← Back to Blog

98% Browse, 2% Buy: What AI Agents Are Doing on Your WooCommerce Site

Quick Answer: Only about 2% of AI agent activity reaches checkout — 79% is concentrated on product and search pages (HUMAN Security / MetaRouter, 2025). AI bots crawl retail sites 198 times more per visit than Google, bot traffic rose 5.4x in 2025, and 80% of agents don’t declare themselves. The 98% browsing isn’t wasted traffic — agents are building recommendation databases, comparing prices, and evaluating product data. Stores with structured data are being indexed for future purchases. The browsing agents of today are the buying agents of tomorrow.

What percentage of AI agent traffic actually makes purchases?

Only about 2% of AI agent activity reaches checkout — 79% is concentrated on product and search pages, meaning 98% of what agents do on your site is browse and compare, not buy (HUMAN Security / MetaRouter, 2025). That ratio is the most underreported metric in ecommerce right now.

The number sounds alarming until you understand what the 98% is doing. AI agents aren’t bouncing — they’re cataloguing. Each visit is a data-collection run: parsing product specifications, recording prices, noting availability, building the internal index that future purchase recommendations will draw from.

Think of it this way. When Google crawls your site, it’s indexing for search results. When an AI agent crawls your site, it’s indexing for purchase recommendations. 79% of agent activity concentrated on product and search pages (HUMAN Security, 2025) tells you exactly what they’re after — the commerce layer, not your about page.

79% of AI agent activity is concentrated on product and search pages, while only 2% reaches checkout — 98% of what agents do is browse and compare, not buy (HUMAN Security / MetaRouter, 2025).

Here’s the thing — the 2% that does buy converts at a significantly higher rate than human traffic. The other 98% is building the pipeline that feeds that 2%.

What are AI agents doing when they browse product pages?

Agents parse structured data, compare prices across competitors, evaluate product attributes, and build recommendation lists — they read the DOM and product feeds, not your marketing copy or design. Your hero image and persuasive product description exist for humans. Agents read the JSON-LD, the schema markup, and the raw data layer.

From January to August 2025, approximately 87% of all pages browsed by agents were related to products (HUMAN Security, 2025). They’re not reading your blog. They’re not browsing categories for inspiration. They’re going directly to the product data and extracting what they need: price, SKU, availability, specifications, reviews.

The practical implication: a WooCommerce store’s AI-readiness isn’t measured by how good the site looks or how persuasive the copy reads. It’s measured by how clean, structured, and machine-readable the product data is. If your product feed has gaps — missing specifications, inconsistent pricing, no structured reviews — agents build an incomplete picture of your catalogue. When a shopper asks that agent for a recommendation, your products aren’t in the running.

Related: Crawler, Search Bot, or Shopping Agent: The Three Species of AI Traffic Visiting Your WooCommerce Store

How much more do AI bots crawl compared to Google?

AI bots crawl retail sites 198 times more per visit than Google — they traverse product pages at a speed no human matches, reading dozens of product pages in seconds (MetaRouter, 2025). Your Googlebot load is a rounding error compared to what AI agents are pulling.

198x more crawling per visit than Google (HUMAN Security / MetaRouter, 2025) means your server infrastructure is already handling AI agent traffic at scale, whether your analytics reports it or not. A single agent session might hit 47 product pages in 3 seconds — no scroll events, no mouse movements, no pause to read. Pure data extraction.

For WooCommerce store owners, this has two immediate implications. First, server resources: if your hosting is sized for human traffic patterns plus Googlebot, it’s undersized for the actual load hitting your product pages. Second, caching strategy: AI agents make the same structured-data requests at high frequency, so a well-cached product data layer handles the load efficiently, while uncached dynamic pages spike your server costs.

Visitor TypePages per SessionSession PatternWhat They Read
Human shopper3–8Browse, compare, decideImages, copy, reviews
GooglebotVariesCrawl scheduleHTML, structured data
AI shopping agent47+ in secondsRapid product traversalDOM, JSON-LD, feeds

Why do 80% of AI agents not declare themselves when visiting websites?

Because there is no incentive to identify during discovery — protocols for identification exist only at the payment layer, and an agent works best for its user when it looks like a regular visitor during browsing (MetaRouter / PPC research, 2025). The result: 80% of AI agents do not declare themselves properly when visiting websites.

This isn’t nefarious. It’s structural. During the browsing phase, an agent that identifies itself as a bot risks being blocked, rate-limited, or served a different version of the page. Since the agent’s job is to gather accurate product data for its user, presenting as a regular browser session gets better data. The identification layer kicks in later, at checkout, where payment protocols like ACP require the agent to declare itself.

80% non-declaration (PPC research / MetaRouter, 2025) means your analytics is classifying most agent visits as regular user sessions. GA4 sees a “user” who viewed 47 product pages in 3 seconds with no scroll events and no mouse movement — and counts it as one very fast human shopper. The traffic looks like Direct. The behaviour looks like a bot. Your analytics can’t tell the difference.

Related: The Complete AI Bot and Agent User-Agent Reference for WooCommerce

How can you detect AI agent traffic on a WooCommerce store?

