I'm building the product that gets B2B companies cited inside AI search engines.
Not ranked. Cited.
There's a difference - and most marketing teams haven't figured it out yet.
CPO at SCAILE. Posting what I learn building at the intersection of LLMs, content and B2B distribution.
Follow if you're into the technical side of how AI engines actually decide what to trust.
xai took grok 4.5 public this week, a cheap cursor-trained coding model at $2/$6 per million:
- another near-frontier model at commodity price
- another reason "which model" is the wrong question
- what matters is whether any of them can find and read your content
the models keep getting cheaper. being citable is the part that doesn't.
Lipocheck is a health app in a category owned by names with 100x their budget. They came to us because paid was getting too expensive to acquire a single download.
We built their AI search layer. Within weeks, downloads started coming from people who'd asked an AI what to use, gotten Lipocheck's name, and installed it. No ad in the loop. The lesson wasn't about health apps. In a category you can't outspend, being the answer is the only affordable way in.
how models decide your page is current enough to trust:
1/ a visible, real date on the page, not just in the sitemap
2/ recent internal links pointing to it
3/ the content references current facts, not 2023 ones
undated pages read as risky. the model quietly prefers the source that looks maintained.
→ free scan of your site's AI readability: https://t.co/69Qi6Lp5zI
client's best proof point, a 40% improvement, lived only inside a chart image. looked great to humans. to the retriever it was a blank rectangle with alt text that said "chart." their strongest number was invisible to every model. we put it in a plain sentence. it started getting quoted within the month.
quick test: could a model answer the buyer's question using only your page, nothing added?
☐ the question is stated somewhere on the page
☐ the answer is a plain sentence, not implied
☐ no "it depends" without the depends spelled out
☐ no key detail living only in an image or pdf
if it has to guess, it'll guess with someone else's page.
→ free scan of your site's AI readability: https://t.co/69Qi6Lp5zI
miora came to us invisible in their category. Inside 15 days they were the #1 answer in AI search for their space, ahead of MyFitnessPal. No ad spend, no relaunch. They just became the name the model gives when someone asks what to use.
I help companies get named by ChatGPT, Perplexity and Google's AI when buyers ask for the best option in their category.
the single highest-roi change i make to a client page: a two-sentence answer at the very top, before the intro, before the context. models grab it, humans skim to it. everyone writes the payoff last. write it first.
HeyHoney is a DTC brand in a crowded consumer category. The kind of space where everyone assumes the game is paid ads and influencer seeding.
Here's what actually moved the needle. Buyers in that category increasingly ask an AI for recommendations before they ever hit a product page, especially for anything they're slightly embarrassed to google. If the model doesn't mention you, the ad you paid for is just retargeting people who already found you.
We built the content that got HeyHoney into those answers. Over 600,000 organic visitors followed, and more than 150,000 in new ARR, none of it from raising ad spend.
The takeaway isn't "do content instead of ads." It's that in 2026 the recommendation happens inside the AI, and if you're not there, everything downstream costs more. The moment the model decides who to name is the one worth owning.
https://t.co/BM0hbMAEaF
spent this week making a client's docs read cleanly to a model instead of to a human skimming. citations roughly doubled in ten days. the content didn't get better, it got easier to quote.
we build the system that gets AI models to actually cite your company instead of a competitor.
spent yesterday building out a batch of "x vs y" pages for a client and watched them start getting pulled into answers within days.
makes sense once you see it. when someone asks an AI "is x or y better for [use case]", the model wants a page that already answers exactly that. a comparison page is the query, pre-written.
most companies won't name competitors on their own site. turns out that's the exact page the model reaches for. we've been publishing the pages everyone else is too polite to write.
what a page has to pass before we ship it:
☐ answer in the first 200 words
☐ every paragraph stands alone out of context
☐ dated, and actually recent
☐ nothing that contradicts the client's other surfaces
☐ a claim a neutral third party would back
miss three and it won't get cited. writing quality won't save it.