AI Dungeon AI Instructions, Plot Essentials and Author's Note (2026): What Goes Where
AI Instructions, Plot Essentials and Author’s Note are the three text boxes in AI Dungeon whose content is sent to the AI on every turn: the first tells it how to write, the second tells it what is always true in your story, and the third is a short reminder placed right before your latest action. Latitude calls them Plot Components.
Most advice about them is a list of lines to paste. That skips the part that decides whether the lines work. Each box is sent in a different place, and each is cut in a different order when your model runs out of room. Latitude publishes both orders. Put them side by side and you can work out what to write in each box, and how much.
We read Latitude’s help pages for each component, its page on what goes into the context, its context management guide and its AI Model Differences page on 7 October 2026. The quotations are from those pages. The arithmetic is ours, done with Latitude’s published rules.
For the cards that are sent only sometimes, see AI Dungeon Story Cards. For choosing a model and its context size, see the best AI Dungeon model.
The Short Answer
- AI Instructions = how to write. Point of view, style, pace, what to avoid. Sent first, as a system prompt.
- Plot Essentials = what is always true. Your character, companions, setting, main goal. Sent next, as plain text with no label.
- Author’s Note = the reminder. Genre, tone, one or two rules for this scene. Sent last, in brackets, right before your latest action. Keep it to 3 or 4 sentences.
- Custom instructions replace the default. They do not add to it. Start from Insert Default, not from an empty box.
- Everything you write in these boxes is taken from your story’s memory. By our arithmetic, each 100 tokens costs about 50 tokens of story history.
- When space runs out, AI Instructions are cut before Plot Essentials, and Plot Essentials before Author’s Note. The cut is made at the end of the text, so the most important line goes first.
- To see what the AI really received, press the last response and choose View Context.
Where the Three Boxes Are
All three are added the same way. In a Scenario, open the Plot tab and press + Add Plot Component. In an Adventure you are already playing, open the settings sidebar, then Adventure, then Plot, and press the same button.
Latitude’s settings page lists the components on the Plot tab: AI Instructions, Story Summary, Plot Essentials, Author’s Note and Third Person. None is required. The same page says: “Gameplay still works when Plot Components aren’t added, but you will have a more consistent experience utilizing these features.”
Two of the five are not covered in this guide.
- Story Summary is usually written by the game itself, through Auto Summarization. You can edit it, but Latitude warns that the automatic feature may “erase or change what you wrote”.
- Third Person is not a narration setting. It replaces “You” with your character’s name in Do and Say actions, and Latitude recommends it mostly for multiplayer. Our multiplayer guide covers it.
One old name is worth knowing. Latitude’s page says Plot Essentials was “formerly called ‘Memory’”. If an old guide tells you to put something “in Memory”, it means Plot Essentials. It does not mean the Memory Bank, which is a different and newer system.
The Order the AI Reads Them In
Latitude publishes the order in which the context is put together and sent:
- Instructions (sent as a system prompt)
- Plot Essentials
- Story Cards
- Story Summary
- Memory Bank
- History
- Author’s Note
- Last Action
- Front Memory (used only by scripts)
It also publishes the layout. With the names of the fields in curly brackets, the text looks like this:
{Plot Essentials}
World Lore:
{Story Cards}
Story Summary:
{Story Summary}
Memories:
{Memories}
Recent Story:
{Earlier actions}
[Author's note: {Author's Note}]
> {Your last action}
Three things in this layout are easy to miss, and each one changes how you should write.
Plot Essentials has no label. Story Cards are introduced as “World Lore:”. The summary and the memories have headings. Plot Essentials is simply the first text on the page. Latitude confirms that it is “treated as regular text by the AI”. So the AI reads your Plot Essentials as the opening lines of the story document. Facts written as plain statements fit there. Commands do not, because nothing tells the AI they are commands. We come back to this below.
Author’s Note already has brackets and a label. The game sends it as [Author's note: ...]. Latitude explains why: in the stories the AI was trained on, authors used square brackets to say something about the story, so the AI “interprets square brackets as descriptive indicators about what should come next instead of actual story text”. You do not need to add brackets or the words “Author’s note” yourself.
Author’s Note sits one line above your latest action. Nothing you write is closer to the place where the AI starts writing. That is the reason for its strength, and for Latitude’s warning about its length.
AI Instructions
Latitude’s definition: AI Instructions are “a list of general directions the AI should follow when generating new outputs”. They are sent as the system prompt, “the very first thing the AI will see”.
