Players wanted this post so I vomited it forth in like a few hours, I'm mildly dissatisfied with it but oh well
I definitely had my share of campaigns dying after 3 sessions back in the bad ol days of being 12 years old. But in my time as an online GM, I feel like I've had a pretty good success rate in terms of campaigns I make being fairly long-lived and usually reaching conclusions of some kind rather than fizzling out. Part of that is simply that, apart from random freelance word and art stuff, this is what I do, so I have far more time and energy to devote to a campaign compared to people who have jobs or relationships or children or whatever. 'Become a minimalist neet freegan D&D ascetic' isn't really applicable advice for most people, but I feel like it should be mentioned as a caveat or disclaimer.
I have an advantage of having slowly grown a group of regulars, one or two from each new game, who seem to generally enjoy my GMing style. There is a level of trust and shared experience here that random roll20 or OSR discord pickup games don't have, heck, even when my players run for each other, the dynamic changes and so there's more chances of flaking. It could be that there's no meaningful procedural difference in how I start games compared to how others start games, and it's more a matter of social dynamic than anything.
And as final caveat is that players will sometimes just lose the ability to play for whatever reason. We've all been in games killed by scheduling, I've been the player in games who just came less and less due to diminishing interest even in games I thought were good (Acodispo and Spwack's games come to mind), etc etc. Games are not immortal* and I think a good step one when making a campaign is to think 'what would be a good point to end this game on' because if you know roughly where it is going, you may get an idea of where a good place to start it off at might be.
*the open table flailsnails multi-gm shared universe meta-game my server engages in might be immortal as it doesn't have the same death-conditions as regular campaigns but I digress
But for what it's worth, here's the process I've gone through
Initial prep for campaigns usually is a sort of slush pile of handwritten notes written while I donate plasma. This initial slush is anything... 'anorexic centaur noble shamed for her natural weight by human villain who is a toxic friend' 'alternate dungeon entrance is gay pirate brothel' 'time loops for free character revives' 'biomechs actually are nanite swarms mimicking life' 'hot frog' etc etc
I think it's helpful to get these half-formed ideas noted down, so your brain can either move on, or properly start to develop them further. My brain, at least, can chew on a half-formed thought for a very long time without really advancing it anywhere or seeing problems to fix.
Then comes the more serious prep- where 'castle overrun by frog cult' has to actually become a map with notes on what's in each room and so on, 'fairy forest' has to actually get an encounter table and some hex fills, and so on. While you can go wild with big hexcrawls and megadungeons prepped in advance, I think it will usually pay off more to go more in depth with nearby stuff first- prep some town NPCs in the starting village instead of the next town over, add some small 'quests' the players can do, and let distant lands be more vaguely sketched so as not to expend prep.
Also, have minidungeon modules and/or procedural content generators ready for when the players go somewhere less-prepped. Random encounters go here I suppose, but I think they're very good shorthand for populating a world. If I could only prep one thing, I would prep a wandering monster table and treat everything I roll as though it was something I placed on purpose, with motives and backstory and connection to other known things.
Similar advice applies to games with more linear story arcs- prepping the whole thing in advance leads to railroading, so having a vague idea of story beats, but only actually prepping 1-3 sessions in advance lets you keep things reactive to the PC's actions and allow for them to execute their own plans, while still generally heading towards things like 'defeat Ser Hotsalot in the Volcano Tower' or whatever without being opposed to 'players ambush Ser Hotsalot at the pub' instead because you already had Volcano Tower prepped 6 months ago.
I think it's easier to prep as a campaign is being run rather than in a void, because your ideas bounce of the players and they inspire you and you can insert personal moments to shine as things go on.
Uh... speaking of ideas in a void, I'm just going to throw down some specific prep examples for games, then see if there's any keyrecurring features.
Esoteric Oroboro (Or Esoboro) had some procedures I could start off with, rolling an underworld, both as dungeon map and faction relationship web. Esoteric Enterprises has a LOT of tables that help carry the weight of ensuring there is a world to interact with, and was actually a fairly low-prep game thanks to that. I was able to use my older campaign's lore as prep for this campaign, which is such a neat trick that I highly recommend GMs try to set their subsequent campaigns in the same universe unless you absolutely must discard it for a fresh start.
As mentioned, I had a megadungeon and many factions rolled up via tables.
I had to make spell lists for my local gods for the sake of Mystic (Cleric) players, which doubled as spell lists for potential rival cultists.
I had a bonus 'ancient evils escaped from the Reliquary super-prison' running wild through the city- essentially a random encounter table that made a new major threat each session. I was fairly excited for all of them, which is good advice for encounter tables- if you the GM aren't thrilled to roll a result, maybe change the tables to be less realistic and more dramatic.
I had 3-5 jobs lined up for the PCs to introduce them to factions, get them into the dungeon, and get them paid.
Prep for this game took about 2 months and included the following-
A poll offering different campaign choices to the players to ensure there was buy in to the premise before prep started in earnest
A map of the island and mainland, re-used from Betrayal at Queen's Coast
Encounter tables for all regions
Weekly 'events' for the school, a 1d20 table I would roll on 3 times
A megadungeon beneath the school, just some sloppily generated online dungeons maps populated with a simple '1/3 chances of monster, trap, or treasure'
Monsters being drawn from a custom wandering monster table to give the depths a wizard-school basement being raided by supernatural thieves' vibe- animated furniture, flying books, sneaky fey gremlins, and nightmare wizards hinting of something more
Traps being a similar list, and treasures being from my own treasure tables, which were just modified AD&D/BFRPG tables
The incredibly tl;dr glog wizard post a few posts back, made to make glog wizards less self-contained gimmicks and more like a 'wizard' who can potentially learn more things, as well as offering some milestone goals to incentivize behavior beyond money grubbing to increase personal power
Some playtest games with the players to ensure said glog wizard post wasn't complete hokum
A list of tasks they were to complete by the end of the year- 12 spells to form the new 'true' glog class, various dangerous wizards to defeat (again drawn from a past blogpost), 3 abstract goals of solving mysteries and teaching students, and a couple of bonus side quests
And a list of students they were to teach
Betrayal at Queen's Coast
Had a map and random encounters, to make traveling the land recruiting noble aid more exciting
Had a starting scene of being pursued across the land, only to be caught in a haunted mansion and the menaces of the haunted mansion ready
Had some 'villain' counterparts of the heroes ready as minions of the villainess
But then mostly prepped ahead of the players 1-3 sessions in advance as mentioned, reskinning modules and dungeons, making or improving scenes...
It becomes difficult to reconstruct what was prepped before-game and what was prepped mid-game as I gaze upon older campaigns, but there does seem to be some reliable markers of what I prep
What I Prep
-A starting SCENE to start the players in the action without mucking about getting to know each other in a tavern
-A starting DUNGEON, usually tied to the above, to give the players something familiar to do immediately.
-Starting CHARACTERS and/or FACTIONS for the players to roleplay with and be made aware of existing power structures and motivations
-A starting TOWN with SERVICES or at least downtime activities, so the players can prepare
-a MAP and ENCOUNTER TABLES to give a sense of a place and a buffer of content to allow the players freedom of action.
-EVENT tables, or a TIMELINE of expected events. At a minimum this is a weather reaction roll and my calendar blogpost to track time, but events of politics, looming threats, or local flavor all are useful.
-An OVERARCHING GOAL, that, while not immediately pursuable, gives direction to the game. It is essentially the diegetic version of the pitch for the game- if you the player want to play this game, you the character should be interested in this goal somewhat.