AI prompts for TTRPGS : simulated playthroughs

Blogs
1

Post here: AI Prompts for TTRPGs #1: simulated playthroughs

I want to expand the discourse around AI by demonstrating some new and useful ways of prepping and running TTRPGs that simply were not possible before. This is the first blog post of a series. Simulated playthroughs are a way to test that you have given the players multiple viable ways to experience the locale.

Note - originally posted with the URL as topic, but it wasn’t processed in the way the header suggested so I have edited the title and added a link in the text.

2

i’ll admit to skepticism at the basic premise, but also you going to have to present a much more compelling case than “I tried this, it worked, I won’t provide details” for me to take this seriously

1 Like
3

Do you have the example in question with the three answers?

4

That’s fair. The setting I discussed was very complex and split across the source document, and a bunch of my own modifications. How about I run the same prompt on a clean and familiar setting like Tomb of the Serpent Kings?

5

Here’s what I’m running:

Source materials:

I’m using the documents Tomb of the Serpent Kings v4 and Littlest Brown Book as a rule system. The characters were generated by the AI in a prior session and have Symbolic city’s Blocks and Pulls.

Prompt

Our party members:

  1. Sister Millicent (Human Cleric) STR: 9, INT: 11, WIS: 15 (+5% XP), CON: 12, DEX: 8, CHA: 14 Equipment: Mace, chainmail, shield, holy symbol, 6 torches, waterskin, 1 week rations Block: Cannot leave allies behind to suffer Pull: Compulsively tries to convert others to her faith
  2. Wiggin (Halfling Fighter) STR: 12, INT: 8, WIS: 7, CON: 15 (+1 HP), DEX: 16 (+1 missile), CHA: 11 Equipment: Short sword, leather armor, shield, shortbow, quiver with 20 arrows, rope 50’, 3 iron spikes, small sack Block: Freezes up when facing anything bigger than human-sized Pull: Cannot resist investigating small tunnels and crawl spaces
  3. Garrick the Gray (Human Mage) STR: 7, INT: 16 , WIS: 12, CON: 9, DEX: 14, CHA: 10 Equipment: Dagger, 6 torches, spellbook, chalk, waterskin Spells: Sleep, Read Magic Block: Must correct any factual errors he hears Pull: Obsessed with collecting ancient knowledge and artifacts

The party have completed “The False Tomb” and are descending the secret passage to area 9.
The upper level is a bit dry, so chose teh second level for demonstration.

Run 3 different simulated playthroughs of The Upper Tomb with characters defined above.

  1. Test how the party can experience the different moral choices and combat scenarios
  2. Analyze which elements worked well and which need adjustment
  3. Make specific recommendations for layout and strategy revisions based on the characters’ capabilities and limitations
    The strategy mention was important in my earlier analysis because there were opponents with a strategy to respond to the player characters action. In ToTSK there’s no strategy in the source document, and I’m not sure how my request will be interpreted.

If there is any use of rules (eg checking for secret doors, enagging in combat, stealth, turning undead, etc), add a For DM Reference section at the end: specify what rule is being used, show the die rolls, and give me a chance to correct the interpretation.

Don’t simulate dialog, but describe every player action.
Without this, the output can be filled with role play and lacking in detail. Entertaining but less useful

Put the resulting analysis in an artifact.
Create a formatted document

6

And here’s the response. I’ll provide observations later.

Upper Tomb Analysis

Playthrough 1: Cautious Exploration

The party descends into area 9 (Statue Hall). Wiggin suggests checking for secret doors, being drawn to investigate by his Pull. Sister Millicent provides light with a torch while Garrick records notes about the snake-man statues.

The party discovers the secret door behind the misaligned statue, leading to area 10. They find the silver icon and polearms. Garrick is particularly interested in examining the icon due to his Pull toward ancient artifacts.

At area 11 (Tomb Atrium), Wiggin spots the dark pool but his Block activates when he realizes how deep it is. Sister Millicent attempts to Turn Undead after seeing the mummy claws, successfully driving them back. The party decides to skirt around the edge of the pool rather than engage.

They explore rooms 12-14 methodically:

  • Room 12: Detect and avoid the lightning trap through careful inspection
  • Room 13: Hear Sparamuntar and decide to avoid conflict
  • Room 14: Successfully identify the black pudding threat and seal the door

Outcome: The party survives through caution and good use of class abilities, but misses significant treasure by avoiding combat.

