The Overwrite Problem
The hardest game in AI-mediated discovery isn’t building a prior.
It isn’t defending one.
It’s changing one.
Because the internet remembers.
And the models inherit it later.
The Wrong Prior
Most discussions about priors assume two positions.
Either:
You have one.
Or:
You don’t.
But there is a third state.
And it may be the most common of all.
You have a prior.
It’s just the wrong one.
The model already understands you.
It simply understands an older version of you.
The company you used to be.
The category you used to occupy.
The story you used to tell.
The Sediment Problem
Businesses change.
They pivot.
They acquire.
They expand.
They reposition.
Leadership evolves.
Products improve.
New capabilities emerge.
But the corpus remembers.
Old articles remain indexed.
Old interviews still circulate.
Outdated reviews persist.
Legacy positioning survives in presentations, directories and third-party sites.
Every version of your past remains available.
And over time, those fragments accumulate.
Like sediment.
The result is a prior that may be:
→ sharp
→ coherent
→ confidently wrong
The Cruel Irony
The stronger the prior…
the harder it becomes to change.
Fresh retrieval argues for the new reality.
The existing prior argues for the old one.
And the prior can outvote the context window.
At least for a while.
This creates a strange situation.
The very thing that once gave you an advantage…
becomes the thing slowing your evolution.
You’re Competing With Yourself
Startups compete against obscurity.
Incumbents often compete against memory.
Not competitors.
Not algorithms.
Themselves.
Against:
→ their old category
→ their old positioning
→ their old reputation
→ their old customer stories
→ their old success
The challenge isn’t being discovered.
It’s being rediscovered.
Run the Overwrite Test
Ask multiple AI systems:
What is this company?
What is it best known for?
Who is it for?
Run the test with search enabled.
Then without search.
Pay attention to the answers.
Three outcomes are possible.
Accurate and confident
Your prior aligns with reality.
Uncertain
You may not have a strong prior yet.
Confident and wrong
You have an overwrite problem.
The system understands you.
But it understands the wrong thing.
And that may be the most dangerous position of all.
Because confidence can disguise drift.
Overwriting Takes Time
You cannot announce your way into a new prior.
You cannot optimise your way into one overnight.
You cannot erase years of accumulated sediment with a press release.
The loop is longer than that.
New experiences.
↓
New stories.
↓
New reviews.
↓
New coverage.
↓
New conversations.
↓
New corpus.
↓
Future weights.
The internet becomes memory.
The models inherit it later.
Overwrite happens through repetition.
Not declaration.
The Patience Problem
Modern businesses are conditioned to expect immediate feedback.
Campaigns produce clicks.
Ads produce impressions.
Dashboards update daily.
Priors move differently.
They evolve on training-cycle time.
Months.
Sometimes years.
Patience isn’t virtue in this game.
It’s structure.
The clock itself operates differently.
The Strategic Shift
There are now three prior games.
Build
You don’t have a prior.
Your challenge is to become distinctive enough to matter.
Defend
You have the right prior.
Your challenge is to preserve coherence.
Overwrite
You have the wrong prior.
Your challenge is to change what the world remembers.
Each game demands a different strategy.
The mistake is assuming they are the same.
Conclusion
Discovery used to reward visibility.
The prior era rewards understanding.
But understanding creates its own inertia.
The hardest thing to change isn’t what people think about you today.
It’s what tomorrow’s models remember about who you used to be.
Because the internet remembers.
The models inherit it later.
And sometimes the most important transformation isn’t becoming known.
It’s becoming known differently.