In 1787, Catherine the Great sailed down the Dnieper to inspect her new lands.
Her minister, Grigory Potemkin, had a problem.
The lands were a mess. Poor, empty, unimpressive.
So he built fake villages along the riverbank.
Painted facades. Happy “peasants” who were actually his own men. As the boat passed, he’d dismantle the village, race it downriver, and rebuild it round the next bend.
The Empress was delighted. Look how prosperous we are.
It worked. For about a week.
Then everyone found out, and “Potemkin village” became a phrase that’s outlived the man by 240 years.
That’s the thing about manufactured reality.
It always looks brilliant right up until the boat stops moving.
I thought about Potemkin this week.
One of my team flagged that a competitor of ours is building villages.
Not on a river. On Reddit and Hacker News.
One post a day. Every day. Always from a different faceless account.
Always one of three things.
“I solved this exact problem and used their product.”
A neutral-sounding “top five API gateways” list, where, funnily enough, the same name always wins.
Or a long, AI-generated “paper” that cites their own content as a source.
Some posts even have fake comments. Other accounts turning up to praise the original. The peasants, waving from the riverbank.
My first instinct was competitive.
There’s a version of me that wanted to build a few villages of our own.
I’m glad I sat on it.
This was never about Reddit
Here’s what most people miss.
The competitor doesn’t care if a human reads those posts. The engagement is terrible. The accounts are nobodies.
As a way to persuade a buyer today, it’s worthless.
But that isn’t the game.
The game is training data.
People don’t search any more. They ask. They ask ChatGPT, Claude, Gemini “what’s the best API gateway for AI agents,” and they take the answer as fact.
There’s a whole industry growing around this now. They call it Generative Engine Optimisation.
And the models learn what to say partly from what they read across the web.
LLMs trust what they see repeated. Say the same flattering thing often enough, in enough places, and you hope the next model swallows it whole. Some people call it the “consensus engine” effect.
So that’s the play.
Build a fake consensus. Paint the villages. Hope the Empress sails past.
It’s keyword stuffing. Reincarnated for the age of AI.
Why the villages fall down
A colleague of mine lived through the early SEO wars.
He said something simple.
We’ve seen this film before.
There was a time you could stuff a web page with hidden keywords and rocket up Google.
It worked. Briefly.
Then Google got wise, the cheats got buried, and everyone who’d bet the farm on it had to start again from nothing.
The same correction is coming for AI search.
For three reasons:
Firstly, the platforms have to stop it. An answer engine that recommends products based on bot-farmed nonsense is a worse product. The model makers will hunt down coordinated, zero-reputation content, because their whole business is being trusted. Even the honest GEO crowd admit it: anyone guaranteeing AI placement with bot farms is lying to you.
Secondly, fakery is fragile. A post from a nobody account goes nowhere. No upvotes. No replies. No links. No citations. It sinks. A genuinely useful answer gets quoted again and again, and that’s the durable signal the models weight most. You can’t fake a reputation for being helpful. You can only earn one.
Finally, the dishonest version is dangerous. Anonymously trashing a named rival isn’t clever marketing. It’s a legal risk with your logo hidden behind it.
I’m completely happy making a hard case against competitors on our own comparison pages. In public. With our name on it. Facts evidenced.
That seems fair and useful, sock puppets are not.
What we do instead
A couple of weeks ago I set up a small task that scours Hacker News and Reddit every morning and surfaces threads where someone has a real question we can actually help with.
Most of what it finds is junk. Including, beautifully, the competitor’s own fake posts.
But nearly every day it finds one real person.
Someone asking “how do I get off Apigee without losing my mind.”
And I can drop in a few sentences that genuinely help.
Sometimes I recommend a tool that isn’t even ours, based on more than a decade in this space. Sometimes I mention us, when it’s relevant.
That’s the whole strategy.
Be the most useful voice in the room. Consistently. With your real name on it.
If the models learn from that, wonderful.
If they don’t, I’ve spent my time helping people instead of painting villages.
The honest bit
I won’t pretend the cynical route has no appeal.
There’s a real fear of being outflanked. If they game the indexes for eighteen months and we don’t, maybe they take mindshare we never win back. “We took the high road” is cold comfort when you’ve lost the quarter.
I take that seriously.
I will happily run a series of genuinely good, clearly-authored pieces to make sure we show up where buyers and models are looking.
That’s fine, but the line I won’t cross is anonymity plus dishonesty.
The moment your tactic only works if nobody knows it’s you, you’ve already told yourself everything you need to know.
The internet has spent thirty years learning that fake consensus gets found out.
AI hasn’t repealed that law.
It’s just moved the boat faster.
Be useful. Sign your name.
Further reading:
- What is Generative Engine Optimisation? — Search Engine Land
- Generative Engine Optimisation: how to win in AI search — Backlinko
- The 2026 guide to AI search visibility — LLMrefs
- Tyk’s approach to governing AI traffic — Tyk AI Studio
- How Tyk compares on API management — Tyk