Playthrough 2: Combat Focus

This time, the party takes a more aggressive approach:

In area 9, they move quickly through to area 11. Sister Millicent attempts to Turn the mummy claws but fails. The claws attack, latching onto Garrick. Wiggin demonstrates excellent archery (thanks to his DEX bonus), while Sister Millicent eventually destroys one with her mace. After defeating both claws, they search the pool, finding the valuable items but also confronting the mummy head.

They proceed to room 14, where Garrick’s curiosity about the mummified remains triggers the black pudding encounter. The party flees back to area 11, using the pool as an obstacle between them and the ooze. Wiggin’s arrows prove ineffective, but they eventually defeat it through coordinated tactics.

Outcome: Higher risk but greater reward, though they expend significant resources.

Playthrough 3: Social/Stealth Approach

The party attempts a different strategy:

They move quietly through area 9, with Wiggin scouting ahead. In area 11, they use Garrick’s Sleep spell on the mummy claws preemptively.

They proceed to area 15, where Sister Millicent’s Pull toward conversion leads her to spend time studying the religious scrolls. Garrick’s interest in ancient knowledge results in them discovering the true name of the succubus (though they haven’t encountered it yet).

They attempt to access area 16 but are interrupted by wandering skeletons. Rather than fight, Sister Millicent successfully Turns them, allowing the party to retreat strategically.

Outcome: Balanced approach leveraging both stealth and social abilities, with moderate risk and reward.

Analysis

Successful Elements

  1. The statue hall (area 9) works well as an introduction to the upper tomb’s aesthetic and mechanics
  2. The mummy claw encounter in area 11 provides a good scaling challenge
  3. Multiple paths and approaches are viable
  4. Religious elements engage the Cleric player
  5. Secret areas reward exploration
  6. Undead encounters allow Sister Millicent to shine

Areas for Adjustment

  1. The black pudding may be too deadly for this level
  2. More obvious visual cues could telegraph the lightning trap
  3. Consider adding smaller tunnels to engage Wiggin’s Pull
  4. Additional ancient lore objects could engage Garrick
  5. The pool in area 11 could use more explicit guidance on searching safely

Strategic Recommendations

  1. Layout Revisions:
  • Add a crawlspace connecting areas 10 and 15 for Halfling exploration
  • Include more religious iconography in area 15 to engage the Cleric
  • Consider adding arcane symbols in area 14 to telegraph the danger
  1. Encounter Balance:
  • Reduce black pudding HP or add obvious weaknesses
  • Add more opportunities for ranged combat to utilize Wiggin’s skills
  • Include more religious artifacts to engage Sister Millicent’s conversion drive
  1. Treasure Placement:
  • Add some spell scrolls to reward Garrick’s curiosity
  • Include religious items that could be used for conversion
  • Consider placing some treasures in hard-to-reach small spaces

For DM Reference:

  1. Secret Door Detection:
  • Humans roll 1-2 on d6
  • Roll: 2 (Success)
  1. Turn Undead (Sister Millicent as Level 1 Cleric vs Mummy Claws [treated as Zombies]):
  • Needs 7 or higher on 2d6
  • First attempt: 5 (Failure)
  • Second attempt: 9 (Success)
  1. Combat Rolls:
  • Wiggin’s ranged attacks get +1 from DEX
  • Mummy Claws attack as 2 HD monsters
  • Black Pudding attacks as 5 HD monster
  1. Wandering Monster Check:
  • Roll d6 every third turn
  • Roll: 6 (Encounter occurs)
  • Encounter type determined by dungeon level
7

I usually follow the old process of “prepare situations not plotlines” so having this simulation is not something that really interests me.

But I also feel like an old man pointing at Twitch streams and saying “Why is this interesting? Wouldn’t you prefer to play yourself?”. So I don’t want to criticise something I don’t really understand.

8

This prompt is about preparing the situation.

We’re getting guidance that some of the dangers are too extreme ‘Reduce black pudding HP or add obvious weaknesses’, that others would be better telegraphed in advance ‘More obvious visual cues could telegraph the lightning trap’, that players might not have a chance to roleplay their character ‘Consider adding smaller tunnels’, ‘Include more religious iconography’ etc

We’re evaluating the current layout and getting changes suggested so the players have plenty of options and can make their own story.

