Roughly 54.6% of U.S. adults aged 18 to 64 had used a generative AI tool by mid-2025, according to the Federal Reserve Bank of St. Louis. Tabletop Keepers are part of that wave, and horror scenario design is one of the more practical places to apply it. The catch: a Call of Cthulhu campaign assembled carelessly with a language model tends to fall apart at the exact moment it should tighten – the reveal.
The fix isn’t complicated. It comes down to sequencing. Lock the truth first. Let the AI build the scaffolding around it.
Why Does AI-Generated Horror Often Fall Flat?
Because the ending gets invented too late. When a language model is asked to generate scenes and solve the mystery at the same time, the two tasks compete with each other. Clues stop aiming at anything specific, NPCs feel interchangeable, and the climax arrives without the buildup to support it.
This isn’t a flaw unique to horror writing. A recent Federal Reserve note on AI adoption found that about 41% of U.S. workers were using generative AI for work tasks by late 2025 – yet most of that use still leans on human oversight rather than full delegation. Scenario design works the same way. The AI handles volume and detail. The Keeper handles judgment.
Step One: Fix the Core Truth Before Prompting Anything
Three elements need to be locked before a single prompt gets typed:
- The Mythos threat driving the events
- The antagonist’s timeline of actions leading up to session one
- The cosmic secret that ties every clue back together
A working example: in a 1920s New England setting, the hidden truth might be that a small-town coroner has gotten hold of a Mi-Go brain cylinder and is using it to extract secrets from murder victims before burial. That single sentence becomes the anchor for everything built afterward.
Pro tip: Write the truth down as a fixed paragraph and paste it into every subsequent prompt. This keeps the AI from quietly drifting away from canon over a long session.
Building the Investigation: What Should the AI Actually Generate?
Once the truth is locked, the AI stops guessing and starts populating. This is the stage where a Call of Cthulhu campaign generator earns its usefulness – not by inventing the mystery, but by dressing it.
How Do You Structure Node-Based Locations and NPCs?
Use a web, not a line. A node-based structure gives investigators several places to go rather than one scripted corridor, and every node still connects back to the same central truth.
A practical prompt might request four locations tied to the coroner plot: the city morgue, his boarding house, a local pub, and an abandoned warehouse near the docks. For each one, the AI can supply a key NPC, a private secret, a personal fear, and a reason that character is tangled up in the larger plot.
This is exactly where a Call of Cthulhu NPC generator pulls its weight. Instead of a string of forgettable background characters, each NPC carries something worth finding. The morgue attendant might be terrified of the coroner but too broke to walk away. The landlady might have smelled something wrong through the floorboards and said nothing, out of superstition more than loyalty.
What Is the Rule of Three, and Why Does It Matter Here?
The Rule of Three means every essential clue needs at least three separate paths leading to it. This single principle solves the most common failure point in homebrew mysteries: a vital clue locked behind one skill roll that, if failed, stalls the entire session.
Take a detail like strange metallic scars near the base of a victim’s skull. That fact could surface through:
- Morgue records, via a Medicine or Science roll
- Gossip overheard at the pub, via Psychology or Fast Talk
- The coroner’s personal letters, if investigators search his room
Three doors, one destination. Miss two, and the third still gets the table where it needs to go. When prompting an AI, it helps to specify the clue’s content first, then ask for three distinct discovery methods – vague requests tend to produce vague, interchangeable results.
Drafting Handouts and Red Herrings
Physical props do quiet work at the table. A torn diary page or a shaky handwritten note handed directly to players creates a moment of engagement that spoken description rarely matches on its own.
AI tools handle period-appropriate language reasonably well here. A short journal entry from the coroner’s assistant, written in an unsteady hand, hints at guilt without explaining everything outright – and that restraint often lands harder than a document that spells out the full truth.
Red herrings need the same discipline. A believable false lead should be active enough to draw suspicion but unconnected to the real solution. A grave robber working the same cemetery where the coroner disposes of evidence fits naturally – guilty of something, just not the thing investigators assume.
Turning the Outline Into a Playable Call of Cthulhu Campaign
With the truth locked, locations mapped, clues tripled, and handouts drafted, the last step is shaping everything into a session a group can actually sit through.
Every solid Call of Cthulhu campaign needs a hook that gives investigators a reason to get involved in the first place. A missing person, a blood-stained letter, or a panicked relative all work well – none of them require elaborate backstory, and each implies stakes immediately.
Three or four investigation sites is usually the right count. More than that dilutes focus and stretches a single session across several sittings without payoff. Each site should offer at least one clue, one NPC worth talking to, and a small note of unease – a strange smell, an object out of place, a witness who won’t quite meet anyone’s eyes.
The climax deserves restraint. A rushed confrontation with a Mythos entity feels cheap if the buildup hasn’t earned it. Starting with small, explainable oddities and escalating slowly toward the impossible tends to hit harder than an early reveal. By the time the coroner is caught mid-ritual, cylinder in hand, players should already suspect something is deeply wrong – even before they can name it.
A few habits keep an AI-assisted outline from reading generic:
- Balance sanity loss against physical danger so investigators rely on observation over combat
- Keep dead ends rare; if a clue seems to run out, quietly relocate it to wherever players are already searching
- Let some beats stay implied rather than explained through dialogue
Locking the truth first, letting the AI populate the structure around it, and checking every clue against the Rule of Three turns a pile of generated material into something that actually plays well. Try mapping out one node-based location tonight – four sites, one NPC each, three clue paths per key fact – and see how much faster the rest of the outline comes together.
Frequently Asked Questions
Can an AI generate a full Call of Cthulhu campaign on its own?
It can generate the surrounding material – locations, NPCs, handouts – but the core truth and villain motive should be fixed by the Keeper first. Left unsupervised, AI tends to invent endings that don’t match earlier clues.
How many locations should one investigation outline include?
Three or four is generally enough for a single session or a short arc. More locations dilute pacing and risk stretching the mystery past the point where tension holds.
What happens if players fail every roll tied to a clue?
With the Rule of Three in place, at least one alternate path should still be open. If all three somehow fail, it’s reasonable to let the clue surface anyway, at a cost – lost time, a minor injury, or unwanted attention.
Is a Call of Cthulhu NPC generator reliable for building believable characters?
It’s reliable for generating structure – secrets, fears, motives – but the Keeper still needs to check that each NPC’s details don’t contradict the locked core truth. Treat AI output as a draft, not a final answer.
Should red herrings be resolved by the end of the session?
Not necessarily. A red herring that stays slightly unresolved often adds more unease than one that gets neatly wrapped up, as long as it doesn’t block progress toward the actual solution.
