How to Use AI for Lorekeeping That Lasts

The village burned three sessions ago.

Its bell still hangs in the blackened tower. A survivor now sits on the city council. One player carries the blame. The faction that supplied the oil has gained a new foothold on the border.

Six sessions from now, the characters may no longer be anywhere near that village.

The world should still remember.

This is where how to use AI for lorekeeping becomes more than a question of convenience. It becomes a question of memory.

For a Game Master or worldbuilder, lorekeeping is not merely collecting place names, family trees, timelines, and fragments of myth. It is the practice of preserving relationships between events, people, promises, wounds, beliefs, institutions, and consequences.

AI can help carry that administrative weight.

It can retrieve forgotten details, compare records, identify connections, organise faction information, trace consequences, and help maintain a campaign whose history may eventually span hundreds of sessions.

What it cannot do is decide what matters.

That remains the work of the storyteller.

AI for Lorekeeping Begins With a Clear Canon

Before asking an AI to organise your setting, establish what counts as true.

This sounds obvious, yet many continuity problems begin when different kinds of information quietly collapse into one another.

An improvised rumour becomes historical fact.

An abandoned campaign idea survives in three documents.

A player theory gets repeated often enough that everyone remembers it as something an NPC actually confirmed.

A priest states with complete confidence that the dead cannot cross running water.

Ten sessions later, the campaign notes simply say:

The dead cannot cross running water.

Something important has been lost.

Not information.

Uncertainty.

A useful lore system therefore needs more than a list of facts.

At minimum, distinguish between things such as:

Established canon — what the world currently treats as true.

Character belief — what a particular person believes.

Public knowledge — what society generally thinks is true.

Rumour or folklore — what circulates without reliable confirmation.

Faction knowledge — what a particular organisation knows or believes.

GM possibility — an idea that has not yet become part of the world.

In investigative games and horror campaigns, these distinctions become especially important.

A witness may be wrong.

A government may lie.

A family may misremember its own history.

A corporation may deliberately alter records.

A legend may preserve something that official history forgot.

AI can help label, retrieve, and compare these categories.

It should not flatten them into one supposedly authoritative truth.

Mystery needs room to breathe.

A Contradiction Is Not Automatically a Problem

This becomes particularly important when campaigns grow old.

Two records may disagree.

Two NPCs may remember the same event differently.

An early session may establish something that later material appears to contradict.

The instinct is often to repair the inconsistency immediately.

That is not always necessary.

First ask:

What kind of contradiction is this?

Perhaps one source is wrong.

Perhaps both describe different perspectives.

Perhaps one character knew less than another.

Perhaps the situation changed over time.

Perhaps an institution deliberately falsified its records.

Perhaps a local legend transformed a historical event.

Perhaps important events occurred between version A and version B that have never yet been described.

And in settings where reality itself can become unstable, the contradiction may occasionally be genuine.

The important principle is:

Explain before you erase.

A living world does not need perfect records.

It needs to understand why its records disagree.

That distinction becomes enormously useful when AI is involved. An AI system may quickly detect that two notes conflict. The storyteller should decide whether that conflict represents an error, incomplete information, changing history, deliberate deception, or a new question the campaign has just created.

Sometimes the best continuity repair is not correcting A or B.

It is discovering C.

What must have happened between A and B for both to make sense?

That missing history may itself become a future story.

Build a Lore System Around Consequence

The most useful lore entries are rarely encyclopaedia articles.

They are records of pressure and change.

A faction entry should not stop with its history and heraldry. It should show what the faction wants, what it fears, what it believes, whom it owes, what resources it controls, what it misunderstands, and how recent player actions have altered its position.

A city entry should not merely describe architecture.

It should remember shortages.

Political tensions.

Destroyed neighbourhoods.

Religious disputes.

Refugees.

Rumours.

New laws.

People who have reasons to love or hate the characters.

This is the principle behind living-world design: lore becomes most useful when it describes movement rather than decoration.

The old empire’s burial rites are interesting.

They become dramatically useful when a disputed succession depends upon those rites, a player character has unknowingly broken them, one religious faction insists the coronation is now invalid, and the dead have begun appearing outside the palace gates.

