Character.AI Lorebook Ideas: Templates That Actually Work (2026)

On Sunday, September 27 I paid Character.AI $9.99 for a month I already plan to let lapse, for one reason: I wanted to know whether the Lorebook import box would swallow a plain text file. It did. Nine entries went in on a single paste. So what follows is the post I went looking for in July and never found, a set of Character.AI lorebook ideas written out as finished entries, in the exact format the import box accepts, ready to copy.

Nearly every lorebook tutorial, mine included, explains what a good entry does and then leaves you staring at an empty content field. That gap is the whole reason this post exists. Below are nine templates plus one small complete lorebook, twelve entries in total. Each one stands alone, uses eight keywords or fewer, and can go into Character.AI without you editing anything except the names.

If you need the mechanics first, they live in my Character.AI lorebook guide, written the weekend the feature launched.

Copy the Format Before You Copy the Ideas

Here is the choice that makes this post different from every other list of Character.AI lorebook ideas you'll find. Instead of describing entries in prose, I wrote all of them in the plain text layout Character.AI reads on import. You highlight a block, paste it into the import box, and the entry arrives with its title, keywords and content already in the right fields. Nothing to retype.

The format is four lines and two of them are optional:

Lorebook Name: Whatever you are calling this world
Lorebook Description: One line for anyone who browses it.

Entry Title: The first thing
Keywords: keyword one, keyword two, keyword three
Content: What the Character learns when one of those keywords shows up in chat.

Name and description go at the top once. Every entry after that starts with Entry Title:, then Keywords: separated by commas, then Content:, which can run over as many lines and paragraphs as you want. A new entry begins at the next Entry Title line. Character.AI documents all of this in its Importing Lorebooks article, along with two details I like: files up to 10 MB are accepted, and the whole thing is processed locally on your device.

Three fields per entry, that's the entire data model. Title, content, and up to eight keywords. The title is for you and for anyone browsing your lorebook, and your character never sees it, which I didn't register until my second lorebook. There is a second reason to draft in a text file rather than the editor: Character.AI has no lorebook export button, a gap I complained about at length while working out what the Disney hire means for people with worlds parked on the platform. Write outside the app and you keep the only copy that survives a lapsed subscription.

Nine Entries, One Paste

Back in July, while my old subscription was still live, I hand-copied the nine Briceport entries worth keeping out of Character.AI into a text file, one field at a time, because there's no export. Roughly 40 minutes of clicking between tabs. Briceport is the port town behind Sabine, my long-running detective, and that file is the only copy of her world I own. In August I rebuilt the same nine as World Info entries in SillyTavern on my laptop and kept adding keywords over there, which matters in a minute.

Sunday afternoon I reformatted the text file. Mostly mechanical work: adding Entry Title:, Keywords: and Content: in front of things I had originally written as a loose outline with dashes. That took about twenty minutes. Then I paid for the month, opened Import, pasted the whole file, and watched all nine entries land in the editor at once.

Two of them arrived broken, which was on me.

My July notes had separated some keywords with slashes instead of commas, so one entry came in with a single absurd keyword reading Briceport/the port/the docks/customs house. And for the Corwin entry I had pasted in the longer keyword list from the SillyTavern install on my laptop, where nothing stops you adding more. It had grown to eleven. Character.AI caps an entry at eight, so three had to go. Fixing both took under five minutes, and the entry that had been my worst keyword mistake in July, the one where I used the word watch and set the thing off on nearly every message, came through already corrected.

Monday, September 28, I did the second test: I exported the World Info book from SillyTavern as JSON and imported that instead, as a throwaway lorebook, to see what survived the trip. More on that below, because it's the part most people migrating between tools get wrong. If you're weighing the two platforms generally, I timed the whole switchover in Character.AI versus SillyTavern.

One change I made this time that I couldn't have made in July: I set Briceport to Private before publishing. In August my single real gripe about this feature was that everything you published sat on your profile for strangers to read. That has since been fixed for subscribers, and I'll come back to it.

Which of the Eight Entry Templates to Reach For

Two different things in Character.AI are called templates, and the overlap confuses people. When you create a lorebook you choose + New empty lorebook or a template that gives you a pre-built structure. Separately, inside the editor, under + Add a new entry, there are eight entry templates. Those eight are what this table is about.

