A few weeks ago, a data audit of my brand’s AI visibility produced a number I had to read twice. My company carries twenty-five SKUs. ChatGPT had catalogued more than seven hundred.
The same LLM a customer asks to recommend a product for a specific benefit can’t figure out which SKUs belong to which brand, because legacy product names, retailer-feed variations, and inconsistent metadata have fractured a company into hundreds of weak signals across the web.
AI models are doing the work of a customer-service rep, a salesperson, and a category expert all at once. To those models, too many brands look like noise. And chances are, yours is one of them.
Are Consumers Truly Discovering Products Via AI?
The shift is structural, not a glitch. AI platforms like ChatGPT, Gemini, Claude, and Perplexity are now the first stop in a growing share of consumer product discovery, and they don’t retrieve brands the way Google search originally did. They retrieve solutions, recommending purchases at the product level based on a structured catalog of attributes the model can actually read. If the model can’t figure out which attributes apply to your product, you simply don’t show up.
AI shopping traffic is rising fast, and the impact is showing up in how consumers discover, evaluate, and spend. According to Adobe Digital Insights’ April 2026 Quarterly AI Traffic Report, 39% of consumers have used AI assistants for online shopping, AI-driven retail traffic increased 393% year over year in Q1 2026, and AI-driven visits delivered 37% higher revenue per visit than non-AI traffic in March 2026. An April 2026 report from EMARKETER and Publicis Commerce shows the same shift is influencing purchase decisions: nearly 1 in 5 shoppers now start their journey inside an AI assistant, 23.3% view AI as a recommendation engine, and 49% would consider a different brand or product if AI recommended an alternative.
How Did Product Discovery Shift From “Brand” To “Solution”?
With LLMs, product discovery works differently than it did with legacy search. Consumers are no longer typing keywords and clicking through pages of results. Instead, they’re inputting their exact parameters and letting the LLM do all the research, returning three to five results to choose from.
On the backend, the LLM deploys fan-out queries across the web, retail catalog feeds, structured data sources, and trusted citation hosts. From there, it composes a recommendation from product-level attribute matches. There is no “brand homepage” stop. There is no “first-position SEO play.” There is a structured comparison of attributes scored against a stated need.
Brands fail this comparison in three predictable ways, according to the experts I interviewed:
1. They are invisible – the AI can’t read the catalog feed at all because it’s malformed, missing, or behind a login wall
2. They are ignored – the AI can read the feed but the attribute data is so low-quality the product never makes the recommendation cut
3. They are misaligned – the data is clean enough to surface, but the product lands in the wrong category, in conversations the brand can’t win.
When evaluating partners to help solve this, look for a B2B data platform focused specifically on AI-discovery optimization rather than traditional SEO: one that can audit your catalog against the attribute schemas AI models actually read, identify which category neighborhoods your products are landing in, and rank fixes by impact on recommendation share.
For CPG brands, look for a partner that understands the variable attribute schemas in your category: supply-chain provenance, ingredient sourcing, certifications, allergen flags. CPG arguably has the most ground for an AI model to get wrong, which makes choosing the right data infrastructure partner a high-stakes decision.
On the other hand, there is the greatest opportunity here for smaller companies to have an outsized impact. LLMs aren’t evaluating market share when recommending products.
The product with the cleanest structured data and the most trusted third-party citations wins, regardless of legacy market share
Why Are Some Brands Competing In The Wrong “Neighborhoods”?
The most common version of the misalignment failure mode is one brands rarely see coming. When product data isn’t structured consistently, an LLM can’t pattern-match and determine that two listings with slightly different names are one and the same product, and the resulting confusion can put a brand in entirely the wrong category.
Because AI models cluster product categories in ways that don’t match retail-shelf logic, brands routinely discover their products are competing in conversations they have no business being in, while missing the conversations that should have been theirs by default.
In one analysis conducted by the CPG AEO platform Novi, 85% of laundry detergent products that AI could not correctly place were routed to the dishwasher detergent shelf, even when the model was given a high-confidence category label. The reason: both categories rely on nearly identical attribute language, including “surfactants,” “enzyme action,” “cleaning power,” and “concentrated formula.” If the content says “powerful clean” but doesn’t clearly say “clothes” or “dishes,” AI collapses the two categories together. AI reads what is written.
The opposite can also happen. Plant-based milk and nut butter share ingredient language (almond, cashew, macadamia), which caused 16% of misclassified plant-based milk products to initially route to the nut butter shelf. But when the model was given maximum confidence in the category label, every plant-based milk product landed correctly. The word “milk” gave the model a clear beverage signal. The broader lesson: category labels only help when they carry a distinct semantic cue. Otherwise, AI follows the attribute language it can read.
What Trust Signals Are AI Models Actually Reading?
When an LLM builds context around a product recommendation, it doesn’t weigh all sources equally. It weighs the sources with the most credibility. A large advertising budget means nothing to an LLM. The keyword optimization SEO teams have invested in for years isn’t sufficient either. What matters is whether trusted third parties have described your product in legible, attribute-rich language the model can index.
ChatGPT leans heavily on Reddit. In some categories, paper goods and wipes among them, Reddit accounts for 52% of citation sources. But the move that gets a product cited is not what most marketing teams assume. ChatGPT doesn’t browse Reddit in real time, and upvote count is not a ranking signal. The post title, indexed by the underlying web search, is what matters. A well-titled Reddit thread is an AI-engine optimization (AEO) artifact, not a social-media one.
Gemini leans on YouTube, which routinely outranks major retailers as a citation source, specifically product-review videos that carry a reputation for unbiased feedback. A brand without a presence in the YouTube creator review economy may be invisible to Gemini’s recommendation engine regardless of retail shelf placement.
What Should Brands Do Tomorrow Morning?
When brands first discover the gap between how they see themselves and how AI platforms see them, the next steps can feel overwhelming. But the root cause, and the first place to focus, is almost always the underlying product data.
Start with a catalog audit. Run your top twenty SKUs through the major AI shopping assistants and document the discrepancies: wrong product counts, wrong category placements, missing attributes. That output becomes your prioritized fix list.
Next, clean the feed before amplifying anything else. A clean structured-data feed, with consistent naming, complete attribute schemas, and no login-walled listings, is the foundation every other tactic builds on. Third-party citations and AI-optimized content compound quickly once the catalog is legible to the model.
Then seed the trust sources strategically. A well-titled Reddit thread in the right subreddit and a single thorough YouTube creator review can outperform months of paid media for AI visibility. This is not influencer marketing logic. It is structural. The model will find it.
Established brands need to audit their portfolio for the three failure modes and fix catalog feeds before competitors do. Insurgent brands have a rare window: the established competitor’s ten-year SEO moat doesn’t transfer to AI. Retailers have the sharpest incentive of all: their own AI visibility is downstream of the brands they carry.
Brands that treat AI discovery as a data infrastructure problem (not a marketing one) will capture the recommendation share that legacy players leave on the table. The next product question your customer types won’t be typed into Google. The brands that have done this work will be the ones that show up when it matters.