In the Rexborn Framework, history, influence, social systems, factions, player action, and consequence interact.

You do not need an enormous simulation to adopt the underlying habit.

After something happens, ask:

What changed?

Who knows?

Who cares?

What do they believe happened?

What do we still need to understand?

Only after those questions have answers should the world move again.

Research Should Not End as a Document

This is where AI-assisted lorekeeping can become much more powerful than campaign note-taking.

Suppose the players discover that a ruined monastery once belonged to an order believed to have disappeared three centuries ago.

The immediate response does not need to be five pages of monastery history.

The discovery creates questions.

Who remembers the order?

Who inherited its property?

Which religious institutions would dispute the discovery?

Does a noble family base part of its legitimacy on the accepted version of this history?

Who owns relevant archives?

Who would finance further excavation?

Who would prefer the ruins to disappear again?

Which local legend may actually preserve a distorted memory of the order?

Those questions determine the research.

And that research may reveal actors who did not previously matter.

A minor ecclesiastical archive.

An obscure aristocratic family.

A university department.

A criminal antiquities dealer.

A forgotten branch of an existing order.

An old woman whose family has maintained an apparently meaningless ritual for generations.

This is one of the central transitions in a living-world campaign:

Research reveals actors.

And once an actor has enough motive, knowledge, resources, and ability to affect the world, that actor may need to become part of the faction model.

From Research to Factions

A faction cannot meaningfully act simply because the campaign document says it exists.

It needs identity.

Ask:

What does it want?

What does it know?

What does it merely believe?

What does it misunderstand?

What resources can it reach?

What does it fear losing?

What limits its power?

Where is it internally divided?

Who does it depend upon?

Who does it resent?

Who owes it?

Whom does it owe?

What would make it escalate?

AI can be extremely useful here because it can compare large amounts of research and campaign history without forcing the GM to manually reread every note.

But again, AI should not simply be asked to invent a faction.

The stronger use is to give it existing research and ask it to identify the motives, relationships, contradictions, resources, and uncertainties already present.

This matters because factions do not act according to objective reality.

They act according to the reality available to them.

A bishop may respond to a false rumour.

A corporation may launch an operation because an analyst misunderstood a report.

A cult may correctly understand an artefact that every respectable scholar has dismissed.

A government may spend twenty years responding rationally to information that was wrong from the beginning.

That is not broken worldbuilding.

That is history.

Research gives factions identity.

Influence Is Where Lore Begins to Move

Once factions have enough identity to act, lore stops being purely archival.

Now relationships matter.

Influence matters.

Access matters.

Reputation matters.

Knowledge matters.

Consider a simple example.

The players expose a corrupt magistrate but protect his daughter from retaliation.

The magistrate loses political influence.

His allies become frightened.

His rivals gain confidence.

His daughter may develop loyalty towards the characters.

A criminal organisation loses one route through the city bureaucracy.

A religious institution acquires evidence it could use against the magistrate’s remaining supporters.

A smuggling network benefits from the temporary instability.

The player action was one event.

Its consequences are distributed.

This is where AI becomes useful as a comparison engine.

Rather than asking:

What happens next?

give it the current world state and ask:

Given these established goals, resources, fears, relationships, and knowledge states, which factions are materially affected by what the players have done?

Then ask for several plausible responses.

Do not introduce new villains merely to create movement.

Do not overturn established motives because a more dramatic option appeared.

Trace consequences through the world that already exists.

Research gives factions identity. Influence gives them agency.

Player Action Comes Before the World Tick

Player characters are not simply another faction.

They disturb systems.

They expose secrets.

Kill people.

Save people.

Make promises.

Break promises.

Move resources.

Create alliances.

Destroy institutions.

Change reputations.

Reveal information.

Sometimes they change the world merely by discovering something that already existed.

An ancient ruin may have remained buried for a thousand years without altering the political situation around it.

The moment the players prove it exists, everything may change.

Historians care.

Religious authorities care.

A government cares.

A family with an ancient claim cares.

Collectors care.

