AI is changing how shoppers discover products, but traditional retail measurement still gives full credit for the purchase to the last click before checkout. That final click might come from Google search, a paid ad, or a direct visit—even if AI influenced the shopper’s decision earlier. As a result, retailers might be underestimating AI’s role in shopping decisions.
We’ve seen the same measurement gap beyond retail. At Stripe, we look beyond direct referrals from AI tools and track other signals that AI might have influenced the user journey, such as activity tied to AI partnerships, AI-assisted product usage, or other verified indicators linking a sign-up to AI tool usage. In H1 2026, approximately 20% of Stripe’s self-serve sign-ups showed at least one of these AI signals. That was approximately 12 times the share attributed to AI tools as the last-click referrer, suggesting that last-click reporting captures only a small fraction of AI’s measurable influence.
As product discovery shifts into AI, retailers need product data that models can understand, measurement that captures influence beyond clicks, and purchase paths that convert wherever shoppers choose to buy. Stripe catalog feeds are one way to share structured product, pricing, and inventory data with supported AI commerce channels, but feeds are only one part of a broader answer engine optimization (AEO) strategy. Retailers also need to understand how their products appear in AI responses, make sure the information is accurate, create a smooth path to purchase, and measure the business impact.
This guide lays out how to measure AI’s effect on product discovery, answer accuracy, purchase flow, and business results.
AEO builds on SEO, but the shopper interaction is different
Answer engine optimization, or AEO, builds on the same basics as search engine optimization: accurate information, crawlable pages, clear language, and useful content.
The difference is that answer engines often summarize, compare, and recommend products before a shopper ever clicks through. That means retailers need to measure more than traffic.
A measurement framework for AEO
AEO measurement should answer four questions:
- Can AI find your products?
- Does it describe them accurately?
- Can shoppers move easily from an AI response into a buying channel?
- Do those interactions lead to sales?
1. Measure discoverability
Discoverability tells you whether your brand and products show up when shoppers use AI to research, compare, or look for a specific item. Start with two baseline metrics:
- Visibility or “mentions”: How often your brand or product appears in a defined set of shopper prompts
- Citation share: How often the response cites or links to your pages, listings, or other sources you control
To measure either, first create a prompt set. A prompt set is a representative group of questions a shopper might realistically ask an AI tool while researching or deciding what to buy. You can test prompts manually by entering them into the AI tools your customers use and recording the results.
To be effective, your prompt set should reflect how shoppers actually describe what they want and weigh their options, not product taxonomy or merchandising terms. If you only track the terms your team uses internally, you might miss the questions shoppers are actually asking and fail to see where your products show up. Instead of a category label like “storage furniture,” think in terms of the search a shopper might use, like “small entryway bench with shoe storage.”
The prompt set should include a mix of category questions, comparisons, problem-led queries, use cases, budgets, sizes, locations, and seasonal needs. For example:
- “Best carry-on luggage for a weeklong business trip”
- “Waterproof hiking boots for wide feet under $200”
- “Where can I buy a modular couch that ships quickly to Chicago?”
The more specific these prompts are, the more useful they can be, especially for situations where you know your products should be a strong fit. For example, in the example above, a furniture retailer might know that many customers are shopping for a modular couch before moving into an apartment. It could track prompts like “what’s the best couch or sofa for someone moving into an apartment?” or “I’m moving into my first apartment in a few months. What couches should I consider?” This helps show whether your products are being recommended for a real customer need—moving into an apartment—not just a broad product keyword like “modular furniture.”
In our own AEO work at Stripe, we don’t treat the prompt set as static. We start with a representative set of questions, then refine it over time using entry point data, customer research, crawler behavior, and user testing. SEO data is also a valuable starting point: traditional keyword research shows what shoppers search for most often, while Google Search Console clicks and impressions highlight the queries where your site already performs well. Retailers can turn these insights into tracked prompts and continue updating based on site search, paid search queries, reviews, customer service questions, and real shopper language.
Because most answer engines still provide limited first-party prompt data, AEO visibility tools like Profound, Scrunch, Peec AI, and Semrush can help you build and test a prompt set that represents the range of ways shoppers actually search for your products, compare results across models, and track changes over time.
Once you have a prompt set, run it at least once a week to see how results change over time. Check more often, potentially daily, if you’re actively publishing or updating content, moving quickly on an AEO strategy, or when an AI provider releases a new model.
Models might pull from product pages, publisher coverage, creator videos, reviews, forums, marketplaces, and local business profiles. Think in terms of source consensus. AI answers are more likely to be consistent when the same accurate product information shows up across your own sites and pages in addition to the third-party sources models seem to trust. That source mix varies by model and category.
At Stripe, we review source patterns by model rather than assume the same publishers, communities, or platforms will improve visibility everywhere. Retailers should do the same. In broad retail categories, models might lean on marketplaces, review sites, YouTube, or Reddit; in niche categories, a smaller trade publication or trusted creator can carry more weight than a major news outlet.