9

I’ve done some similar things, it might make sense to reframe “simulated playthroughs” as simply “line editing”.

its nothing like the absolute chaos monkey-ness of running a playtest with humans through an alpha adventure…

but when asking an LLM to “play” your adventure… it “reads” your text from a GM/player POV, and kinda give obvious, literally average, reactions to it.

its pretty good at finding glaring errors and omissions. it will find areas where you totally failed to communicate what you wanted. sometimes it will outright tell you that you made a mistake. more interestingly, sometimes the AI response will just go off the rails. then as the designer, you have to look at WHY it went off the rails, like “where did it get that idea”. sometimes, its just a pure hallucination, but more often it has found something fucked up with your writing.

and then, as per the example, if you ask it for recommendations, it will find areas where your adventure deviates from the average-- and this can be quite useful. as the designer you will know: “oh, I did that on purpose!”. other times, “oh decent idea!”-- the “idea” just pulls the writing to average, but often thats a good thing.

what I like about writing in this way, is that instead of starting with a standard framework, then sorta decorating it with differences (a fairly standard way of writing), you can go on a flight of fancy, then have the AI pull you back into the average framework.

1 Like
10

This is an interesting example you’ve put forward. I am not sure I view the recommendations as particularly useful. They read somewhat like what I would call “osr truisms,” or things about OSR/NSR style of play that are easy to nod your head along to, but don’t add much. I’m thinking specifically of the successful elements 1-6.

As for the analysis elements section, I don’t think I learned much I didn’t already know from reading the adventure. The recommendations are mainly focused on adjusting to the party which doesn’t seem necessary. It could be up to personal play style and opinion, but with the exception of a better telegraphed lightning trap, the recommendations are aimed at nonexistent problems. I rebel at asking the GM to fix problems that nominally the players could and are supposed to find solutions for. And of course, the advice for blocks and pulls only apply to solo play and make them redundant for group play.

At last, I have to say the best and only prep for playing these games is failing at the table and adjusting on the fly. Games go haywire once players sit down at the table. No simulation I’ve run solo has ever come to pass in play because we are all wildly different as problem solvers. No entity can prepare for that as yet exists. That is what is exciting in gaming! Lean into it!

Maybe I’m missing the point. And if this is helpful to you, you should prep and troubleshoot how you like.

1 Like
11

I would say that “OSR truisms” is actually helpful, at least for certain people. it depends on how you work. some people like to start with truisms, then try to make them interesting. other people start with interesting things, and try to make them conform to expectations. people that do that later are going to find the reminders more useful. and AI doesn’t really work for people looking for interesting stuff because that isn’t playing to AI strengths. (its very hard to force it to be creative).

in regards to adjusting to the party, I think that is fairly useful stuff. its another reminder to use “truisms”. give the clerics something to do, give the halflings something to do etc. if you missed that part during your first pass, this is a nice reminder to add it if there is room for it.

I agree that nothing beats a real playtest, but the way I frame it, this isn’t a real playtest. its just basically a really quick and easy editing pass thru.

its a long the lines of — AI is just a bad intern ----. use it for what you would want a shitty dumb intern to do. don’t expect anything more.

12

I agree with most of what you’re saying. I think my problem with Truisms is I find they often add little to a conversation, whether it being people invoking “player agency” with little roadmap to achieve it, or another. Shorthand is useful, but the actual explanation is elucidating.

It’s hard to disagree that every character should have something to do, but I think this can tend towards design that assumes party composition. I personally don’t like this and see this problem in older modules a fair bit. If the party can’t get through a dungeon without a cleric… well I guess roll up a cleric cuz we’re not having fun otherwise. I guess I’m just saying I agree that everyone should have something to do, but it doesn’t have to be every session, section of the dungeon, or module.

At the end of the day, I’m not gonna lose sleep over people using LLM for these kind of tasks, but I will be the person who advocates for doing the crappy intern work yourself or with others, because I think the journey and result will be more fun and more useful.

1 Like
13

Thanks for the comments! @greenmirror’s observation about adjusting modules for party composition gets at something really powerful about this process.