The templates don't change how an entry behaves. Character.AI describes them as a starting point, and every one of them collapses into the same three fields when you save. I start from a blank entry most of the time, but thinking in the System / rules category improved how I write, because a rule with no stated cost isn't a rule yet.

Character.AI Lorebook Templates, Ready to Paste

Nine lorebook entry examples, from a restaurant kitchen to a town that gossips. They are deliberately not one connected world, because you probably want a pattern to copy rather than my setting. Swap the proper nouns, keep the shape. Each block is a complete entry in the import format, so you can paste one on its own or stack several under a Lorebook Name line.

1. The correction entry (write this one first)

Entry Title: Correction: Marit is a line cook, not the chef
Keywords: Marit, line cook, the kitchen, Fourth Street Grill, her job, head chef, promotion
Content: Marit works the grill station at the Fourth Street Grill. She does not run the kitchen, has no say over the menu, and does not decide who gets scheduled. When her job comes up she talks about her station, her hours, and the people above her, never about running the place. Never promote Marit, never give her staff to manage, and never describe the restaurant as hers.

Character.AI's own writing advice puts corrections first and I've come around to agreeing. Pick the detail the model keeps quietly upgrading. Job titles are the classic: give a character any workplace and the AI will promote her by message thirty. Notice that the last sentence is an instruction, not a fact, which is the single biggest upgrade available to any entry you write.

2. A side character written as behavior

Entry Title: Danut, the shift lead who never says no out loud
Keywords: Danut, shift lead, the schedule, cover my shift, swap shifts, Danut said
Content: Danut runs the evening schedule and has never refused a request directly. He says he will look at it, writes nothing down, and the schedule quietly stays as it was. Everyone under him has learned to ask twice and then ask in writing. When Danut is in a scene he is friendly, unhurried, and slightly too agreeable, and people leave the conversation unsure whether they got what they asked for. He is never openly hostile and never yields on the first ask.

Biography is the weakest thing you can put in a side character entry. What you want is how the person behaves when they walk into a scene, plus one thing they never do. Danut is a workplace character and the entry never states his age, his history or his appearance, none of which would change a single line of dialogue. If you like pushing behavior out of the model this way, the same instinct drives my Character.AI prompts guide.

3. A location that changes how people talk

Entry Title: The Sunderly Rooms
Keywords: Sunderly, Sunderly Rooms, residential hotel, front desk, room 41, the fire escape
Content: The Sunderly Rooms is a twelve-story residential hotel with monthly rates, one working elevator, and a front desk staffed by whoever owes the manager money that week. Sound travels: every conversation in room 41 can be heard in 39. Residents prop their doors open in summer because the windows are painted shut. Scenes in the Sunderly should feel overheard. Anyone talking in a hallway assumes at least one person is listening, and speaks accordingly.

Most location entries are travel-brochure copy and do nothing. This one spends half its words on a consequence: the place is acoustically leaky, so everyone in it is on guard. Give your location a property that constrains dialogue and it earns its keyword. Swap the hotel for a shared office, a barracks, a ward, whatever you have.

4. A faction with an actual power structure

Entry Title: The Tuesday Panel
Keywords: Tuesday Panel, the panel, Holloway, funding request, review meeting, next cycle
Content: The Tuesday Panel is four senior staff who decide which projects get money for the coming quarter. There is no application form and no published criteria. Holloway chairs it and has never been outvoted. Requests are not rejected, they are held for the next cycle, which means the same thing and keeps everyone polite. Anyone with business in front of the Panel speaks carefully in public and lobbies in private. Nobody says the word no in a Tuesday meeting.

Labels like influential or corrupt leave the model guessing. Answer four questions instead: who decides, who owes whom, what nobody says out loud, and what happens when someone pushes. This one is a workplace committee, but the same skeleton works for a royal court, a guild, or a group chat of eleven-year-olds.

5. A world rule with a price attached

Entry Title: RULE: what a favor from Quillon costs
Keywords: Quillon, the ledger, owe Quillon, ask Quillon, a favor from Quillon, Quillon's price
Content: Quillon can fix almost anything asked of him, and every fix goes into his ledger. Payment is never money. It is a memory, chosen by him, taken quietly, and the person who paid does not notice the gap until somebody else points at it. Debts are always collected, usually at the worst available moment. Never let Quillon solve a problem for free, and never let his help arrive without a cost surfacing later in the story. When his name is said out loud, he hears it.

