Good habits

The people who get the best answers from a connected brain share a few habits. None of them is technical. This page collects them, and the handful of things that quietly make a connection worse.

Asking

• Ask about your work, not about the world. The brain holds what your company knows. "What is our onboarding process?" is a brain question. "What is an onboarding process?" is not, and the instructions block tells the AI to skip the brain for it.
• Name the department or the entity when you know it. "In the finance hub, what is our invoicing cadence?" lands faster than "how do we invoice?". The brain is organized by department, and the AI navigates it rather than searching a pile of text.
• Ask for the page. "Where is that written?" turns a plausible answer into a verifiable one. A good answer carries a path like wiki/processes/p-invoicing.md.
• Ask for what changed. "What changed in product this week?" uses the change history, which is authoritative: who did what, when.
• Say "check the brain first" when the AI answers from memory. Once. If it keeps happening, the instructions block is missing from that client.

Correcting

• Correct durably, out loud. "That is wrong, X moved to Y in August, update the brain" is a durable correction. "Hmm, not sure about that" is not, and the AI is told to treat it that way.
• Let it fold the correction in. With the instructions block, the AI writes the fix into the page and reports one line. Read the line; it names the page.
• Review what it filed on a company brain. Depending on department settings, changes wait for a reviewer. Owners and admins see them under Review in the dashboard. Confirm or undo; every write is reversible.

Keeping the connection healthy

• One block, all brains. If you have several brains, the dashboard generates one instructions block with routing lines. Paste that one and delete any older block.
• Keep the connector on. Turning it off "to make the AI faster" is the most common way a brain goes unused. The negative clause in the block already keeps the AI from over-using it.
• Re-sign in when asked. Sessions expire; clients show a login button or an authentication prompt. It takes ten seconds and nothing else changes.
• Refresh tools after an upgrade. When the brain gains a tool, ChatGPT and some editors need a refresh or restart to list it. The dashboard announces upgrades under Review.
• Use skills for anything you do twice. If you find yourself pasting the same format or procedure, ask the AI to save it as a skill. The next run is one line.

What not to do

• Do not paste secrets into a chat that writes to the brain. API keys, passwords and personal identifiers do not belong in pages. Company brains can hold sensitive-looking writes for review; do not rely on it as a filter.
• Do not bulk-copy another system into the brain. The brain links to your CRM, tracker or docs through system cards and reads them through your own connector. Copying their contents in creates two stale versions.
• Do not add to the instructions block. It is short on purpose. Tool lists, protocols and "always use the brain" make it worse, not better. Why.
• Do not share a connector. Each person connects with their own account and sees exactly what they are allowed to. A shared login turns per-person access into nobody's access.
• Do not treat a blank answer as a failure. "The brain has nothing on that" is the correct answer when nothing is written. It is a prompt to write the page, not a bug.

A weekly rhythm that works

More on keeping the brain itself healthy: Best practices and Health checks.