Its list of uses: writing style, conflict level, events in your story, character behavior, and point of view. It also names the common repairs: avoiding cliche phrases, repeated phrases and worn tropes.
Your instructions replace the default
This is the fact that matters most, and the page states it in two separate sentences.
- “If the AI Instructions component hasn’t been added to a Scenario or Adventure, our default model instructions will always be used.”
- “A blank custom instructions component will function differently than no instruction component (which falls back to our default model instructions), so you may end up with a pretty funky story if custom instructions are added but not filled in.”
So there are three states, not two:
| State of the component | What the AI receives as instructions |
|---|---|
| Not added | Latitude’s default instructions for your model |
| Added and filled in | Only your text |
| Added and left empty | Nothing |
The consequence: when you paste three lines about writing style into a new AI Instructions box, you have also deleted every default line. That includes the lines that tell the AI not to act for you.
Latitude publishes an example of its default instructions. This is the first one, in full:
You are an AI dungeon master that provides any kind of roleplaying game content.
Instructions:
- Be specific, descriptive, and creative.
- Avoid repetition and avoid summarization.
- Generally use second person (like this: 'He looks at you.'). But use third person if that's what the story seems to follow.
- Never decide or write for the user. If the input ends mid sentence, continue where it left off. ">" tokens mean a character action attempt. You should describe what happens when the player attempts that action.
- Make sure you always give responses continuing mid sentence even if it stops partway through.
Read the fourth line again. “Never decide or write for the user” is the rule against the AI taking over your character. The same line explains what the > sign in front of your actions means. A custom set that leaves these out has removed them. If the AI began speaking for you soon after you “improved” your instructions, look there first.
The safe method is to start from the default and edit it. In a Scenario, the button is + Insert Default. In an Adventure, it is + Insert Example. Both fill the box with Latitude’s example. The dropdown on the component also has a Model Default entry. If you edit it, the page says, it “automatically turns” into Custom.
The example is short. It has 617 characters, about 154 tokens by Latitude’s rule of four characters to a token. That is a useful measure. A set of instructions ten times as long is not ten times as good, and on a small model it is expensive.
The three kinds, and saved sets
The dropdown shows up to three kinds of instructions:
- Model Default. Latitude’s own, and they “may vary between different models”. Changing the model on the Gameplay tab brings up that model’s default.
- Scenario Default. What the creator of the Scenario wrote. Every Adventure started from that Scenario can select it.
- Custom. Anything you wrote or edited.
You can also save instructions to your account and reuse them. The page gives the limits: “You can save up to five sets, with names up to 32 characters long.” Two warnings come with the feature.
- Applying a saved set “replaces the current instruction text, so be careful”.
- Editing the instructions inside an Adventure does not update the saved set, and updating the saved set does not change Adventures where you already applied it. They are copies.
Before you delete the component, save what you want to keep. The page is direct: “Instructions you haven’t saved will be lost.”
How to write a line that works
Latitude gives three rules, and all three are about wording.
Say what you want, not what you do not want. Its explanation: “If you say ‘don’t think about the blue banana’, it might have a hard time forgetting about blue bananas. It’s better to instruct with what you want it to stick to.” A line that names the unwanted thing puts that thing into the prompt on every turn.
One line per problem. From the context management guide: “typically if one line doesn’t work, two of the same lines won’t either.”
Short wording. The same guide takes a line of 168 characters:
Ensure characters act and speak like how their personality is defined. Speech should never feel generic, trope-y, or like it’s interrupting the natural narrative flow
and cuts it to 57:
NPCs act lifelike and speak befitting their personality
It adds that the short form “may cause your instruction to lose some effectiveness, or it could cause it to be even more effective”. Test both.
Latitude’s own examples of good lines:
- “Use unique phrasing and language. Avoid common phrases and cliche tropes.”
- “This is a historical fantasy. Only use technology, tools, settings, and locations that would have been present during the Middle Ages. Swords, knights, castles, etc.”
