How to Make AI Recommendations for Your Shop Measurable — Without Clean Click Tracking

Why Click Tracking Breaks Down for AI Recommendations

Imagine someone asks ChatGPT"Where do I get sustainable running shoes under 120 euros?" and your brand gets named. The user then types your shop name straight into the browser, or clicks a link with no UTM attached. In your analytics that shows up as "Direct" or "Organic." The revenue is there, but the source is invisible. This is exactly where classic e-commerce tracking, built for years around clean referrers and campaign parameters, falls apart.

The problem is structural, not just a tooling gap. Many AI answers lead to no click at all: the user reads that your shop carries a few relevant models, weighs the options mentally, and decides later. Days can pass between the recommendation and the purchase, often on a different device entirely. No cookie, no referrer, no session chain survives that gap. For an online shop, the channel growing fastest right now is also the one you can see the least of.

The consequence is dangerous. If you don't measure AI visibilityat all, it looks like nothing in the report — and you end up cutting budget from exactly the channel bringing in new customers. Before you go chasing a perfect tracking setup that doesn't technically exist, you need replacement signals: indirect measurements that, taken together, add up to a reliable picture.

Step 1: Measure the Mention Itself, Not Just the Click

The first metric worth measuring isn't traffic — it's the mention. Before anyone can click, the AI has to name your shop at all, and that's something you can test systematically. Build a list of 30 to 50 purchase-intent prompts your real customers would actually type: "best cable management for a desk," "buy vegan protein powder without sweetener," "gift for coffee nerds under 40 euros." Run these prompts weekly through ChatGPT, Perplexity, Google AI Overviews and Gemini.

Track three things per prompt: are you named, at what position, and in what context — as the recommendation, as one example among several, or only as a passing footnote? From this you build a simple share-of-voice metric: the percentage of your purchase-intent prompts where you show up, measured against your three biggest competitors. That number is stable and trackable, even when not a single click gets logged.

This matters for shops especially: separate generic prompts from category-specific ones. "Where do I buy sneakers" is nearly impossible to win; "sustainable barefoot shoes for wide feet" isn't. Your niche is where you can realistically show up in AI answers, so measure the niche first and work outward from there.

Step 2: Server Logs Beat JavaScript Tracking

AI systems leave a trace before they ever recommend you: their crawlers fetch your product pages. GPTBot, OAI-SearchBot, PerplexityBot, and Google-Extended all access your shop directly. These hits land in your server logs, not in Google Analytics, because bots don't execute JavaScript. If you want to know whether AI systems even know your catalog exists, the server log is your most honest data source.

Filter your logs for these user agents and see which pages get fetched, and how often. A typical e-commerce pattern: the homepage and bestsellers get crawled constantly, while deep category and filter pages barely register. That explains why the AI knows your flagship products but never surfaces your long-tail niches — and it turns into concrete work: fix your internal linkingadd a clean sitemap, and cut the parameter clutter in your URLs.

It's also worth a second look at Perplexity and Bing-based systems, which sometimes fetch pages live. If a Perplexity crawl shows up in your logs shortly after a user prompt, and a purchase with no referrer follows soon after, that's a strong indirect signal — with AI traffic, you rarely get closer to real attribution proof than that.

Step 3: Ask Buyers Directly With a Post-Purchase Survey

The most underrated measurement method in e-commerce is also the simplest: one question after the purchase. Add "How did you find us?" to the thank-you page or the first transactional email, and include "ChatGPT / AI assistant" explicitly as an answer option. Many shops discover this way, for the first time, that a meaningful share of new customers came through an AI recommendation — a channel that never once showed up in their analytics.

This self-reported attribution isn't perfect, but it closes exactly the gap that cookies and referrers leave open. What matters is linking the answers to actual order value. Then you see not just that AI customers exist, but whether they spend more or less, what they buy, and whether they come back. For an internal budget conversation, that's often more persuasive than any dashboard.

Keep the question short and optional so it doesn't get in the way of your conversion. A five-option dropdown is enough. Review it monthly and watch the trend — a rising AI share is your clearest proof that work on AI visibility is paying off.

Step 4: Read Cohort and Direct-Traffic Patterns

Even without clean tracking, AI traffic leaves patterns in your aggregated data. Watch for a growing share of "Direct" sessions with unusually long, focused dwell times on the exact product pages that match your AI recommendations. If your barefoot-shoe page suddenly gets more direct visits right as you start appearing there in Perplexity answers, that's not coincidence — it's a correlation worth documenting.

