Are AI Assistants Recommending You?

The moment that decides whether a customer buys your product may have already happened — inside an AI assistant’s response, before the buyer ever sees your listing. On August 6, 2026, Cloudflare launched a tool that lets you see whether that moment went your way.

The AEO Visibility Dashboard, now in early access, probes the leading AI assistants with real customer queries and measures whether your brand gets cited, mentioned, or passed over. It is the first product to surface recommendation data at scale — not impressions, not clicks, but the actual share of voice you hold when an AI tells a shopper which product to choose.

The problem this solves

For two years, marketplace sellers have operated blind on one critical question: when a buyer asks ChatGPT, Claude, or Google AI Mode “which product is best for my situation,” does the assistant recommend you? You could check manually — ask the question yourself, note the answer, repeat across engines and phrasings — but the answers drift. AI assistants rarely respond the same way twice, and one-off checks tell you nothing about trends, competitors, or the prompts you are losing.

Existing tools filled pieces of the gap. Google Search Console’s Generative AI report shows impressions but not which queries triggered them or where you ranked. Cloudflare’s Agent Readiness scanner checks whether bots can technically read your site. The Ceramic.ai AEO reporting partnership from July tracked crawl-to-referral ratios — how often bots fetch your pages versus how many visitors they send back.

None of them answered the question sellers actually care about: am I being recommended?

The AEO Visibility Dashboard changes that. It probes Anthropic’s Claude and OpenAI’s GPT with category-relevant prompts — product comparisons, recommendations, and advice queries that mimic real shopping discovery — and extracts four metrics from the responses:

• Citation Rate: the share of answers that cite your site as a source
• Mention Rate: how often the assistant names your brand, whether or not it links to you
• Prominence: when you are cited, how early and how much of the answer is attributed to you
• Share of Voice: your slice of citations against competitors in the same category

Why mention rate and citation rate tell different stories

The distinction between Mention Rate and Citation Rate is where the tool gets strategically interesting. Cloudflare frames it directly: assistants naming your brand far more than they cite you means you are on their radar but not yet earning the citation — a specific, targetable gap.

For a marketplace seller, this maps to a concrete scenario. Imagine you sell handmade leather wallets on Etsy. A buyer asks Claude: “What are the best handmade leather wallets under $100?” If Claude says “brands like Your Shop Name and a competitor are popular options” but only links to the competitor’s standalone website, your Mention Rate is high but your Citation Rate is near zero. The assistant knows you exist. It just does not trust your listing enough to send the customer there.

That trust gap — between being known and being cited — is exactly what Generative Engine Optimization addresses. It is the gap between brand awareness and brand authority in the eyes of an AI model, and until now, sellers had no way to measure it.

How the measurement works under the hood

Cloudflare builds a category-level benchmark before scoring any individual site. It queries the assistants with likely prompts — without specifying your brand — and records which sites appear, where they rank, and how prominently they feature. This panel runs once per category and is reused across all accounts in that domain, so results load instantly rather than waiting for live model queries.

To handle the inherent variance in AI responses (the same prompt can return different brands on different runs), Cloudflare uses its AI Gateway to prompt each assistant multiple times across different models. Workers AI then reads each response and scores how citations and mentions appear — not a model grading its own output, but a separate evaluation layer extracting structured signals from the text.

Alongside the recommendation metrics, an AI Operator Activity panel shows real crawl and referral traffic per operator (OpenAI, Google, Anthropic, and others). The pattern Cloudflare flags as worth acting on: an operator that crawls thousands of your pages but refers no one back — consuming your content without sending customers.

The readiness-to-recommendation gap

Here is where marketplace sellers should pay close attention. Cloudflare’s own data reveals a structural shift: fewer than half of all HTML page requests now come from a human. The machines reading your site are no longer just indexing it — they are evaluating whether to recommend it.

But being readable is not the same as being recommended. An anonymized FirstShelf audit of 24 marketplace listings over 90 days found an average quality score of 50.5 out of 100, with entity authority — the structured data signals that tell an AI who you are, what you sell, and why you are trustworthy — scoring just 32.5 out of 100.

That means most listings are technically accessible to AI crawlers but lack the structured evidence that would make an assistant confidently recommend them over a competitor. The gap between a site that passes an agent readiness check and a brand that wins AI recommendations is the same gap between having a door and having a reputation. The AEO Visibility Dashboard now lets you see which side of that gap you are on — and the breakdown of where the work is.

The weakest dimension, entity authority at 32.5, is precisely the signal that determines whether an AI model treats your listing as a specific, credible, recommendable option or a generic entry it can safely skip. Semantic density and platform compliance score higher but still leave substantial room before listings become the kind of evidence-rich pages that earn citations.

What marketplace sellers should do now

Request AEO Visibility early access

The dashboard launched in early access on August 6, 2026. If your store is on Cloudflare, check the Overview tab in your dashboard. If not, the metrics it surfaces — Citation Rate, Mention Rate, Share of Voice — are the new KPIs for AI-era discoverability, and you should begin tracking them however you can.

Close the mention-to-citation gap

If your Mention Rate is high but Citation Rate is low, the assistant knows your brand but does not cite you as a source. The usual cause is thin product pages — listings that mention a product but lack the structured attributes, specifications, and evidence that make an AI confident enough to link. Complete your product schema, fill in every attribute (brand, GTIN, size, material, returns policy), and ensure your listing answers the comparison questions buyers actually ask.

Watch your crawl-to-referral ratio

Use the AI Operator Activity panel to identify operators that read your content but send no visitors. A high crawl volume with zero referrals means your content is being consumed without compensation — the same dynamic Cloudflare flagged in its July partnership with Ceramic.ai. If you run your own Shopify site, consider whether Content Signals and freshness hints can reduce wasteful re-crawls.

Fix agent readiness fundamentals

The Diagnostics tool groups fixes by effort. Start with the quick wins: a valid robots.txt that explicitly allows AI search crawlers (OAI-SearchBot for ChatGPT search, Google-Extended for AI Overviews), an XML sitemap, and serving clean Markdown to agents. These are the technical minimum for being readable by the systems that now decide recommendations. For GEO for marketplace sellers, these fundamentals are non-negotiable.

Treat entity authority as the recommendation lever

FirstShelf’s audit data shows entity authority at 32.5 out of 100 across 24 marketplace listings — the lowest-scoring dimension. This is the structured evidence layer (product identifiers, brand entities, specifications, trust signals) that tells an AI model you are a specific, credible, recommendable option rather than a generic listing. Closing this gap is the most direct path to moving from mentioned to cited to recommended.

How FirstShelf can help

FirstShelf’s Buyer Match Audit evaluates exactly the structured-data layer that determines whether AI assistants cite and recommend your products. Our audit of 24 marketplace listings found entity authority averaging just 32.5 out of 100 — the same signal that separates a brand an assistant mentions from one it confidently recommends. The audit identifies your specific missing attributes, incomplete identifiers, and trust gaps, then prioritizes the fixes most likely to move your Citation Rate and Share of Voice.