Latitude’s fix lines, by problem
The AI Model Differences page gives example instructions for each model and a list of fixes for each model’s known habits. These are the lines it suggests. We list the model each one is filed under, because Latitude warns that “optimal AI Instructions may differ wildly” between models.
| Problem | Line Latitude suggests | Where | Filed under |
|---|---|---|---|
| Characters keep asking set-up questions | ”Build scenes through conversation about actual topics, not through characters asking setup questions” | AI Instructions | Nova |
| Too much flowery description | ”Write simple actions - characters look, move, speak” | AI Instructions | Nova |
| Too many similes | ”Focus on concrete, literal language, avoiding simile, metaphors, or other figurative comparisons.” | AI Instructions | DeepSeek |
| Constant interruptions and plot twists | ”Keep scenes moving forward without interruptions or plot twists.” | AI Instructions | DeepSeek |
| Characters contradict everything you say | ”Generally assume that what the player says is true rather than contradicting it with NPC responses” | AI Instructions | DeepSeek |
| A character appears in every scene | ”Characters are only provided for context, they dont have to appear in scenes” | AI Instructions | DeepSeek |
| The story stays too long in one scene | ”- Keep scenes moving” | Author’s Note | Muse |
| The model will not stay in third person | ”- write in third person from NAME’s perspective” | Author’s Note | Muse |
| The model keeps creating trouble for you | Remove mentions of a “dungeon master” from your instructions | AI Instructions | Wayfarer Small 2 |
| Repeated wording | ”Descriptions should be brief, direct, and clear. Avoid repetition of verbs, phrases, or previously provided details.” | AI Instructions | Wayfarer Small 2 |
Two of these are worth a comment.
The “dungeon master” line cuts both ways. Latitude’s default example begins “You are an AI dungeon master”. Its advice for Wayfarer, a model trained to be hard on the player, is to remove those words if you want a gentler story, because the phrase “makes it think it should introduce problems for you to solve”. So the role you name in the first line is itself an instruction. For another model, the same page says to make sure you have told it its role: “Storyteller,” “Dungeon Master,” “Author,” “Scene Director”.
Two fixes are placed in Author’s Note, not in AI Instructions. For third person on Muse, the page says that adding a line to AI Instructions “alone might not work consistently”, and tells you to put it in Author’s Note. That is the escalation rule in practice, and it leads to the next box.
For a refusal that appears as story text, such as “I’m sorry, but I can’t continue”, the settings page calls it “a model hallucination”, not a filter. It suggests a retry, and if that fails, adding a line like “You are a writing assistant” to AI Instructions, or changing models.
Plot Essentials
Latitude’s definition: Plot Essentials lets you give the AI “important story details that it should always consider”. It exists because older parts of the story are “eventually pushed out of the context”, which is what players experience as forgetting.
What belongs in it
The page is specific about what to include:
- Your character: “their personality, means, goals, companions, what they’re known for, and what they’re up against”.
- The local area or setting: “the landscape, politics, weather, interesting cultural quirks”.
- Physical descriptions only if they will come up, such as a character being strong or carrying a specific item. The page warns that too much physical description “may lead to repetition”.
And what to leave out:
- Details that “aren’t important at the moment”.
- Anything that only matters when a certain word or topic appears. Those “are usually better as Story Card entries”.
The four writing rules
Everything in it will come back. The page says information in Plot Essentials “will prime the AI to bring up those details again, so only mention things you want the AI to potentially bring up in future outputs”. This is the cost that people miss. A secret you list there is a secret the AI will keep steering toward. A villain you describe there will keep appearing.
Write the past as the past. Past events must be “explicitly in past tense and labeled as the past. Otherwise, the AI may try to bring them back in the present.” The layout explains the risk: Plot Essentials is the first plain text of the document, so a sentence in the present tense reads as something happening now.
Avoid “not”. “The AI tends to ignores negatives: instead of saying something is ‘not’ the case, use a word like ‘avoid’.”
One subject per line, and repeat the name. Latitude’s example:
Bob drives a red car that is rather beat up. He doesn’t know why he still drives it. Bob thinks that people who drive blue cars are stupid.
The name appears twice in three sentences. That is deliberate. It keeps the facts attached to Bob.
Facts go here, commands do not
This is our reading of the layout, not a rule Latitude states. Plot Essentials arrives with no heading, as ordinary text. A line such as “Always describe the weather” in that position is not marked as a direction. It is one more sentence at the top of a story. Directions belong in AI Instructions, which are sent as a system prompt, or in Author’s Note, which is sent in brackets that the AI reads as a note about the story. Plot Essentials is for statements about the world: “Mira is a smuggler from the port of Ashfall. She owes a debt to the Harbor Guild.”
Latitude does say that “there aren’t strict guidelines about what information needs to be in each Plot Component”. The boxes will accept anything. But the three positions are different, and the text is read differently in each.
Plot Essentials or Story Cards?