Write rules as constraints, never as lore. A rule entry should say what can happen, what it costs, and what must follow. The reason this matters: without a stated cost, a language model will cheerfully use the supernatural thing as a clean solution and drain the tension out of your story. The last line is there because a small enforceable creepy detail beats a paragraph of atmosphere.

6. An item that carries suspicion

Entry Title: Kessler's field notebook
Keywords: Kessler, field notebook, green notebook, Kessler's notes, missing pages
Content: Kessler carried a green cloth notebook in his coat for nineteen years: names, dates, plate numbers, and one column of figures nobody has decoded. Pages 40 through 52 were cut out with a blade rather than torn. Whoever holds the notebook is assumed to know something, whether they do or not, and anyone who recognizes it changes the subject. Never explain the notebook in full. Each time it comes up, surface one more small detail and stop there.

Items are the template I use least, because most objects don't need a keyword. This one qualifies because holding it changes how other characters treat you. The instruction to reveal one detail at a time is doing real work, and it's the kind of pacing control that normally requires you to keep nudging in chat.

7. A {{user}} entry that gives you a reputation

Entry Title: What Ashgrove already thinks about {{user}}
Keywords: Ashgrove, Ashgrove Road, the post office, new in town, who are you, people talk
Content: Ashgrove is a town of eleven hundred where an arrival is noticed within a day. By the time {{user}} has been here a week, the post office, the garage and both churches have already worked out a version of where {{user}} came from and why. Nobody asks directly. People are warm, curious, and already slightly informed. {{char}} has heard at least one rumor about {{user}}, never repeats it unprompted, and lets it shape how carefully {{char}} speaks.

The {{user}} placeholder is the most underused thing in the whole feature. It resolves to whoever is chatting, persona and all, so an entry like this hands the reader a standing in your world before they type a word. Note the careful {{char}} usage: I only assert things that are true of every linked character. Write "{{char}} owns the garage" and link five characters, and all five of them believe they own the garage. If you write detailed personas, this pairs with the custom prompts I built for myself.

8. A study-buddy rule for language practice

Entry Title: RULE: how {{char}} corrects my Spanish
Keywords: Spanish, en espanol, corrigeme, conjugation, preterite, did I say that right, grammar
Content: When {{user}} writes in Spanish, {{char}} answers in Spanish first, then adds one short English note naming the single worst error and the corrected phrasing. One correction per reply, never a list. If a sentence is right, {{char}} says so in three words and moves on. {{char}} never switches the whole conversation into English because {{user}} struggled, and never waves a mistake through as close enough. Vocabulary stays at an intermediate level unless {{user}} asks for harder words.

Nothing says lorebook entries have to be fiction. This is a behavior contract for a practice partner, and it fixes the two things tutors-by-chatbot always get wrong: dumping six corrections at once, and switching to English the moment you struggle. Keywords here fire constantly by design, which is the one case where a broad trigger is correct. Swap Spanish for whatever you're learning and keep the one-correction rule. The same logic shows up in my notes on prompting techniques nobody writes down.

9. An event people date things by

Entry Title: The Everly Street fire, March 2019
Keywords: Everly Street, the fire, March 2019, the warehouse, the insurer, arson
Content: A furniture warehouse on Everly Street burned on a Thursday night in March 2019. Two people were inside and both got out. The fire marshal called the cause undetermined, and the insurer paid in eleven days, which people here found interesting. Nobody says the word arson out loud. The fire is how the town dates everything else: before the fire, after the fire. Any character who lived here then has an opinion and will not hand it to a stranger.

Here is my test for whether a piece of history deserves an entry: do characters measure other things against it? If yes, write it. If it's only backstory, leave it in the character definition. Novelists make the same call every time they decide which backstory reaches the page, which is roughly the argument I make in Character.AI for writers.

A Whole Small World in One Block

I closed the August guide by promising a second lorebook in a completely different genre. Here it is, built across Sunday and Monday while the subscription was fresh. Sabine's world is rain and harbor politics, so I went the other direction: a quiet, cramped, unglamorous science fiction setup with exactly one person in it. Name, description and three entries, all pasteable as a single block.

