Back to blog

SEO CRO

AEO for Ecommerce: How to Get Recommended by ChatGPT and Gemini

April 25, 2026 · 10 min read

Ask ChatGPT to recommend a probiotic for a dog with a sensitive stomach and it answers with two or three named brands, a sentence of reasoning for each, and perhaps a citation. There is no page two. There is no position seven. You are either in the answer or you do not exist for that shopper.

A growing share of product discovery now happens inside these answers rather than on a search results page. In the analytics we review for DTC brands, referrals from chatgpt.com, perplexity.ai and Gemini are still a small slice of total sessions — but they convert unusually well, because the assistant has already done the qualifying. The visitor arrives having been told you are the answer.

Answer Engine Optimization is the discipline of earning those recommendations. It has also, predictably, become a magnet for snake oil. This post covers what actually influences whether an LLM names your brand, what transfers from SEO, the tactics worth doing, and the promises you should ignore.

---

How answer engines decide who to recommend

When an assistant answers a commercial query, it is not consulting a ranked index the way classic Google does. Most production answer engines do something closer to this: rewrite the query, retrieve a set of candidate documents from a search index or a live crawl, then synthesise an answer, deciding which sources to trust and which brands to name as it writes. Your job is to be easy to retrieve, easy to verify, and easy to quote.

Five questions an answer engine effectively asks before naming a brand. Fail retrieval and the rest never gets evaluated.

Entity clarity is the part most brands skip. If your homepage says one thing, your Trustpilot profile another, and a three-year-old press mention a third, a model triangulating across sources finds noise where it needs agreement. Consistent naming, a thorough Organization schema with sameAs links, and a one-paragraph description of what you sell repeated everywhere you control. It reads as pedantry, and it works.

Corroboration matters more than anything you publish yourself.

Models are trained, and increasingly instructed at retrieval time, to weight independent sources: review platforms, press coverage, Reddit threads, niche forums. A brand that exists only on its own domain looks, to a system built on cross-checking, like a brand making claims about itself. The brands we see named repeatedly in answers are the ones being discussed in places they do not control.

---

AEO vs SEO: what transfers, what changes

The good news for anyone who has invested properly in search: most of the foundation carries over. Answer engines lean on conventional search indexes for retrieval, so crawlability, speed, clean information architecture and structured data still gate whether you are a candidate at all. We covered how this plays out inside Google's own products in our piece on AI Overviews and ecommerce SEO.

The technical and authority foundations transfer; the competitive logic does not. Answers reward the source, not the survivor of a ranking battle.

The mindset shift is the important part. Classic SEO rewarded covering a topic competently enough to rank somewhere on page one, and ten brands could share that prize. An answer has room for two or three names. Being the source other pages cite — the brand with the original number, the definitive comparison, the documented test — is what gets you onto that short list. Being the tenth competent listicle gets you nothing.

---

The tactics that actually compound

Everything below is work you would defend even if answer engines vanished tomorrow. That is deliberate. It is also the test we apply before recommending any AEO work to a client: if a tactic only makes sense as a trick played on a model, it will not survive the next model version.

Publish original data and benchmarks

LLMs quote sources that give them something concrete to quote. A supplement brand we worked with published a plainly formatted page of third-party lab results across its range: heavy metals, potency, the unfashionable details. Within a few months it was being cited by name in Perplexity answers about supplement quality. Survey your customers, benchmark your category, test your products against competitors and publish the table. If you are the origin of a number, every answer that uses the number tends to carry your name. This is the same logic that underpins a sound DTC content strategy; it simply pays out in citations as well as rankings now.

Maintain comparison pages that mirror real queries

"X vs Y", "best fresh dog food for puppies", "is this brand worth it": answer-engine queries read like questions put to a knowledgeable friend, and comparison pages are the documents that answer them. The catch is they must be honest. A comparison that declares your product the winner on every row reads as marketing, gets corroborated by nothing, and is ignored. Concede the criteria where the competitor genuinely wins. Models cross-check, and the page that matches what Reddit already says is the page that gets cited.

