How to Set Up a Character.AI Lorebook and Test Keyword Triggers

1. Open an empty Lorebook

Select Create > Lorebook > + New empty lorebook. Expected result: an editable blank workspace.

Follow the official Lorebooks guide. If the option is missing, stop: do not infer that upgrading will unlock it. Ask official support whether your account is eligible, identifying your platform and missing control without sharing credentials.

Use a disposable fictional world for this first attempt. Starting small makes a wrong answer easier to investigate than importing a large setting with overlapping names.

2. Add one fictional entry and specific keywords

Choose + Add a new entry. Expected result: separate title, content and keyword fields populated for one location.

Use this original example with fictional adult characters. In the content field, enter the whole description below; enter each keyword separately using the editor's controls. Up to eight keywords are documented; two are sufficient here.

The entry-writing guide favors actionable constraints. This example gives the response a concrete object and a behavior to follow, rather than asking the model to remember a vague atmosphere.

3. Choose visibility before linking

Open Lorebook Settings. Expected result: an intentional visibility choice, not an accidental public release.

Private is subscriber-only and restricted to your private Characters. Public is discoverable. For public entries, Hint Only hides content, not titles or keywords; it does not change chat behavior. Leave Let others link their Characters off for this isolated test.

If your intended privacy setup is unavailable, stop instead of publishing publicly to proceed. Use invented names throughout; a setting label is not a reason to upload unpublished personal correspondence or somebody else's private history.

4. Link your Character and publish

Select Linked Characters > Add Character, choose one you own, then Publish. Expected result to verify after review: published lore attached to the intended Character.

Check the selected identity carefully if you maintain similar Characters. Use a private test Character for private lore. Autosaving preserves a draft, but drafts cannot be used in chat. Do not treat returning to a saved workspace as completion.

Confirm publication after review before testing; selecting Publish alone does not establish that lore is live. Pause if publication is unconfirmed or the selected Character is wrong. Record the unfinished checkpoint without copying chat contents into a support request.

5. Run a harmless keyword test

Expected result: the reply reflects the bell or waiting rule; a mismatch remains a failed or inconclusive sample.

Suggested test prompt: We arrive at Glass Harbor Observatory. Describe the entrance and what visitors should do before entering. Do not put the bell color, ownership or waiting rule in the prompt: supplying the answers would weaken the check.

Read the next reply for the distinctive green brass bell and waiting outside. Record correct, partial, contradictory or absent details. Mentioning an observatory alone is weak evidence because that setting is already in the prompt.

A keyword can occur in your message or the Character's. You can try the second phrase in a fresh chat, but these example phrases overlap: results cannot identify which keyword matched. Avoid teaching the answer through corrections. This suggested test is not an activation diagnostic; replies cannot prove activation, permanent memory or future consistency, or identify a failure's cause.

6. Publish revisions with Update

After editing published lore, select Update. Expected result: your intended revision becomes the published version.

For a controlled second pass, change only the fictional bell color. Keep a note of the intended version and repeat the same prompt in a fresh conversation. Existing chat text can otherwise contaminate your interpretation. Discard changes retains the previous publication instead of releasing edits.

Troubleshoot limits and keep Definition separate

Investigate one variable at a time; do not rebuild your Character around one inconsistent reply.

Recheck the selected account, Character, published version and keyword spelling before expanding the example. Remove competing test facts from your own prompt. The cited guides do not establish a complete matching algorithm, case-sensitivity contract, activation log or numeric entry-capacity specification. Do not invent those limits or promise a trigger.

Use Character Definition for core identity, voice and behavioral instructions; the field supports up to 32,000 characters. Prioritize essentials instead of filling the allowance. If Lorebooks are unavailable, keep a short essential setting there and retain the full world offline; this is not equivalent keyword retrieval.

For shared lore, {{char}} means the Character in the current chat, not a fixed owner. Keep Mara Vale explicitly named as observatory owner; writing {{char}} owns it would assign ownership to each linked Character.

Privacy, payment and when to stop

Stop for unclear visibility, unresolved access or repeated contradictions; paying is not a reliability fix.

Keep real addresses, passwords, financial data and identifying relationship details out of entries and prompts. These three feature guides are not a privacy-policy audit or a confidentiality guarantee. Review current provider privacy and deletion terms separately before sharing sensitive information.

No subscription price is asserted here. Before any independent purchase, verify account eligibility, displayed total, renewal interval, cancellation and refund terms. Ask support about missing controls without buying another plan. If the test remains unreliable, pause and use your offline world notes rather than escalating spending.

Setup summary

Finish with a small published example, a recorded test result and a clear stopping point.

Retain only details your own checks support. Add another entry after the first example is understandable, and repeat the same checks rather than assuming a larger world will behave consistently.

Frequently asked questions