What a week.

Three firesides, two parties, and one omakase dinner later, we bid SF Tech Week farewell. If you joined us at an event for the first time, welcome. If not, keep an eye out in our weekly newsletters for future programming.

We’re in LA this week. No planned Unicorner programming but you can catch us at events. Reply letting us know where we should meet you.

Ethan and Arek 🦄

When a buyer asks ChatGPT, Claude, Gemini, or another AI assistant which product to choose, the model names a few brands and skips the rest. Petra Labs gets enterprise brands onto that list, combining proprietary technology with an embedded team that works as an extension of the client’s marketing department. Its platform builds a custom map of the prompts each client needs to win, starting from real AI and Google search data and the language buyers use on platforms like Reddit and LinkedIn. It then connects that map to the client’s own data, such as CRM records and sales call transcripts, to pinpoint the exact pain points customers have. The platform also tracks which sources the models rely on, including the ones they check while “thinking” that don’t appear in the final answer. Petra’s team then does the work, from content and technical fixes on the client’s site to press placements, to put the brand among the first names in the answer. Finally, a custom attribution model, built for each client, finds the AI-driven visits that standard analytics count as direct or organic traffic and connects them to pipeline and revenue.

Check it out: petralabs.com

You’ve had that idea for months. Maybe years. The one you keep meaning to build once you learn to code and find the time.

You don’t need to wait anymore. 

With bolt.new, you describe your idea in plain English, and it gets built in minutes. No coding knowledge required. No developer to hire. 

Now, the only thing standing between you and your idea is an excuse. And that excuse doesn’t exist anymore.

It’s time to stop sitting on your craziest ideas and start shipping them.

Build it now.

This is sponsored content.

Petra runs an agency model priced by the effort it takes to win a category. Engagements start at $15,000 per mont. Most clients pay between $250,000 and $600,000 (or more) per year depending on how many brands or product lines Petra tracks, how competitive the queries are, and how complex the attribution setup is. Petra’s team handles everything from strategy to measurement. It ensures its clients never compete against each other by working with only one brand per category.

Sam Akkawi spent five years at Strategy&, PwC’s strategy consulting arm, advising private equity firms on more than 50 deals in marketing data and technology. Across thousands of conversations with marketing leaders, he kept hearing the same complaint about how nobody could tie AI search spending to revenue. Rishi Penmetcha and Ramie Khoury came from Traba, a startup that supplies on-demand workers to warehouses and distribution centers. There, Penmetcha built AI agents after engineering roles at Palantir and Ironclad, and Khoury led the growth team. The three started Petra in 2025, running their first client programs by hand before automating and scaling the process. 

Brands spent 15 years learning to rank. Now they have to learn to be recommended. 

For years, search meant a page of 10 blue links, and SEO was how brands won it. Google's ranking inputs (backlinks, keywords, page content, site health) behaved the same way across industries, so agencies and tools could build businesses on shared playbooks. That model depended on clicks, but clicks are falling now. Google users clicked a traditional result 8% of the time when an AI summary appeared, compared with 15% when none did. About 60% of searches now end without a click, cutting organic web traffic by 15% to 25%. 

The demand is moving into the answers themselves. ChatGPT passed 1 billion monthly users in May 2026, and AI-referred traffic to U.S. retail sites grew more than 1,200% between October 2024 and July 2026, starting from a small base. Those visitors are worth more: they converted 60% more often than non-AI traffic, and spent 59% more time on the site. Lastly, AI-referred visitors are 28% more likely to add items to their shopping carts. It’s hard evidence that showing up in AI search matters more and more to brands, and they’ve been quick to catch on.

Petra's founders estimate that seven months ago only 30% to 40% of the companies they spoke with believed AI search mattered, compared with 90% to 95% today. Most have responded by buying monitoring tools or trying to work out Answer Engine Optimization in-house. AEO is the work of getting your brand to show up in the answers AI tools give, the way SEO gets you onto Google’s results page.

Petra bets that brands get stuck on attribution, i.e., knowing which actions, like PR spend or influencer deals, moved revenue. LLMs learn from Reddit threads, press coverage, YouTube, and product docs. And what brings traffic, conversions, and sales for a sportsbook, for example, may do nothing for an insurer. So you can follow every traditional SEO best practice and still get nothing. In other words, the SEO playbook was shared, but in AI search, a shared playbook does not result in a win for everybody. 

Petra went in the opposite direction.

Zapier, which built one of the best-known SEO programs on the internet, tried other AEO vendors before hiring Petra to write for its blog and guide its video team. Another client, Novig, shared how, thanks to Petra, the prediction market went from almost no citations to the most-cited sportsbook operator in general betting questions that don't mention a specific brand in about three months.

AI platforms increasingly restrict third-party access and keep their analytics in-house, which leaves marketers without an equivalent of Google Search Console or Meta Ads Manager for AI search. Answers are also inconsistent. Asked for brand recommendations repeatedly, ChatGPT and Google's AI have less than a 1% chance of returning the same list twice, and Google's AI Mode and AI Overviews cite different sources for the same query 87% of the time. Different engines, and even different ways of accessing the same engine, recommend different brands, while changes to a brand's site can take days to show up in answers. That means a tracker that checks one engine a handful of times is mostly measuring noise. 

Petra fills that gap with three systems. The first is an attribution engine that logs every intervention, from a press placement to a Reddit thread, and ties it to changes in AI recommendations, site traffic, and revenue. The second runs large-scale simulations of purchase-intent prompts across models to see which brands and sources get recommended. The third is a classification layer that tracks brands, products, and features in AI answers and which sources are gaining or losing influence. 

Since a jump in traffic can mean a qualified buyer for one company and a non-qualified one for another, Petra builds each attribution model around the client's own data. The result shows a client which pieces of work moved its numbers, something a visibility dashboard can't do. 

Petra's next stage starts with converting launch demand. Its July 2026 launch produced over $8 million in pipeline, and the founders' near-term focus is closing it through referrals and a tighter sales process. Because Petra takes only one brand per category, it will grow by entering many categories, betting on each client being large enough to justify the exclusivity. 

Besides that, if brands trust Petra’s attribution models to show which marketing initiative brings more AI revenue, Petra will be less an AEO vendor and more the place companies decide how to spend their marketing money. 

The longer-term step is turning attribution into a product, which is what would let a two-person team serve hundreds of clients, as the founders intend. Petra has published how it measures visibility, but clients will rely on its models without a team around them only once it can show, with evidence beyond its own case studies, that those models separate AI search from the rest of a client's marketing.