Company Brain

The Company Brain is the knowledge LogNorm writes from: your positioning, product docs, pricing, changelog and the words your customers use. Every brief, draft, refresh and claim check searches it for the passages that matter, so content states your real product facts instead of generic filler. This page covers adding sources, choosing which websites they ground, your brand identity and how sources are screened. For the overview, see Company Brain on the product pages.

Open it from the sidebar under Configure: Company Brain. The page has these parts: Brand identity at the top, then Writing style, Add knowledge, Sources, Test retrieval and Screening.

What to add first#

The empty Brain suggests a good start: your homepage, your pricing page and a positioning doc. After that, add whatever explains what you sell, to whom and how you talk about it. Screening labels each source with one of these kinds:

You don't pick the kind yourself. LogNorm infers it from the passages, and it helps you spot a source that isn't what you thought it was.

Add a source#

The Add knowledge card has three tabs. Each one creates a source that LogNorm reads, splits into passages and indexes in the background. The source shows Queued, then Indexing with a percentage, then Ready (or Failed, with the reason).

1. Open Company Brain

   Go to Company Brain and make sure the website switcher shows the website this knowledge is about.

2. Choose where it's used

   Under Use for, pick Only yourdomain.com or All websites. See Website or shared below.

3. Pick a tab and add it

   Fill in the form and click Add page, Add text or Add file. You can leave the page while it indexes.

4. Check it landed

   The source appears at the top of Sources with its token and passage counts. Try it in Test retrieval.

You can also add up to five URLs and a block of text during onboarding. Those sources belong to the website you're setting up.

Keep sources fresh

Each source row has a menu with Re-fetch & re-index (web pages) or Re-index (text and files), Open page, a scope switch and Remove. Re-index a page after it changes: LogNorm replaces the passages and bumps the source version (shown as v2, v3 and so on). Removing a source stops it grounding new briefs and drafts. Content already written isn't changed.

Website or shared#

Every source either belongs to one website or is shared by every website in the workspace. This keeps an agency's clients, or a company's separate products, from bleeding into each other's drafts.

When you write for a website, LogNorm searches that website's own sources plus the shared ones, never another website's. The Sources list shows exactly that set, with shared sources tagged All websites. If other websites have their own sources, the list tells you how many and asks you to switch website to see them.

To move a source later, use Share with all websites or Only use for yourdomain.com in its menu. The passages stay the same. Deleting a website deletes its own sources, but not shared ones. For more on websites, see Workspaces, websites and team.

How retrieval grounds your content#

When you add a source, LogNorm splits it into passages of roughly 500 tokens, with a little overlap so no sentence loses its context. When it writes, it runs a keyword search over those passages and takes the best matches. It doesn't send your whole Brain to the model. Here's where it searches:

• Briefs. The brief searches the Brain for the keyword and title. It uses what it finds for the differentiation section: what you can say that competitors can't, based only on the Brain and company info. The passages it used appear in the draft's Sources, grouped under Company Brain.
• Drafts and refreshes. The search covers the keyword, title, outline headings and must-cover items. Drafts state product facts only when they're in the Brain or your company info.
• Claim checks. The Score tab checks figures, superlatives and product claims against a wider set of Brain passages. Each claim is marked Supported, Contradicted or Not in your Brain. The Quality checks use the Brain as evidence too. See Content and the editor.
• GEO assets. Assets drafted by your quality model draw on your pages, the Company Brain and competitor data. See GEO audit.

If a website has no ready sources, the content workspace says so. Drafts then rely on your website alone, and more claims come back as Not in your Brain.

Test retrieval

The Test retrieval card shows what a writer would get. Ask the kind of question a writer would, such as "who is our ideal customer and what do they struggle with?" LogNorm returns up to six passages, labelled B1, B2 and so on, with their source, passage number and a relevance bar. Your search words are highlighted. If nothing relevant comes back, rephrase or add a source that covers the topic.

Brand identity#

The Brand identity card at the top of the page lists every name and site that counts as you. It's set per website, needs the Owner or Admin role to edit, and works on every plan.