Lorebook Name: The Leaward Run
Lorebook Description: A slow cargo hauler on a six-week loop, one crew member aboard, and a station that answers seven minutes late.

Entry Title: The Leaward
Keywords: Leaward, the hauler, cargo bay, hab ring, six-week loop
Content: The Leaward is a forty-year-old cargo hauler working a six-week loop between Bellweather Station and the outer refineries. Four holds, one habitable ring, no medical bay beyond a wall kit. Two crew berths, one of them occupied. Everything aboard is loud: the recyclers, the ring bearings, the fans. Silence on the Leaward means something has stopped working, and Ilka hears the silence before she hears the noise.

Entry Title: Pell, the dispatcher at Bellweather
Keywords: Pell, Bellweather, dispatch, the log, Pell said
Content: Pell has handled the Leaward's dispatch for eleven years and has never met Ilka in person. He is dry, faintly formal, and pads every message with the weather at the station because he knows she wants something ordinary to read. He files her hours generously. Pell will not lie to the company, but he is very good at choosing which sentence goes in the log. When he stops joking, something is wrong upstream.

Entry Title: RULE: the seven-minute lag
Keywords: Bellweather, dispatch, the lag, comms, ask Pell, radio
Content: Messages between the Leaward and Bellweather Station take seven minutes each way. Live conversation is impossible. Ilka talks in recorded blocks and gets answers to questions she has already stopped caring about. Every decision on this ship is made alone and defended afterward. Never write Pell replying immediately, and never let Ilka ask permission when the situation needs a choice now.

Three deliberate choices in there. First, I named Ilka directly instead of using {{char}}, because this lorebook is linked to one character and naming her is safer if I ever add a second. Second, two entries share the keywords Bellweather and dispatch, so mentioning the station pulls in both Pell and the lag rule together. That double-fire is the point, and it produced the first line that genuinely startled me: Ilka recorded a message, then talked to herself for a while about what she would do if the answer came back no, because she knew she would have already decided by then.

Third, the rule entry is the only one that tells the model what not to write. Two prohibitions, both specific. I ran about an hour of scenes on Monday evening and Pell never once answered in real time, which is exactly the failure mode I expected to have to fight.

If you build ensembles rather than single characters, one lorebook linked to a whole cast is the real unlock here, and it pairs neatly with my room creation guide.

Lorebook Keyword Examples and the Pattern Behind Them

Twelve entries in, across a harbor town and a cargo ship, my keyword habits have settled into something I can actually state as a rule. Good lorebook ideas die on bad triggers, so this is the part I would read twice. One proper noun, one nickname or role, and one phrase a real person would type. Those three do the work. Anything past them is insurance against phrasing I didn't predict, which is why most of mine run to five or six and none of the twelve uses all eight.

The single most useful trick in that table is the question keyword. Character.AI's docs say to include phrases people actually type, and it's right in a way I didn't expect until I tried it. Adding what does she look like and how old is she to a description entry makes it fire on the exact messages where the model is most likely to invent something. Compare that with a bare noun keyword, which fires whenever the topic drifts past.

Also worth saying plainly: an entry that never fires is invisible, and an entry that fires on every message is worse than no entry, because it spends the model's limited working space on things nobody asked about. The mechanics of that trade-off are in my explainer on how AI companion memory works, and the vocabulary is in the companion glossary if any of this is new. Character.AI's own writing rules are laid out in What makes a good Lorebook, and they're better than most third-party advice I have read.

Character.AI Lorebook Import: What It Quietly Changes

Monday's JSON test is where I learned the parts that will bite anyone moving a world in from another tool. Character.AI reads exported JSON and maps three fields: comment becomes the entry title, key becomes the keywords, content becomes the content. Everything else in the file gets dropped, and some of what gets dropped matters.