Smugglers care.

A cult cares.

Someone who spent thirty years insisting that the ruins were imaginary suddenly cares very much indeed.

The ruin did not change.

The world’s knowledge of the ruin changed.

That distinction matters.

Before running the next World Tick, update the world state.

What happened?

What became known?

Who learned it?

Who misunderstood it?

Whose resources changed?

Which relationships changed?

Which faction gained influence?

Which lost it?

Did the event expose a previously unknown actor?

Did it create a research question whose answer could change our understanding of the past?

Only after those questions have been considered should the world move.

The World Tick Is Not a Plot Generator

A World Tick asks what happens when the rest of the world receives a moment to act.

It should not ask:

What would make next session exciting?

It asks:

Given everything we now understand about this world, what would these actors actually do next?

A frightened institution may destroy documents.

A weakened faction may seek an alliance.

A cult that has lost its ritual site may relocate.

A corporation whose rival acquired forbidden technology may accelerate a programme it previously considered too dangerous.

A political movement may exploit refugees from the burned village.

An ally empowered by the players may begin behaving like the threat they originally needed protection from.

Sometimes almost nothing happens.

That can also be correct.

A living world does not require constant escalation.

It requires believable movement.

AI is particularly useful at this stage because the administrative burden can become considerable. It can compare factions, relationships, outstanding debts, recent events, information flows, and previous consequences.

But it should remain constrained by the established world.

The AI helps calculate possibilities.

The storyteller decides which consequences are meaningful, plausible, and appropriate.

Lorekeeping Must Also Remember Success

Campaign continuity often tracks disasters very well.

Villages burn.

Friends die.

Villains escape.

Curses spread.

Failure leaves obvious scars.

Success should do the same.

Destroying an organisation creates a power vacuum.

Saving a politician may strengthen their faction.

Recovering forbidden research changes who possesses it.

Exposing a conspiracy may cause one institution to collapse while making another much stronger.

Killing a villain may scatter their personnel, archives, money, methods, and unfinished projects across the world.

The organisation can die.

Its knowledge does not have to die with it.

That means lorekeeping should not ask only:

What problem did the players create?

It should also ask:

What balance did their victory change?

A living world remembers success just as aggressively as failure.

Capture the Session Before Memory Cools

All of this depends on having usable evidence from play.

As soon as practical after a session, assemble a raw account.

GM notes.

Player recaps.

Chat logs.

Encounter outcomes.

Improvised NPC names.

Important decisions.

Promises.

Threats.

Discoveries.

Anything that may later matter.

AI can turn this material into a structured session record containing:

events,

new facts,

uncertain facts,

new NPCs,

changed relationships,

player commitments,

faction consequences,

open questions,

and possible areas requiring research.

But review the result before treating anything as canon.

The review matters.

AI can misattribute an action.

Infer something nobody actually established.

Confuse player speculation with character knowledge.

Turn ambiguity into certainty.

Treat it like an exceptionally quick assistant who was not emotionally present at the table.

It can organise the evidence.

You decide what happened.

Recording and Transcription Can Extend the Memory

For groups comfortable with recording their sessions, another layer is increasingly practical: automated transcription.

There are tools capable of recording online sessions or turning existing audio into searchable transcripts.

That transcript can then become another source for AI-assisted lorekeeping.

But a transcript is evidence, not canon.

A four-hour roleplaying session contains rules discussions, jokes, abandoned ideas, player speculation, misremembered names, out-of-character comments, statements made by lying NPCs, and theories that may be completely wrong.

Preserving everything without classification merely allows technology to preserve noise more efficiently.

A more useful workflow is:

Recording → Transcript → Session Extraction → Human Review → Canon Update

The transcript can recover things ordinary notes often miss.

A promise made in passing.

The exact wording of a threat.

An NPC name improvised under pressure.

The first time a player mentioned a suspicion.

A tiny emotional exchange that becomes important twelve sessions later.

The moment a faction first learned something.

AI can search that transcript and extract commitments, discoveries, character interactions, open questions, and possible changes to faction knowledge.

The GM still determines which of those things entered the actual world state.