Build cohorts by landing page. New direct visitors who arrive on a deep product or guide page instead of the homepage often behave like AI-recommended users: they show up pre-informed, bounce less, and convert faster. Compare this cohort over time against your prompt-test results from step 1. If both curves rise together, you've largely solved your attribution puzzle.

Pair this with brand-search data from Google Search Console. A rise in searches for your shop name plus a product category is a classic side effect of AI recommendations — the AI names you, and the user googles you afterward. This "assisted brand lift" is measurable, even when the original AI contact never was.

Step 5: Optimize What the AI Actually Pulls From Your Shop

Measuring is half the job; the other half is giving the AI something worth citing. AI systems recommend shops whose product information is unambiguous, structured, and fact-rich. In practice that means product descriptions built on concrete facts instead of marketing prose — real dimensions, materials, use cases. "Waterproof to 10 meters, 42 grams, fits wrists from 14 to 20 cm" gets cited. "The perfect companion for your adventure" does not.

Implement structured data properly: Product schema with price, availability, reviews, and GTIN. This helps classic SEO too, but more importantly it hands AI crawlers machine-readable facts they can drop straight into an answer. Add real comparison and guide content — "Model A vs. B for beginners" — because pages like these are exactly what assistants cite when someone is looking for purchase advice.

Measure the effect of this work by running step 1 again. If you clean up a category's facts and then show up in more prompts three weeks later, you have a clean before-and-after comparison. That's what turns GEO for your shop into an iterative, verifiable process instead of a guess.

Step 6: Build an AI-Visibility Dashboard That Needs No Pixel

Bring the individual signals together into one view. A practical AI dashboard for an online shop has four rows: share of voice from your weekly prompt tests; crawler hits from the AI bots in your server logs; the self-reported AI share from your post-purchase survey; and brand search plus direct-traffic cohorts. None of these numbers proves anything on its own, but together they point in a reliable direction.

Update the dashboard monthly and tie it back to revenue. The sentence your team needs isn't "we had 800 AI clicks" — it's something closer to "our AI share of voice climbed from 18 to 31 percent, and the self-reported AI share of new customers went from 5 to 9 percent, at a stable order value." That's the language that makes AI visibility worth budgeting for.

Keep the method honest: mark clearly what's measured and what's estimated. Precisely because clean click tracking doesn't exist here, you build trust by naming the uncertainty instead of faking a precision the data can't back up.

Step 7: Connect AI Visibility to Cart Value

Visibility alone doesn't help much if you don't know what it does to the cart. So take a second look at the order value of purchases that land with no classic click source attached. If you built direct-traffic cohorts in step 4, attach the average cart size, the return rate, and the number of line items per order to each one. That way you can see whether AI-recommended buyers reach for higher-value items or just the entry-level ones.

In practice, a clear pattern often shows up: someone arriving via an AI recommendation already knows your product by name and buys more purposefully. In cases we've seen, the direct-traffic cohort's average cart runs noticeably above the site average, because the AI recommended a complete set rather than individual parts. Segmenting order value by cohort makes relationships like this visible — entirely without a pixel, using nothing but your own shop and order data.

Limits: What This Method Cannot Prove

Be honest with yourself about the gaps. What you're measuring here are approximations, not clean attribution chains. A direct visit could come from an AI recommendation, but just as easily from a bookmarked page, a newsletter, or a conversation offline. Treat your numbers as a trend, not proof. A rise in brand-related direct visits alongside more mentions in AI answers is a strong signal — but it isn't a verdict.

Second, your optimizations are chasing a moving target. AI models get retrained, answers fluctuate, and what gets recommended today may look completely different in two months. Set fixed measurement points — the same test questions to the assistants every four weeks, for example. That's how you separate real improvement from random noise, instead of rebuilding your entire catalog around a one-off outlier.

Questions Shop Owners Actually Ask

Is this worth it for a small shop? Yes, especially then. You don't need an expensive analytics setup — your server logs, your order data, and a handful of honest test questions each month are enough. The time cost looks more like a couple of hours a week than a dedicated tool budget. With a smaller catalog the leverage is even bigger, since individual recommended products register more strongly in percentage terms.

How often should I measure? Set a fixed rhythm: check logs and cohorts weekly, repeat the AI test questions monthly, reconcile the dashboard against revenue quarterly. And if the AI never names your product at all? That's your most important finding — go back to step 5 and check whether your product copy, spec sheets, and FAQ actually answer the questions buyers are asking the assistants.