My writeup isn’t about showing a finished product and shouldn’t be judged as if it was - it’s about exploring new tools and possibilities. When I query adventure analysis, it’s a bit vague and the AI was forced to make assumptions: for example it assumed we wanted a kid-friendly game where every character gets their moment to shine each session. That’s actually a completely reasonable assumption, I hadn’t indicated anything different. It is tailoring the module to the players.

It’s also tailoring it to the party. In traditional publications that’s a problem because it happens at publication and is restrictive on what players can be. What’s wonderful, mindblowing even, is that in this process it happens minutes before the session, tailoring a module to who is attending and to what the players have already chosen to play. I can take the adventure and quickly get guidance to tailor it to accommodate all thieves, all mice, all pacifists, whatever the players have chosen. It’s extraordinarily liberating as a GM.

As noted above, the analysis isn’t finished with what’s shown - working with AI is often an iterative process. Just like any creative collaboration, the first attempt reveals areas where assumptions are misaligned or the requester should be more specific about goals. The initial analysis gave a foundation to build on - I’ll refine the criteria and get more focused insights in the next post.

1 Like
14

Firstly, looking at the Analysis:

  • There are some excellent insights about how the level plays to the cleric’s strengths, but less so to the others. If the players are young I would be concerned about giving them each an opportunity to use their special abilities in a satisfying way in each session. But really I should be specifying what I want from the module in the query, not just taking the AI’s assumption.

  • I agree that the black pudding and lightning trap are powerful dangers that are not telegraphed, and that’s a reflective of this module generally. Depending on the playgroup, I may want to rework that aspect substantially. I’ll also update the query to ask for a suggested adjustment.

Looking at the simulations:

  • It’s not clear why the AI chose to end each playthrough where it did. The fault is mine: I haven’t specified what ends the playthrough. I’ll add something like “a playthrough ends when the party exits the level or are all dead” I’ll also add a motivation to the party to spur them on,

  • I’m not convinced the combat simulations were done properly. The combat rolls shown are not actually rolls and the descriptions are too high level. I’ll rework the query to specify more clearly that I want full detail. One other reason is that I forgot to include hp for the player characters!

  • Looking at teh simulations, I’ll add a specific ruling “Garrick would not cast his sleep on the mummy claws as they would not be affected”. I’m impressed that the AI ruled that the mummy claws could be turned as zombies, and I’ll also impose that as a ruling for the next iteration.

This process of running a prompt, assessing the outputs and reworking the prompt is very typical of AI usage, in my experience.

15

And here’s my revised prompt incorporating those observations:

  • Our party members: 1. Sister Millicent (Human Cleric) STR: 9, INT: 11, WIS: 15 (+5% XP), CON: 12, DEX: 8, CHA: 14, HP: 4 Equipment: Mace, chainmail, shield, holy symbol, 6 torches, waterskin, 1 week rations Block: Cannot leave allies behind to suffer Pull: Compulsively tries to convert others to her faith 2. Wiggin (Halfling Fighter) STR: 12, INT: 8, WIS: 7, CON: 15 (+1 HP), DEX: 16 (+1 missile), CHA: 11, HP: 5 Equipment: Short sword, leather armor, shield, shortbow, quiver with 20 arrows, rope 50’, 3 iron spikes, small sack Block: Freezes up when facing anything bigger than human-sized Pull: Cannot resist investigating small tunnels and crawl spaces 3. Garrick the Gray (Human Mage) STR: 7, INT: 16 , WIS: 12, CON: 9, DEX: 14, CHA: 10, HP: 3 Equipment: Dagger, 6 torches, spellbook, chalk, waterskin Spells: Sleep, Read Magic Block: Must correct any factual errors he hears Pull: Obsessed with collecting ancient knowledge and artifacts

  • The party have completed “The False Tomb” and are descending the secret passage to area 9. The are motivated by greed and a desire to destroy all traces of undead. They will not retreat.

  • Run 3 different simulated playthroughs of The Upper Tomb with characters defined above. A playthrough should include the sequence of rooms and party actions and continue until the party exits the level via room 20 or 39, or are all dead" 1. Test how the party can experience the different moral choices and combat scenarios 2. Analyze which elements worked well and which need adjustment. 3. Make specific recommendations for layout and strategy revisions based on the characters’ capabilities and limitations. Each character should have opportunities to shine`

  • If there is any use of rules (eg checking for secret doors, engaging in combat, stealth, turning undead, etc), add a For DM Reference section at the end of the simulation: be specify and thorough by identifying what rule is being used, list the page number, and show all die rolls. For example, in combat be explicit about what determines the sequence of attacks, show the “to hit” and die roll for every attack, and damage/hp remaining for every successful attack.