What to track
- Visibility rate for priority prompts, by model and market
- Citation rate for first-party product, category, editorial, and help pages
- Share of voice versus the key retailers and marketplaces that shoppers compare you with
- The domains and creators cited most often in your category
- LLM crawler hits, including which pages answer engine crawlers fetch, and how often (This is an early signal that a page might be informing answers, not proof of visibility, traffic, or conversion.)
- Trends by model, market, product category, and shopper need
2. Measure your answer accuracy
Visibility only helps if AI describes your products, policies, and business accurately enough for a shopper to make a decision.
Your brand can appear in an AI answer without the model linking to or naming the source it used. Even a cited answer can get key details wrong, such as an outdated price, the wrong variant (like recommending the queen size when the shopper asked for a twin), missing availability, an incorrect delivery estimate, or an outdated return policy. Review a recurring sample of priority prompts and compare the answers against your current product and policy data.
We’ve seen that visibility alone doesn’t guarantee accurate representation. AI models can rely on older training data or outdated third-party sources when forming a recommendation, even when updated information is available. Once a model has made that decision, it might not revise its recommendation, even when it encounters newer information. At Stripe, for example, AI systems sometimes surfaced current documentation while recommending legacy integration paths. For retailers, the equivalent is appearing in an AI response with the wrong price, outdated availability, incomplete product details, or a recommendation for the wrong product configuration.
Answer accuracy depends on product data quality. Titles, descriptions, variants, prices, stock status, images, delivery estimates, return policies, and reviews all affect whether a product appears and how it’s described. Product feeds deserve particular attention because they can give AI commerce channels a definitive view of your catalog. For US businesses selling physical goods, Stripe’s catalog feed can syndicate structured product, pricing, and inventory data to supported AI commerce channels. Keep product and policy data complete, machine-readable, and current across the pages, feeds, and marketplace listings models might rely on.
3. Measure the handoff
The handoff is the point where a shopper moves from an AI response into a buying channel: a product page, marketplace listing, local profile, checkout, or store. This is where AI visibility turns into a retail interaction.
Retailers need to measure whether AI mentions a product, in addition to whether shoppers can actually act on the information they’re given. As discovery shifts into AI, the purchase path has to work across every channel where a shopper might land.
Test this from the shopper’s point of view. Can they find the right product, confirm price and availability, choose the right variant, and complete the next step without running into conflicting information or a dead end? That experience should be reviewed across every channel you sell in.
Referral data will still be incomplete. Some answer engines don’t consistently pass referrer or UTM data, and shoppers influenced by AI might later arrive through direct traffic, branded search, a marketplace, or a store visit. As a result, the handoff is best measured in two ways: what shoppers see after clicking through, and the referral and conversion signals you can observe.
4. Measure the business impact
Visibility, representation, and handoff metrics tell you whether the system is working. Business impact shows whether AEO is driving measurable results.
For visits you can attribute directly, compare AI-driven traffic with other channels on:
- Conversion rate and revenue per session
- Average order value
- New versus returning customer mix
- Product category and inventory mix
- Returns, cancellations, and customer service contacts
- Repeat purchase and customer lifetime value, for categories with recurring purchase behavior
It can also be helpful to evaluate these metrics by the LLM model. Performance often varies by model, which can help surface gaps and clarify where to focus your optimization efforts. For example, if one model drives a disproportionate share of traffic over the others, it might be a sign to prioritize improving how your brand appears in its responses.
If you don’t have a dedicated data science team, start with a practical three-part approach.
As your measurement tactics get more advanced, you can add methods that estimate influence beyond the click. Multitouch attribution spreads credit across several touchpoints. Incrementality testing asks whether AI drove demand that would not have happened otherwise. Media mix modeling estimates each channel’s contribution using aggregate performance data. At Stripe, we took that approach in 2026, adding AEO as a separate input to our media mix model after seeing that click-based attribution missed part of AI-driven demand. Retailers don’t need to start there. Begin with measurable referrals and conversion, add customer-reported and correlated demand signals, and then use incrementality testing or media mix modeling as your data capabilities grow.
Performance will vary by AI platform and market. At Stripe, AEO-referred traffic often converted to sign-ups at higher rates than SEO, but earlier analyses showed weaker seven-day first-charge and 30-day activation rates than organic search. To improve how answer engines surfaced and described Stripe, we expanded and refreshed content, made key documentation more accessible to AI crawlers, and corrected outdated third-party information. As AEO referral channels matured, downstream conversion became more comparable to, and in some cases exceeded, SEO. Retailers should compare traffic quality alongside volume and break results down by answer engine, country, category, landing page, and shopper intent before deciding where to invest.
The AEO space is changing quickly, and measurement will keep improving. Retailers don’t need perfect attribution to start. A practical first step is to see how answer engines describe your products, fix any inaccurate or incomplete information, and then track whether those changes show up in traffic, conversion, and sales.
Additional resources
Learn how to get ahead in agentic commerce.