the reader. Automated checks run on every piece and you see what they found; you decide what publishes. The standard is the product, and we'd rather show you the standard than a number we can't back.
Each article makes the next one better
What we publish becomes part of what the next piece is built on, so your library gets stronger as it grows. That's the difference between churning out volume and building authority that lasts.