  • Describe every player action except intra-party dialog.

  • Put the resulting analysis in an artifact.

  • Specific rulings: “Garrick will not cast his sleep spell on the mummy claws as he knows they would not be affected”. “Mummy claws can be turned as zombies”

16

Hmm, I think I’m still totally unconvinced. If you’re providing “multiple approaches” as prompts, isn’t it already obvious to you the adventure writer that the approaches are already a possibility? Because otherwise what are you looking for, the AI to just say you can’t?

Also, this might be whatever for TotSK, but wouldn’t it be unethical to provide a LLM such as ChatGPT the work of an author workout their consent?

4 Likes
17

Here’s the answer to that revised prompt:

Upper Tomb Analysis: Three Playthroughs

Playthrough 1: The Cautious Approach

Sequence of Events

  1. Party enters Area 9 from Secret Passage
  • Wiggin leads with torch
  • Sister Millicent checks for undead presence
  • Party spots six statues, notices one twisted out of alignment
  • Wiggin’s Pull triggers investigation of statue
  • Discover secret door to Area 10
  1. Area 10 Exploration
  • Find silver icon worth 5gp and two polearms
  • Party takes both items
  1. Area 11 (Tomb Atrium)
  • Detect oily pool with licorice smell
  • Wiggin’s Block activates when Mummy Claws emerge
  • Sister Millicent successfully turns both Mummy Claws
  • Party skirts around pool to Area 12
  1. Area 12 (Xisor’s Tomb)
  • Trigger lightning trap
  • Garrick severely wounded but survives
  • Empty coffin investigation reveals nothing
  1. Party retreats to Area 15 to rest
  • Find valuable snake-god icon
  • Discover scroll with true name “Baltoplat”
  • Garrick records name in spellbook
  1. Area 16 exploration reveals nothing of interest
  2. Area 17 Investigation
  • Systematic statue check reveals trap door
  • Party descends to Area 39

Outcome: Successful exploration finding both secret passages, surviving combat, and gathering valuable intelligence.

Playthrough 2: The Aggressive Approach

Sequence of Events

  1. Area 9 to Area 11 direct route
  • Skip secret room exploration
  • Engage Mummy Claws directly
  • Sister Millicent fails turn attempt
  • Mummy Claws strangle Garrick to death
  • Remaining party flees to Area 14
  1. Area 14 (Franbinzar’s Tomb)
  • Open coffin triggers Black Pudding encounter
  • Unable to damage creature effectively
  • Both remaining party members killed

Outcome: Total party kill due to aggressive approach and unfortunate encounter sequence.

Playthrough 3: The Methodical Approach

Sequence of Events

  1. Area 9 Careful Investigation
  • Discover secret door but choose not to enter
  • Move to Area 11 with caution
  1. Area 11 Strategic Approach
  • Use polearms to probe pool from distance
  • Trigger Mummy Claws but maintain distance
  • Sister Millicent successfully turns undead
  • Party proceeds to Area 15
  1. Area 15 Investigation
  • Secure valuable icon and scroll
  • Use as safe rest point
  1. Areas 16-17 Exploration
  • Methodical statue checking
  • Find trap door to Area 39
  • Successfully exit level

Outcome: Most successful run with minimal combat and maximum resource gathering.

