Most retailers think about personalization as product recommendations or email targeting. You’re describing something that reshapes the storefront itself in real time. What signals are you actually acting on, and what does the shopper experience that they wouldn’t on a static site?
Product recommendations and email targeting operate around the edges of a static page. The description, the hero copy, the way the catalog is organized, none of it changes based on who’s viewing. We reconfigure that core layer itself.
We act on a few buckets of signals. How someone arrived: the ad they clicked, the campaign, the intent behind it. What the merchant’s CRM knows: first-time versus returning, loyalty status, past purchases, shipping preferences. Real-time behavior on the site, what they dwell on and what they skip. And for some categories, external signals like local weather.
The result is a storefront that speaks to why someone showed up. Click an ad for a specific concern and the page leads with that concern, not a brand story you scroll past. On a static site, every visitor gets the same page and has to figure out whether the product fits. We flip that and let the page match itself to the visitor.
You’ve made the case that the structured-content foundation behind real-time personalization is also what makes a brand discoverable to AI engines. Walk me through that connection, why does solving for personalization today set a retailer up for AI visibility tomorrow?
The same need lies beneath both problems: product data structured as meaning, not just fields.
To personalize in real time, a title and a price aren’t enough. You need to know what a product is, what it solves, who for, and how it relates to everything else. We hold that in a knowledge graph connecting a product to its use cases, ingredients, the customer goals it serves, and the situations it fits. That’s what lets us serve the right page to the right visitor.That’s exactly what an AI engine needs. When ChatGPT or Perplexity makes a decision about whether to recommend a product, it isn’t influenced by marketing copy; it’s parsing structure: entity relationships, factual attributes, fit to a specific need. So, a brand that has structured its catalog for humans has already made itself legible to the machines. Because it’s the same data problem. You solve for revenue today, and visibility is a byproduct.
Product discovery is shifting from search engines to AI assistants like ChatGPT and Perplexity. For a brand that’s already built this structured, dynamic foundation, what changes about how they show up in an AI-generated answer?
In a traditional search result, a brand is a link, and the shopper evaluates it by clicking through. In an AI-generated answer, the engine evaluates first and decides which products to name and how.
For a brand with a structured foundation, the engine has something concrete to reason about. If a shopper asks for a specific size, price range, or ingredient, the engine can match against data that represents those dimensions and name the product with specifics. Without that structure, you’re asking the engine to infer everything from prose, and often it just doesn’t, so the product never enters the answer. It isn’t rejected; it’s never even a candidate. What changes is your eligibility to be in the answer at all. Structure takes you from invisible, to a product the engine can name with confidence.
Generative Engine Optimization (GEO) gets talked about as a successor to SEO. Where does the old search playbook actually break down once an AI engine, rather than a search index, is deciding what to recommend?
The old playbook was built for a system that ranked pages by keyword match and signals like backlinks. The tactics optimized for that machine: link building, keyword density, exact-match phrases.
An AI engine isn’t ranking pages against a query. It’s building an understanding and generating an answer. That breaks the playbook in two ways. Keyword stuffing is now actively harmful, because what makes text trustworthy to a language model is factual clarity and coherence, not term frequency. And a lot of SEO was about winning the click, but in AI discovery there may be no click at all, the product gets recommended inside a conversation and the shopper acts on it. So, the game stops being about click-through position. It becomes about whether the engine understands and trusts your product.
Smaller retailers assume they can’t compete with Amazon or large platforms on visibility. Where do you see openings for them in a model built on real-time personalization and AI discovery that didn’t exist in the search era?
In the search era, visibility was a function of budget and domain authority. Big platforms outspent everyone on SEO and ads and amassed backlinks a small retailer couldn’t match. Resources tipped the field.
AI discovery changes what the engine rewards. It’s looking for the product that best fits a specific, often narrow, intent. That’s an advantage for a small retailer with real depth in a niche, because a focused grooming or wellness brand often understands its customer better than a giant generalist does. If that understanding is well-structured, the engine can see it.
The other opening is speed. Small teams can restructure their content faster than a large platform can turn its ship, and there’s a window right now to establish yourself before the category crowds. Big platforms still have real advantages, but the advantage has shifted from raw resources to depth and speed, and those are things a focused small retailer can compete on.
You spent years building NLP systems at scale at Dataminr before founding Gigit.ai. What from that work carried over most directly into how you approach personalization and AI visibility for merchants?
Two things carried over the most:
- First, treating real-time constraints as a first-class problem. At Dataminr, the work was detecting real-world events from huge, noisy data streams with very little latency to spare. Reshaping a storefront the instant a visitor lands is the same class of problem: the intelligence has to act before the page renders, not overnight.
- Second, encoding meaning in a machine-reasonable form. My work was turning unstructured text into structured signals that downstream systems could act on reliably, at scale, without a human in the loop. That’s the same craft as turning a merchant’s catalog and content into a knowledge graph. The domain changed, from event detection to product data, but building a representation of meaning a system can reason about is exactly what I spent years on. Most of the work we do at the foundation layer is a direct application of that.
Measuring brand presence inside AI-generated answers is still an unsettled problem. How are you thinking about what’s actually measurable today, on both the personalization ROI side and the AI-visibility side, versus what the industry is overpromising?
The two sides are at very different stages, and it’s worth being blunt about the gap.
On personalization, the measurement is solid and it’s what e-commerce has tracked for years: conversion rate, revenue per visitor, add-to-cart, all against proper controls. You can attribute the lift with confidence, and the gains across grooming, beauty, and wellness have been significant. That’s real, measurable ROI.
AI visibility is genuinely unsettled, and anyone selling clean attribution is ahead of reality. A recommendation can happen inside an AI conversation with no click and no referral, nothing that hits your analytics. You can track whether products surface for given prompts across engines, but not tie it to revenue with the same rigor. My honest take: visibility alone is close to a vanity metric until attribution catches up, and much of the industry overpromises by presenting it as solved. What I’m confident in is the foundation: the content driving measurable gains today is the same asset that drives discoverability tomorrow.