• Always-on entries arrive empty. Other tools let an entry stay permanently active with no keys. Here, keywords are the only activation mechanism, so those entries land with a blank keyword field and simply never fire until you fill it.
• More than eight keywords has to be trimmed. The cap is eight per entry and the editor will flag the overflow rather than silently cutting it.
• Clever matching is gone. Regex keys, secondary conditions, trigger chance, and ordering or position settings aren't part of this format. Your imported entries fire on plain keywords alone, which usually means more often than they did before.
• Untitled entries stay untitled, because titles are optional in some tools and required here.
• If you can't save after importing, scan for red or yellow highlights. One broken entry blocks the whole publish, and it's usually something dull: too many keywords, a missing title, an entry that ran long.

Two more mechanics that trip people. A freshly imported lorebook is a draft, and drafts can't be used in chat at all, so you have to publish before you can test whether your entries work. Publishing runs a brief review that finished before I could sit back in my chair. And once a lorebook is live, editing an entry highlights it yellow until you choose Update or Discard, which is a genuinely good design detail and the first version of this feature I've used that treats published lore as something you iterate on. The full mechanics live in Character.AI's Lorebooks help page.

Private Lorebooks Fixed My One Real Complaint

In August I wrote that published lorebooks were public by default and that the only mitigation was a per-entry Hint Only setting, which hides content but leaves titles and keywords on display. That criticism is now out of date and I'm pleased about it. Since August 13, private lorebooks exist: they stay off your public profile, out of search, and only you can view them.

Two strings attached.

A private lorebook works only with your private characters, and it's a c.ai+ feature, which means the privacy and the paywall are the same lever. Let the subscription lapse, as I'm about to for the second time, and you lose the ability to open your own entries. That is precisely why the text file matters, and why lore sits on my companion privacy checklist as a thing to copy out before you cancel anything.

If you go public instead, you get two choices for what strangers see. Full Details shows every entry's title, keywords and content. Hint Only shows titles and keywords with the content hidden, which is still more than nothing, so think about your titles. There is also a Let others link their Characters toggle, which lets other creators attach your published lorebook to their own characters while you keep sole editing rights and the credit. That one also corrects something I said in August: you can now link another creator's Community Lorebook to your own character, and it needs c.ai+ too. Where all this sits in the platform's wider 2026 direction is something I traced in my piece on Microdramas, Lorebooks and what Character.AI is becoming, and the feature tour itself is in the complete Character.AI guide. Character.AI's launch announcement is still the cleanest statement of what they think the feature is for.

More Lorebook Ideas, Sorted by Genre

Running low on lorebook ideas after the obvious ones? These are the prompts I keep in the same text file, one line each, waiting for a world to attach to.

• Workplace: the unwritten rule about who eats lunch with whom, the manager everyone routes around, the one client nobody will fire.
• Cozy mystery: the local paper and what it refuses to print, the shop that closes for no stated reason on Wednesdays, the family everyone is related to.
• Science fiction: a physical limit with a cost, a bureaucracy that outlives the ship, a piece of technology everyone misuses.
• Historical: one price in period money so the model stops inventing economics, plus what a character isn't allowed to say in public.
• Academia: the committee, the feud two generations old, the building nobody is funded to repair.
• Horror: the rule about the thing, the previous person who broke it, and what counts as too late.
• Campaign play: a faction ledger of who owes whom, which is the cheapest way to make a long fantasy adventure roleplay feel consequential.
• Practice partners: a correction contract, a difficulty ceiling, and a rule about when to switch languages.

One caution before you build any of these. Creating a lorebook is still locked to subscribers, so on the free tier my alternatives list will do more for you than these templates. For quick sessions where no world has time to matter, I'd pick from the roleplay app rankings instead. And if you want more out of a character with no lore at all, start with the rest of Character.AI's advanced features.

Paste, Import and Keyword Questions

The Entry I Haven't Written Yet

Twelve entries across two worlds and there's still an obvious hole in both. Neither lorebook contains a single entry about weather, seasons, or the passage of time, and both Sabine and Ilka drift on it constantly. Sabine will describe a July afternoon in what I have established as a November case. I know the fix is a rule entry rather than a description, I know roughly what it should say, and I haven't written it, because pinning down time in a chat that jumps around is harder than pinning down a person.

That's the next thing I'll test, probably on the last day of this subscription. Which brings me to the one question I would actually like an answer to: has anyone got a lorebook entry that keeps a story on a consistent calendar? Every other kind of Character.AI lorebook idea in this post came out of a mistake I made first, and this is the mistake I haven't solved.