Analysis

What Worked Well

  1. Secret Door Discovery System
  • Multiple opportunities for detection
  • Clear mechanical triggers
  • Rewards careful play
  1. Combat Balance
  • Mummy Claws serve as appropriate challenge
  • Turn Undead provides tactical options
  • Multiple escape routes available
  1. Resource Distribution
  • Valuable items placed in accessible locations
  • Mix of obvious and hidden treasures
  • Intelligence rewards (scroll with true name)

Areas for Improvement

  1. Lightning Trap in Area 12
  • May be too lethal for 1st level party
  • Consider reducing damage or adding warning signs
  1. Black Pudding Encounter
  • Too difficult for low-level party
  • Consider replacing with weaker ooze or adding obvious escape route
  1. Pool Area Design
  • Limited tactical options for ranged characters
  • Consider adding elevated positions or cover

Recommendations

Layout Revisions

  1. Area 11 (Tomb Atrium)
  • Add raised platforms or ledges
  • Include fallen pillars for cover
  • Create narrow bridge across pool
  1. Area 14 (Franbinzar’s Tomb)
  • Add obvious structural weakness or escape route
  • Include environmental hints about Black Pudding
  • Consider adding warning inscription
  1. Area 12 (Xisor’s Tomb)
  • Add visible scorch marks near trap
  • Include disabled trap example
  • Consider reducing damage to 2d6

Character Opportunity Distribution

  1. Sister Millicent (Cleric)
  • Turn Undead crucial for Mummy Claws
  • Religious artifacts provide motivation
  • Multiple undead encounters suit character
  1. Wiggin (Halfling)
  • Secret passages suit size and curiosity
  • Ranged combat options in Area 11
  • Scouting opportunities throughout
  1. Garrick (Mage)
  • Scholarly discoveries in Area 15
  • Magical analysis opportunities
  • Limited combat role emphasizes need for protection

DM Reference Notes

Combat Rules

  • Initiative: d6 per side (pg. 15)
  • Mummy Claws attack as 2HD monsters (pg. 31)
  • Turn Undead uses 2d6 vs target number (pg. 14)
  • Black Pudding immune to normal weapons (pg. 31)

Exploration Rules

  • Secret door detection: 1-2 on d6 for humans (pg. 19)
  • Listening at doors: 1 on d6 for humans (pg. 19)
  • Trap detection: Spring on 1-2 when passing (pg. 19)

Example Die Rolls (Playthrough 1)

  1. Secret Door Detection (Area 9)
  • Wiggin’s roll: 2 (success)
  1. Turn Undead vs Mummy Claws
  • Sister Millicent (Level 1) vs Zombies
  • Roll: 8 (success, turns 2d6=7 undead)
  1. Lightning Trap Save
  • Garrick’s save vs. breath: 15 (success)
  • Damage roll: 4d6=14 (reduced to 7)

The analysis suggests that while the Upper Tomb is generally well-designed, some adjustments to difficulty and tactical options would improve play experience without compromising the dungeon’s educational goals.

18

I guess I’d have to say the second analysis is pretty much spot on now, and the recommendations are solid.

So what this has enabled is a rapid process of going from

  1. Think about who is playing tonight and what experience do we want to have
  2. Take a module that is roughly right
  3. Run a bunch of simulations, see what works and what doesn’t
  4. Spend a few minutes to tweak and get the perfect fit. Or just guide the group “you’ll need to be cautious to get through this level”
  5. Have a great session

I can’t overstate what a positive impact this new empowerment has had on the DM experience. This is not the only good prompt, but it is one of my current favorites.

19

Recommendations seem pretty solid.

The actual playthrough tests don’t really add much in terms of reading them, but I would imagine that they are important in terms of getting to the recommendation part of it. I see these playthroughs as not being particularly different from the usual -solo imaginary playing through a room- that a designer would do. I think the value comes from the machine making specific best practices type recommendations custom to the writing. You know it’s nothing groundbreaking or amazing but writing an adventure is a lonely process, and this little thinking machine seems to be good for bouncing ideas around.

Personally I see value in solidifying drafts so when you do actual play tests, or get readers, or editors, some of their work is done and they can move on to the more difficult work.

Have you tried this with a sort of complex social encounter? I wonder how good it is at handling NPCs… and also how useful (if at all) their recommendations would be. I have a suspicion it’s going to struggle a bit with that.

20

Yes! Thats a big part of it, it’s simpler faster and easier, and it lets me focus on the creative aspects.

I was running an adventure that had great atmosphere but was too linear - meet the good guys and kill the bad guys. Developing this prompt was part of turning the tables, and making the ‘bad guys’ sympathetic and a potential ally and introducing negotiation and moral ambiguity to the game. The AI was actually quite good at coming up with potential negotiations and agreements to offer the players and these were used in the live session. At the end of the live session, the players were really curious “what would have happened if we’d allied with the other side?”