Imagine going to a restaurant and the waiter just tells you what to order. You don't read the menu, you trust the recommendation. The decision space collapses from twenty options to one, and the friction of choosing is gone. That's roughly what's happening in how people shop for consumer products online — they're no longer browsing five sites and comparing; they're asking an AI what to buy, taking the recommendation, and moving on. And it's not a fringe behavior: AI-driven traffic to US retail sites rose 393% year over year in Q1 2026, and over the 2025 holidays AI referrals converted 31% more than other traffic with revenue-per-visit up 254%. For CPG brands this is a structural change in product discovery — and most aren't set up to be the one the AI recommends.
Key takeaways
- People increasingly ask an AI what to buy and take the answer — ChatGPT hit ~800M weekly users and added in-chat checkout for Shopify merchants.
- The traffic is high-intent: AI referrals converted 31% better over the holidays.
- Classic search is becoming answer-first too: Google AI Overviews grew from ~13% of queries in March 2025 to ~34.5% by December, with 26% of those sessions ending in zero clicks.
- Most brands are invisible to it — retail product pages are only ~66% machine-readable to LLMs.
- Getting recommended is knowable work: legible product pages, structured data, and third-party corroboration the AI trusts.
What is actually changing
For two decades, discovery worked one way: search Google, get a ranked list, click through several options, compare, decide. That model assumed the customer wanted to do the comparison work. It's breaking down. A growing share of discovery now happens inside AI assistants — ChatGPT, Perplexity, Claude, Gemini. The customer asks "what's the best gut-friendly soda?" or "a good electrolyte drink with no artificial sweeteners?" and instead of a list gets a recommendation — two or three options with the reasoning attached — and often just takes it. Nearly half of US consumers (49%) used AI in shopping at some point in 2025, and Perplexity's shopping-intent queries jumped 5x after it opened up agentic checkout. For brands in the recommendation, this is a phenomenal channel — the customer arrives already convinced. For brands that aren't, it's invisible loss: the customer was looking for exactly what you sell, never saw you, and bought something else.
How AI assistants actually decide what to recommend
There's a tendency to treat this as magic. It isn't. When an AI is asked "the best electrolyte drink with no artificial sweeteners," it draws on its training corpus and, increasingly, real-time search, looking for products described as being in that category, with those attributes, in clear, parseable language. The brands that get recommended share a few traits, and content-answer fit drives roughly 55% of citation decisions, favoring self-contained answer passages:
- The product page describes the product in plain, direct language. "An electrolyte drink mix with no artificial sweeteners, sweetened with stevia, in five flavors" is gold. "Feel the difference, hydrate like never before" is invisible to an AI looking for something to recommend.
- The brand has third-party validation the AI can find and cite — mentions in publications it considers authoritative, reviews on platforms it trusts, inclusion in "best of" roundups. Reddit alone appears in 68% of AI Overviews, and user-generated content is ~22% of citations.
- The product is described with structured data — schema markup, machine-readable specs, a real FAQ in the language people actually use.
- The brand exists in the cultural conversation — Reddit threads, newsletters, podcast transcripts — not just its own marketing channels.
A brand can have an excellent product, a beautiful site, and real revenue, and still be invisible if the AI can't find a clean, structured, validated reason to recommend it.
What this means for product-page architecture
For a decade the product page was designed for human eyes — hero image, brand voice, social proof, CTA. That still matters; humans still buy. But the page now has a second job: be legible to AI. And the choices that make a page beautiful to humans often make it illegible to machines. Adobe's benchmark found retail product pages are only ~66% machine-readable to LLMs — roughly a third of what a shopper needs to buy is invisible to the engine doing the recommending. The usual culprits: benefits baked into a graphic the AI can't read while the text is poetic brand language; vague positioning ("Hydration, reimagined") as the primary description; and specs buried in an accordion the AI may never expand. The fix isn't to make the page ugly — it's to make sure the primary text that loads first contains a clear, direct, parseable description of what the product is, who it's for, what's in it, and what makes it different, living alongside the brand language. (It's the same discipline that determines whether cold paid traffic converts: say what the product is, in the first viewport, in language a stranger can parse in five seconds.)
What this means beyond the product page
The product page is necessary but not sufficient — the recommendation also draws on what the broader internet says about the brand in places the AI trusts. This work is less direct and more durable. Press coverage in authoritative publications is the most powerful single lever and the hardest to engineer; a mention in a credible outlet carries weight in the recommendation logic far beyond its direct traffic. Reviews and discussion in forums and communities — Reddit threads about the category, newsletters, podcast appearances — feed the AI's sense of whether the brand is real and what category it belongs to. And comparison articles and "best of" roundups, long a part of SEO, are now part of AI search: a brand consistently included in third-party "best electrolyte drinks of 2026" lists is far more likely to be surfaced. None of this is a quick fix — it's a multi-quarter discipline — but brands that started two years ago are reaping it now.
What this does not change
AI search won't replace everything. Google isn't going away, neither is direct traffic or paid social — people will keep finding products through every existing channel for a long time. What's changing is the share of discovery flowing through AI and its trajectory: small today, growing fast. By the time it's the dominant channel, the brands that already optimized for it will have a structural advantage and the late movers will spend years catching up to a baseline that's no longer an edge — the same dynamic that played out with mobile-first design in 2014 and SEO in the late 2000s. The choice in front of CPG founders now is whether to be in the first cohort or the second.
Where to start
The most useful first step is to ask — literally. Open ChatGPT, Claude, or Perplexity and type the query a customer would use to find a product like yours: "the best non-alcoholic aperitif," "a good replacement for sugary sports drinks," "the best clean protein powder." See whether you're mentioned, and what's said.
- If you're consistently absent, the work is structural: product-page legibility, structured data, third-party presence.
- If you're mentioned with thin or wrong detail, improve the signals — rewrite the product-page text, build out the FAQ, get into more credible third-party content.
- If you reliably show up in the top recommendation, you're ahead of the category — keep building.
- Make sure you're crawlable: confirm GPTBot, PerplexityBot, ClaudeBot, and Google-Extended aren't blocked in robots.txt — if they are, you're invisible regardless of content.
- Keep key pages fresh — AI engines favor recently updated content for buying queries.
The brands that win the next decade of CPG are the ones that get found when the customer asks — and the customer is increasingly asking an AI. Make yourself easy to recommend, or keep paying for acquisition to replace discovery you used to get organically. If you want help getting your store recommendable, that's part of what we do.
Frequently asked questions
What is answer engine optimization (AEO)? Structuring your site and content so AI engines (ChatGPT, Perplexity, Google AI Overviews) can retrieve, trust, and cite it when answering a buyer's question — the discovery layer replacing the blue-links list.
Is AI search traffic actually worth anything? Yes, it's high-intent — AI traffic to US retailers rose 393% YoY and converted 31% better than other sources over the 2025 holidays.
Why isn't my brand recommended by AI? Usually because your product information isn't machine-readable — retail PDPs score only ~66% readable to LLMs — and because you lack third-party corroboration the models trust.
How do AI engines choose which brands to mention? Semantic relevance, an authority/E-E-A-T gate, and passage re-ranking, with content-answer fit driving ~55% of citation decisions. Clear text, schema, reviews, and credible mentions all help.
Do reviews and Reddit really matter for AI discovery? Yes — Reddit appears in 68% of AI Overviews and UGC is ~22% of citations. Models want to see your brand corroborated, not just self-described.