At Mythveil, recording sessions is not currently part of our standard workflow. The possibility is nevertheless interesting because it could create a much richer evidence layer beneath normal campaign notes.

The fundamental principle remains unchanged:

AI may help us remember what was said. Human judgement determines what it meant.

Recording also carries an obvious responsibility.

Everyone involved should know that a session is being recorded, understand how the material will be processed and stored, and agree to its use.

A campaign archive should strengthen trust at the table, not undermine it.

Information, Interpretation, and Truth Are Not the Same Thing

Once transcripts enter the process, this distinction becomes even clearer.

A player says:

“I think the mayor arranged the murder.”

The transcript records the sentence perfectly.

That does not make it canon.

It is a player theory.

An NPC says:

“Nobody survived the fire.”

Again, the transcript may be perfectly accurate.

The NPC may still be lying.

Or simply wrong.

That creates several useful layers:

Evidence — what was actually recorded at the table.

Interpretation — what we believe happened.

Canon — what has been established as part of the world.

Faction knowledge — what different actors know or believe.

Public narrative — what society thinks happened.

Bardic Resonance — what history eventually turns the event into.

AI lorekeeping is therefore not simply about storing information.

It is about preserving the difference between information, interpretation, and truth.

Update Entities, Not Just Chronicles

A chronological campaign diary is useful.

It is also terrible when the characters suddenly return to an NPC they last met eight months ago.

The GM does not need Session 37.

The GM needs Captain Elian.

What happened to him?

What does he know now?

What does he think about the party?

What promises remain outstanding?

Which faction has pressured him since their last meeting?

What does he want?

What is he afraid they will discover?

AI can help convert chronological session history into updated entity states.

NPCs.

Factions.

Locations.

Artefacts.

Institutions.

Relationships.

Open conflicts.

If Captain Elian promised safe passage, lost two guards, discovered that the characters support a forbidden claimant, and then spent six months living with the consequences, those developments belong on Elian’s current entry.

This protects emotional continuity.

Players may forget the date of a treaty.

They remember betrayal.

Keep a Ledger of Open Threads

Campaigns become thin when every problem disappears at the end of an evening.

Maintain a ledger for things that remain unfinished.

Debts.

Promises.

Missing people.

Threats.

Prophecies.

Unanswered questions.

Political changes.

Unclaimed rewards.

Damaged relationships.

Unexplained discoveries.

A useful thread does not necessarily need to remain active.

It may be dormant.

Advancing.

Urgent.

Resolved.

Transformed.

Forgotten by the characters.

Not forgotten by the world.

AI can identify loose threads across many sessions and surface ones that have become relevant again.

But not every thread should return simply because it exists.

A campaign that remembers everything equally eventually remembers nothing clearly.

Lorekeeping requires selection.

Some things should close.

Some should fade.

Some should become history.

And some should return when their consequences finally arrive.

Ask AI for Options, Never Verdicts

The quality of AI-assisted lorekeeping depends heavily on the question being asked.

Avoid:

“What happens next?”

That invites replacement authorship.

Instead try:

“Given that the party exposed the magistrate but protected his daughter, identify the factions affected by those actions and propose three believable responses based only on their established motives, resources, fears, and knowledge. Preserve the city’s grim late-medieval atmosphere. Do not introduce new villains or overturn established canon.”

That prompt does something very different.

It tells the AI to work inside the world.

Choose.

Combine.

Reject.

Rewrite.

The goal is not to receive the answer.

The goal is to reveal consequences the storyteller may otherwise have missed.

Bardic Resonance Brings the World Back to the Players

There is still one major problem.

Much of a living-world simulation happens where the characters cannot see it.

A faction changes strategy.

A politician reacts.

A survivor leaves town.

A religious argument spreads.

A criminal organisation acquires new territory.

An old enemy’s research is inherited by someone else.

If all of this remains inside the GM’s lore database, the world may be changing constantly while the players experience none of it.

The world therefore needs ways to speak back.

This is where Bardic Resonance becomes part of lorekeeping.

Events do not survive only as factual records.

They become stories.

