Spent today wiring our data lake: CRM + sales-call transcripts + community questions + GSC + competitor keywords, all feeding content clusters. The unlock isn't more content. It's knowing exactly which question to answer next. Way more fun to build than it sounds.
Pattern in our data: content with small imperfections gets cited more than polished AI copy. Models seem to trust the human seams. We stopped over-editing. Citations went up. Counterintuitive, but it keeps holding.
LLMs don't "read" your page. They retrieve chunks and stitch an answer. If your key claim is buried under a 2,000-word intro, it never makes the chunk. Front-load the answer: one claim, one clean sentence, near the top. That's most of the game.
Fresh content isn't always better than refreshed content for AI citations.
A 2-year-old article with updated data, added FAQs, and improved structure often outperforms a new article on the same topic.
LLMs weight authority and depth. A page with engagement history and ongoing updates reads as more trustworthy than something published last week.
Refresh first. Publish new second.
Most "AI SEO" tools are just keyword tools with a new sticker. Optimizing for an LLM isn't about density. It's about being the cleanest, most quotable source for a claim. The model lifts what's easy to lift. Build for liftability.
We fit €500K ARR a while ago. Wanted to introduce the team that got us here.
CTO: Julius Betzler — built 3 companies from 0 to 1 before this one
CPO: @augustGutsche — obsessed with product since his first semester at uni
Founding Engineer: Dora Pruteanu — top 1% CS at TUM
Head of GTM: Niccolo Casamatta — quit a six-figure IB job to be here
Building this team was harder than hitting the number.
The goal: own content production for AI Visibility in regulated, brand-sensitive markets.
Heads down. See you at 1M.
@GergelyOrosz Meta speedrunning how to destroy engineering morale while posting record numbers. The walk-back came too late for the talent already leaving.
Most "AI SEO" tools are just keyword tools with a new sticker. Optimizing for an LLM isn't about density. It's about being the cleanest, most quotable source for a claim. The model lifts what's easy to lift. Build for liftability.
@paulg Google is slowly turning every product into an ad delivery mechanism. Image search used to be genuinely useful. Now it’s just another place to interrupt the user with irrelevant sponsored results.
AI is one of the biggest levers for lifting the global poor we’ve ever had. Walking out on the guy leading one of the companies pushing it hardest feels like protesting the wrong target.
@Austen History’s richest people have been kings, colonizers, and monopolists. The current one sells cars people want to buy and rockets that actually land. Progress.
@GergelyOrosz Anthropic: ‘Build on our generous subs!’
Also Anthropic: ‘Actually, pay API prices now lol’
Then: ‘Okay fine, we’ll walk it back’
This is why devs are increasingly allergic to depending on any single frontier lab.
@ns123abc Anthropic spent years pushing for heavy AI regulation. Now they’re the ones stuck in the regulatory meat grinder they helped create. Classic monkey’s paw.
@brian_armstrong Option 1 (financial literacy test) feels like the pragmatic middle path. Option 2 is cleaner in theory but harder politically. Either one beats the current wealth-based gatekeeping