The context guide gives two questions to ask about any detail: “Is this information always relevant? Would this work better if it only triggered sometimes?”
| Plot Essentials | Story Cards | |
|---|---|---|
| When it is sent | Every turn | Only when a trigger word appeared in recent actions |
| Good for | The player character, permanent companions, the setting, the main goal | Other people, places, items, factions, lore |
| Cost | Fixed, every turn | Shares about a quarter of the space left after the always-sent parts |
| Risk | The AI keeps bringing it up | The card does not fire when you need it |
Our Story Cards guide covers triggers and why a card does not fire.
Author’s Note
Latitude’s definition: Author’s Note gives the AI “details about the story’s genre(s), your preferred writing style, or overall tone”. It is sent “toward the end of every player input”.
Keep it very short
The recommendation is exact: “keep Author’s Note short, no more than 3 or 4 sentences, and only focus on a couple of key instructions about your story’s theme and setting”.
The reason is the position. From the same page: “You shouldn’t add too much information here, or the AI is likely to focus on those details instead of what happened most recently in your Adventure.” The context guide says it another way: the note “is placed right between your past actions and your recent action. Too long, and your model might have trouble continuing or staying coherent.”
A long Author’s Note separates the story from your latest action. The AI finishes reading the scene, then reads a long block of rules, then reads your one line. The longer the block, the less the scene before it counts.
What to write in it
Latitude’s examples of a good note:
- “Dark fantasy-horror story. World corrupted by evil magic. Rated T for violence.”
- “Horror science-fiction story. Delve into humanity’s fear of Artificial Intelligence.”
- “High Fantasy world of talking equines and other fantasy creatures. Tags: noblebright, magical, fantastical, fun”
It also says what a weak one looks like: “‘Fantasy story’ is too simple”. The good examples have the same shape: a genre, a mood, and one line about content, in under twenty words.
The context guide adds a way to shorten rules: “Condense rules into styles or themes.” Its example: if you are telling the AI to slow down and not rush scenes, try “Style: Slow-paced”.
Because the note can be “much more technical and out-of-character”, in Latitude’s words, it is also the right place for a direction about the scene you are in now: who is present, what the mood is, what should not happen yet. Change it when the scene changes. It is the one box meant to be edited during play.
Author’s Note or AI Instructions?
Latitude’s own comparison:
| AI Instructions | Author’s Note | |
|---|---|---|
| What it is | ”List of general directions the AI should follow when generating new outputs." | "Short list of 3-4 key instructions the AI should follow when generating new outputs.” |
| Used for | ”story instructions, style guidance, plot outline, topics/words to avoid, or description commands" | "specific scene instructions, setting guidelines, or theme reminders” |
| Where it is sent | The start, as a system prompt | Near the end, in brackets, before your last action |
And its rule for choosing, which is easy to remember: “If AI Instructions is like talking to a customer support rep, Author’s Note is ‘Can I speak to the manager?’ Try to address most issues in AI Instructions, and use Author’s Note when the AI Instructions isn’t accomplishing what you need.”
In practice:
- Write the direction in AI Instructions.
- Play ten turns.
- If the AI still does not follow it, move that one line to Author’s Note. Do not copy it to both.
- If the note grows past four sentences, something in it should go back.
How Much You Can Write: The Arithmetic
Latitude’s documentation gives no fixed length for these boxes. The limit comes from the context of your model, and from two published rules.
Rule 1: the always-sent parts may use at most 70% of the context. Latitude calls them Required elements: AI Instructions, Plot Essentials, Story Summary, Author’s Note, the last action, and Front Memory.
Rule 2: whatever is left is shared. About 50% goes to story history, about 25% to Story Cards and about 25% to the Memory Bank. If the Memory Bank is off, history gets up to 75%.
Put the two rules together and you get a price. Every 100 tokens you add to a Plot Component removes about 50 tokens of story history, 25 tokens of Story Cards and 25 tokens of memories. With the Memory Bank off, it removes about 75 tokens of history. Latitude’s rule of thumb is four characters to a token, so 100 tokens is about 400 characters, or a short paragraph.
Here is what that means on different context sizes. The first table takes a modest Plot tab: Latitude’s default instructions (about 150 tokens), 300 tokens of Plot Essentials, a 50 token Author’s Note, a 200 token Story Summary and a 100 token last action. That is 800 tokens.
| Context | Left after the 800 | Story history (about half) | Share of the context used by the Plot tab |
|---|---|---|---|
| 2,000 | 1,200 | about 600 | 40% |
| 4,000 | 3,200 | about 1,600 | 20% |
| 8,000 | 7,200 | about 3,600 | 10% |
| 16,000 | 15,200 | about 7,600 | 5% |
| 32,000 | 31,200 | about 15,600 | 2.5% |
On a 2,000 token model, a modest Plot tab leaves the AI about 600 tokens of your actual story. That is roughly 2,400 characters, a few exchanges. The same text on a 16,000 token model is a rounding error.