A battle becomes a newspaper article.

Then a political speech.

Then a drinking song.

Then a conspiracy theory.

Then something children shout at one another in the street without knowing where the words came from.

A survivor tells one version.

The government publishes another.

The faction responsible denies involvement.

A priest turns the tragedy into a sermon.

A historian later writes about it.

A family remembers something none of those accounts recorded.

The player characters eventually encounter people who know their deeds only through the stories those deeds produced.

AI can be extremely useful here.

Once the actual event and its perspectives are established, AI can help transform it into several culturally distinct memories without pretending that each version is equally true.

That distinction matters.

Bardic Resonance does not merely remind the players what happened. It shows them what the world thinks happened.

The Full AI-Assisted Living-World Loop

At that point, AI lorekeeping becomes much more than maintaining campaign notes.

The cycle looks like this:

Play creates events.

Events create questions.

Questions create research.

Research reveals actors.

Actors become factions.

Factions develop influence.

Player action changes that influence.

The updated world produces a World Tick.

The consequences become history.

Bardic Resonance transforms that history into rumours, propaganda, memories, legends, institutions, relationships, scars, and changing places.

The players encounter that changed world.

Their response produces the next event.

And the world asks another question.

In compact form:

Research → Factions → Influence → Player Action → World Tick → Bardic Resonance → New Questions → Research

AI can assist almost every administrative transition in that cycle.

It can retrieve.

Compare.

Classify.

Cross-reference.

Track.

Search.

Surface forgotten connections.

Maintain conflicting perspectives.

Generate variations.

Identify consequences.

What it should not do is determine what the campaign means.

That remains human work.

What AI Should Not Decide

AI is useful at retrieval, synthesis, classification, comparison, and generating possibilities.

It is much less reliable as the keeper of significance.

Do not casually outsource the central truths of your campaign.

The theme you are exploring.

The meaning of a death.

The moral cost of a decision.

Whether forgiveness matters.

Why a character’s sacrifice changes somebody else.

Why one forgotten village deserves to be remembered when hundreds of others do not.

Those are not information-management problems.

They are acts of judgement.

The same applies to player-facing material.

AI-generated letters, rumours, prayers, newspapers, propaganda, historical extracts, songs, and folklore can be excellent atmospheric tools.

They still need editing.

Voice matters.

Context matters.

Canon matters.

And trust matters.

Let the World Keep Its Scars

Return to the village.

It burned three sessions ago.

Months pass.

The characters travel elsewhere.

The survivor joins the council.

The faction that supplied the oil expands its influence along the border.

A politician uses the tragedy to justify new laws.

Someone builds a shrine beside the ruined tower.

Someone else insists the villagers burned their own homes.

The truth becomes argument.

Argument becomes memory.

Memory becomes culture.

Then, ten sessions later, the characters enter a tavern in another city.

Someone is singing a song about the fire.

The arsonists have become demons.

The surviving councillor has become a saint.

The faction that supplied the oil is not mentioned at all.

The story is wrong.

Mostly.

Then one player hears a line in the chorus.

They recognise it immediately.

It is something their character said while the village was burning.

Nobody outside the party should have known those words.

That is lorekeeping.

Not remembering that Village A burned during Session 14.

Remembering what the fire became.

Keep the records.

Track the knowledge.

Research the questions.

Understand the factions.

Let influence shift.

Let player choices disturb the system.

Then let the world move.

And finally, let the world tell the players what their choices became.

AI kept the record.

The world changed it into memory.

Read. Play. Watch. Support. Or Collaborate.

Read: https://mythveilchronicles.com/

Play: https://startplaying.games/gm/dunchan

Workshop — Campaigns That Remember Your Players: https://startplaying.games/adventure/cmqzlufe80063jj04bafujpq0

Watch: https://www.youtube.com/@MythveilChronicles

Support: https://www.patreon.com/mythveilchronicles_bydunchanhuntergames

Collaborate: https://www.linkedin.com/in/francisco-jos%C3%A9-z-b05aab371/

MYTHVEIL

We create worlds people want to live in.

Stories that remember. Worlds that change