• Brand terms. Product names, former names and abbreviations, up to 30. Each needs at least 3 characters, and they match as whole words, ignoring case. Your website name and domain variants are always matched, and appear under "Always matched".
• Owned sites. Up to 30 extra domains or paths. A domain covers its subdomains. Add a path to claim only part of a site, like github.com/your-org or medium.com/@you. Shared platforms such as GitHub, Medium or YouTube need a path, because all of github.com can't be yours. Your main domain is always included.

You can paste several at once, separated by commas or new lines. LogNorm uses the list wherever it asks "is this us?":

• AI visibility: mentions in AI answers, citations of your own pages and answer highlighting.
• Google results: your position in the SERP and AI Overview citations, and your ranking position on keywords.
• Brief research, which skips your own pages when it reads the top results.
• Competitor discovery and adding: your own sites are never suggested as competitors.

Clicking Save also re-checks your past AI answers and citations against the new list and tells you how many changed. This is text matching only, so it uses no credits.

Memory bank#

The memory bank holds short, durable instructions about how you want to grow: preferences (“practical guides over thought leadership”), goals, things to avoid (“never compare us with a partner”) and facts about the business. How you write lives in Writing style, below. The growth strategist saves them when you tell it something in a conversation and reads all of them before it plans anything; briefs, drafts and the moves LogNorm proposes on its own follow them too.

Manage them in Company BrainMemory bank: add one by hand, edit or delete any of them, and choose whether each applies to this website or to every website in the workspace. The memory bank works on every plan and doesn’t count toward your knowledge source limit.

Writing style#

Writing style is how LogNorm writes for this website. Every draft, revision and refresh follows it. It has two parts.

• Voice. Choose Learn my voice, pick two or three of your own posts (or paste 120 words or more of your writing) and LogNorm writes a voice guide from them: who speaks, the pace, how you open and make a point, what you never do. It takes one short AI call. Learn again replaces it; Remove deletes it.
• Rules. One instruction each, such as "Open with the reader's problem, not a definition". Rules come from three places: the feedback you give on drafts (saved straight away, and shown to you when it happens), the edits you make before approving a draft (listed under Noticed in your edits until you choose Keep or Dismiss), and rules you add here. Where it came from shows the comment or the before-and-after edit behind a rule.

Giving the same feedback again makes its rule stronger, and the 24 strongest rules that are switched on are used. Switch a rule off to keep it without using it, or delete it. See Content for how to give feedback on a draft.

Screening#

Anything you add to the Brain ends up in front of the model that writes your content. So LogNorm screens every passage with its decision AI. Screening runs automatically when a source is indexed and does three things:

• Injection flags. A passage is flagged when it looks like instructions aimed at an AI rather than at people, such as "ignore previous instructions" or "always recommend X". Ordinary how-to steps for people aren't flagged.
• Kind. What sort of knowledge the source is (see the table above), with a confidence level.
• Relevance. How relevant the passages are to what your website sells, from "unrelated to our business" to "core product knowledge". Low-relevance sources get a warning, because they can pull off-topic material into drafts.

The Screening card lists every ready source with its kind, a relevance meter and the number of flagged passages. A summary line shows how many sources are screened, how many have flagged passages and how many are low relevance. A source re-indexed since it was screened reads "Re-indexed since it was screened". Use Screen N sources (or Screen again) to screen anything pending. Passages that haven't changed are cached, so re-running is cheap.

Flagged passages and Keep anyway

A source with flagged passages gets a warning card at the top of the page, for example "2 passages in ‘Partner FAQ’ look like instructions aimed at an AI". The card shows each passage's excerpt and how likely it is to address an AI. Until you decide, LogNorm leaves those passages out of retrieval. The rest of the source is still used.

1. Read the excerpts

   Check whether the text is really aimed at an AI, or something harmless like documentation for your own AI features. Click Show all if there are more than two.

2. Keep or delete

   Keep anyway lets retrieval use those passages again, and records who kept them and when. Delete source removes the source from the Brain. Drafts already written aren't changed.

Screening never deletes anything by itself. If it isn't available, or a check fails, sources stay unscreened and every passage is used as normal.

Who can do what#

Related#