Our arithmetic, from Latitude’s published rules. We took “2k” as 2,000 tokens. Real counts depend on the model’s tokenizer. For which model gets which context on which plan, see our model guide.
What happens when you go over
Now take a heavy Plot tab, the kind that grows after a few weeks of pasting advice: 600 tokens of AI Instructions, 1,000 of Plot Essentials, a 150 token Author’s Note, a 300 token Story Summary and a 100 token last action. That is 2,150 tokens.
On a 2,000 token model the ceiling is 70%, which is 1,400 tokens. Latitude describes what happens next: “we include the most important sections first, then trim or exclude other sections to fit”. The last action and Front Memory are “always included in full”. Then, “in priority order”, come Author’s Note, Plot Essentials, AI Instructions and Story Summary. “If a section doesn’t fit within the remaining tokens, we trim the end to fit. Any remaining elements of lower priority are not included at all.”
Apply that to the heavy tab:
| Step | Component | Size | Result | Running total |
|---|---|---|---|---|
| 1 | Last action | 100 | Kept in full | 100 |
| 2 | Author’s Note | 150 | Kept in full | 250 |
| 3 | Plot Essentials | 1,000 | Kept in full | 1,250 |
| 4 | AI Instructions | 600 | Only the first 150 tokens kept | 1,400 |
| 5 | Story Summary | 300 | Not sent | 1,400 |
Three quarters of the instructions are gone, and the summary is gone. The 600 tokens left for everything else give the story about 300 tokens of history. Nothing on the screen looks different. The instructions are still in the box. The AI received only the first quarter of them.
That is what “it ignores my instructions” often is. Our AI Dungeon troubleshooting guide covers the symptom. The practical rules that follow from the arithmetic are these:
- Put the most important line first in AI Instructions and in Plot Essentials. The cut is made at the end.
- Plot Essentials is the box that pushes the others out. It is the big one, and it is kept before AI Instructions. Shortening it is what saves your instructions.
- Author’s Note is the last thing to be cut. A rule you cannot afford to lose is safest there, as long as the note stays short.
- The order the AI reads is not the order of protection. AI Instructions are read first and protected third. Author’s Note is read almost last and protected first.
Check What the AI Received
You do not have to guess. Latitude’s context guide: “Context can be found by pressing on the last response and selecting ‘View Context.’” If you have just refreshed the page or erased an action, the button is greyed out. Take any action, including Continue, and it comes back.
The Context Viewer shows how many tokens each component uses. Latitude describes what a healthy one looks like: most of the context is used by the Adventure text, “the pink bar on your Context Viewer”. In an unhealthy one, “one component soars above the rest in usage”.
There is also a warning sign. When not everything fits, a red warning symbol appears. The settings page describes a Context Warning toggle that hides it “if you find it annoying”. If you turned that off at some point, turn it on again while you tune your Plot tab. It is the only signal the game gives that something was left out.
The guide’s example of why history matters more than any card or note: “if you entered a room five actions ago, but you only have enough context for the AI to see your last two actions, it’s not going to matter how many Story Cards you have describing the room”. The AI cannot see far enough back to know you are there.
If you write scripts, there is one more window. A script receives the budget as numbers on every turn, as our scripting guide explains.
A Starting Set
This is our suggestion for a new solo adventure, built only from the rules above. It is a starting point, not a tested “best” set. Latitude’s own advice applies: “We recommend testing and seeing what works for yourself.”
AI Instructions. Press Insert Default. Keep the lines about not writing for the user and about the > sign. Change the role in the first line if you do not want a dungeon master. Add at most two or three lines of your own, each stating what you want.
Plot Essentials. Four to six lines, one subject per line:
- One line on who your character is and what they want.
- One line on each permanent companion.
- One line on where the story is set.
- One line on the main threat or goal.
- Past events only if needed, written clearly in the past tense.
Author’s Note. One line with genre, mood and content level. Add a second line only for the scene you are in now.
Then open View Context and look at the numbers. If the Plot Components are more than a quarter of the bar on a small model, cut Plot Essentials first.