An honest concession is an AEO tactic.

Keep product data immaculate and crawlable

Assistants increasingly pull live product information — price, availability, variants, shipping — from feeds and structured data rather than prose. That means complete Product schema with offers, ratings and availability; a clean Merchant Center feed, since Gemini's shopping answers draw on the same plumbing we covered in our Google Shopping feed guide; and product pages that render server-side rather than assembling themselves in JavaScript. Then check your robots.txt. In the audits we run, a surprising number of stores block GPTBot, OAI-SearchBot or PerplexityBot without knowing it, because a bot-management default or a copied file did it for them.

Earn corroboration in places you do not control

Reviews on third-party platforms, press that says something specific rather than reprinting your release, founder participation in the niche communities where your category is actually discussed. None of this can be faked at scale, which is precisely why it carries weight. One practical note from experience: review volume and recency corroborate better than perfection. A 4.6 average across two thousand recent reviews is more useful to a cross-checking model than a suspicious 5.0 across forty.

---

Measuring AI referrals without fooling yourself

You cannot manage what you refuse to measure, but with AEO you also cannot fully measure it, so the goal is honest partial visibility. Start by separating answer-engine traffic from the rest of your referral noise.

Referrer domains shift as the engines change their products. Review the list quarterly rather than treating it as fixed.

In GA4, the practical setup looks like this:

• Create a custom channel group with a session-source rule matching chatgpt|perplexity|gemini|copilot|claude, so AI referrals stop hiding inside "Referral".

• Build one exploration comparing conversion rate and average order value for that channel against the site average. This is the report that justifies, or kills, further investment.

• Watch branded search volume alongside it. A recommendation that is read but not clicked often resurfaces as a brand search days later.

• Re-run a fixed set of buying prompts monthly across ChatGPT, Gemini and Perplexity, and log which brands get named. Crude, but it is the closest thing to rank tracking that exists here.

Be honest about the gaps. A large share of answer-engine influence never produces a referral at all: the shopper reads the recommendation, then searches your brand or types your URL, and lands as organic or direct. Some surfaces strip referrer data entirely, and in-app browsers mangle what is left. Treat the segmented number as a floor, never as the measurement.

---

Honest expectations — and the snake oil to avoid

Calibration first. For most DTC brands we look at, answer-engine referrals are low single digits as a percentage of sessions. They are growing, and they convert well, but nobody should reallocate half a marketing budget on the strength of them. The honest pitch for AEO is narrower: high-intent, compounding, and nearly free if you were doing content and technical SEO properly anyway.

"If someone guarantees you placement inside ChatGPT or Gemini, they are describing a service that does not exist."

Model outputs are non-deterministic. The same prompt, asked twice, can name different brands. Answers shift with model versions, user context, phrasing and memory. Any agency selling guaranteed inclusion, secret prompt-seeding, or a proprietary method for "ranking in ChatGPT" is selling weather control. The same scepticism applies to tools promising precise AI visibility scores: useful as a rough trend line, fiction as a precise measurement.

The tell is always the same. Durable AEO work makes your brand easier to verify; snake oil promises to make a model behave. One of those is in your control.

---

The citable brand wins twice

Entity clarity, original data, honest comparisons, clean feeds, third-party proof — every item on that list also makes you easier to buy from for humans. That is the reassuring thing about AEO done properly: there is no fork in the road where you choose between answer engines and customers. The work is the same work, finally rewarded twice.

It is also groundwork for what comes next. Recommendations are step one; assistants that complete the purchase on the shopper's behalf are already arriving, and the brands with unambiguous product data will be the ones agents can actually transact with. We have written about what that readiness looks like for Shopify stores, and the overlap with this post is not a coincidence.

Start with the unglamorous checks: can the bots reach you, does the web agree on who you are, and is there anything on your domain worth quoting. In our experience those three questions expose most of the gap between brands that get named and brands that do not. They are also exactly the kind of thing a focused audit surfaces in days, not months.