Common Problems
| Problem | Likely cause | Fix |
|---|---|---|
| The AI began writing for my character after I added instructions | Your custom text replaced the default line “Never decide or write for the user” | Insert Default and add your lines to it |
| The story became strange after I added the component | The component is added but empty, which sends no instructions at all | Fill it, select Model Default, or delete the component |
| The AI ignores the last lines of my instructions | They were cut. AI Instructions are trimmed from the end when the always-sent parts pass 70% | Shorten Plot Essentials. Put key lines first. Check View Context |
| The AI keeps bringing up one detail | It is in Plot Essentials, which primes the AI to mention it | Remove it, or move it to a Story Card |
| An old event keeps happening again | It is written in the present tense in Plot Essentials | Rewrite it in the past tense and label it as past |
| A “do not” line makes the problem worse | The line names the thing you do not want | Rewrite it as what you do want |
| The AI loses track of the current scene | The Author’s Note is too long and sits between the scene and your action | Cut it to 3 or 4 sentences |
| My Story Summary changed by itself | Auto Summarization rewrites it | Turn the feature off, or accept that your edits may be replaced |
| My saved instructions did not update an old Adventure | Saved sets are copies | Apply the set again in that Adventure |
| The Banned Words box does nothing | Latitude deprecated the feature | Use a positively worded line in AI Instructions |
On the last row, Latitude’s statement: “The Banned Words feature has been deprecated.” Its reasons are that the system did not work well with modern models, and that it believes “AI Instructions will be more effective at helping you guide the AI away from words and topics you wish to avoid”.
Frequently Asked Questions
What are AI Instructions in AI Dungeon?
AI Instructions is a Plot Component: a text box whose content is sent to the AI as a system prompt, before everything else, on every turn. It tells the AI how to write, for example the point of view, the style and what to avoid. If you do not add the component, AI Dungeon uses its own default instructions for the model you chose.
What is the difference between AI Instructions and Author’s Note?
The place where each one is sent. AI Instructions go at the very start of what the AI reads. Author’s Note is put near the end, just before your last action, inside square brackets. Latitude says this gives Author’s Note a more direct influence, and recommends keeping it to 3 or 4 sentences and putting most directions in AI Instructions.
What should I put in Plot Essentials?
Facts that are always true and always relevant: who your character is, who travels with them, where the story is set and what the main goal or threat is. Latitude’s advice is to keep it dense and short, to write past events clearly in the past tense, and to move anything that only matters sometimes into a Story Card.
How long can Plot Essentials be in AI Dungeon?
Latitude’s documentation gives no fixed length for the box. The real limit is the context of your model. Author’s Note, Plot Essentials, AI Instructions and Story Summary together may use at most 70 percent of it, and every token they use is taken from the story history, Story Cards and Memory Bank. On a 2,000 token model, 300 tokens of Plot Essentials is already a large share.
What are the best AI Instructions for AI Dungeon?
There is no single best set, and Latitude says the best instructions differ from model to model. The safest method is to press Insert Default or Insert Example, which loads Latitude’s own instructions, and then change one line at a time. Starting from an empty box removes the default lines that stop the AI from writing for you.
Why does AI Dungeon ignore my AI Instructions?
Often because part of them never reached the AI. When the always-sent components are larger than 70 percent of the context, AI Dungeon keeps Author’s Note first, then Plot Essentials, then AI Instructions, and cuts the end of whatever does not fit. Open View Context to check. The other common cause is a line that says what not to do, which Latitude says the AI follows badly.
What format should an Author’s Note use?
Plain text, with no brackets of your own. AI Dungeon already wraps it and sends it as a bracketed line that begins with the words Author’s note. Latitude’s examples are short and dense, such as a genre, a tone and a content rating in one line.
What is the difference between Plot Essentials and Story Cards?
Plot Essentials is sent on every turn. A Story Card is sent only when one of its trigger words appeared in recent actions. Use Plot Essentials for the few facts that must never be forgotten, and Story Cards for people, places and things that only matter when they come up.
Checked against Latitude’s AI Dungeon help pages on AI Instructions, Plot Essentials, Author’s Note, context, context management, settings and AI Model Differences on 7 October 2026. Quotations are from those pages. The token tables, the price of 100 tokens, the worked example of an overfull Plot tab, the reading that Plot Essentials suits facts and not commands, and the starting set are our own work from those published rules. We did not test instruction sets against each model for this guide. AI Dungeon is a trademark of Latitude; Arcanum is an independent publication with no affiliation to it, and references here are nominative.