Look for request patterns like rapid page traversal, header signatures including specific user-agent strings, navigation without scrolling or mouse movement, and session shapes that go product-to-product with no homepage or category browsing. Client-side analytics misses most of this — the detection layer needs to sit server-side.

The signals are consistent. Agent sessions show three distinctive fingerprints: speed (dozens of pages in seconds), path (product-to-product, skipping navigation), and absence (no scroll, no mouse, no viewport resize). A human who views 47 product pages generates thousands of scroll and interaction events along the way. An agent generates zero.

Server-side tracking is the detection layer that makes this visible. Because the event data hits your server before the browser processes it, you can inspect request headers, timing, and navigation patterns at a granularity that client-side JavaScript never captures. Transmute Engine captures these server-side signals — request signatures, session shapes, and navigation patterns — that GA4 lumps into regular traffic, giving store owners visibility into the 98% that’s browsing.

LIFETIME LOCK on Harvest Annual Plan

Push your WordPress event data server-side seamlessly into BigQuery so AI can analyse it.

Let's Do It !

How much did bot traffic grow in 2025?

Bot traffic rose 5.4x in 2025 and 80% of websites are exposed to agent spoofing — AI agents now represent a significant and growing share of ecommerce site traffic (MetaRouter, 2025). The growth isn’t linear. It’s compounding as more AI platforms launch shopping capabilities.

5.4x growth in a single year (industry reports / MetaRouter, 2025) tracks with the expansion of AI shopping assistants across ChatGPT, Copilot, Perplexity, and dozens of vertical-specific agents. Each new agent is another crawler hitting every product page it can find.

Bot traffic rose 5.4x in 2025 and 80% of websites are exposed to agent spoofing — AI agents now represent a growing majority of ecommerce crawl volume (HUMAN Security / MetaRouter, 2025).

The spoofing exposure compounds the problem. 80% of websites exposed to agent spoofing (HUMAN Security, 2025) means most WooCommerce stores can’t distinguish a legitimate shopping agent (which you want indexing your products) from a scraper (which you don’t). Without that distinction, store owners either block everything or let everything through — neither is a strategy.

What percentage of agent-browsed pages are product pages?

From January to August 2025, approximately 87% of all pages browsed by agents were related to products — agents are heavily focused on the commerce layer, not content pages (HUMAN Security, 2025). Your blog might drive human traffic. Your product pages drive agent traffic.

87% product-page concentration (HUMAN Security, 2025) confirms what the 79% product-and-search figure suggests: agents are single-minded about commerce data. They don’t browse your about page, read your shipping policy for fun, or check your Instagram feed. They go straight to the data that answers a shopping question.

For WooCommerce stores, this concentration is both a signal and an opportunity. The signal: your product pages are already being evaluated by AI agents whether you know it or not. The opportunity: stores that optimise those pages for machine readability — structured data, complete specifications, schema markup, clean product feeds — are the ones whose products appear in AI-generated recommendations. The 87% tells you exactly where to invest.

Related: What are AI shopping agents on WooCommerce?

Will the 98% browsing to 2% buying ratio change?

Yes — as payment protocols mature and more agents gain purchasing capability, the ratio is shifting; the browsing agents of today are the buying agents of tomorrow, and stores that structure their data now will be the ones agents buy from (commercetools, 2026). The browsing phase is not an end state. It’s a build phase.

The mechanics are straightforward. Right now, most AI agents can browse, compare, and recommend — but the actual purchase still requires a human to click “buy.” As protocols like ACP standardise agent-to-merchant payment flows, the handoff will shrink. An agent that already has your product data, your pricing, and your availability only needs payment capability to close the loop.

The stores that benefit first won’t be the ones that scramble to optimise when agents start buying. They’ll be the ones whose product data is already in the agent’s index — the ones the agent has been browsing, comparing, and cataloguing for months. The 98% browsing phase is the qualifying round. The 2% buying phase is what comes next. And the index the agent builds during browsing is what determines which stores make the cut.

LIFETIME LOCK on Harvest Annual Plan

Push your WordPress event data server-side seamlessly into BigQuery so AI can analyse it.

Start NOW !

Key Takeaways

  • 98% browse, 2% buy: 79% of AI agent activity hits product and search pages, only 2% reaches checkout — agents are cataloguing, not purchasing.
  • 198x more than Google: AI bots crawl retail sites 198 times more per visit than Googlebot, putting real load on product page infrastructure.
  • 87% product focus: agents are almost exclusively interested in the commerce layer — product data, pricing, availability, specifications.
  • 80% invisible: most AI agents don’t declare themselves, so analytics counts them as human visitors with unusual behaviour patterns.
  • 5.4x growth: bot traffic rose 5.4x in 2025 and the growth is compounding as more AI platforms launch shopping capabilities.
  • Browsing builds the buying index: stores with structured product data are being indexed now for purchase decisions agents will make as payment protocols mature.
What percentage of AI agent traffic actually makes purchases?

Only about 2% of AI agent activity reaches checkout. 79% is concentrated on product and search pages, meaning 98% of what agents do on your site is browse and compare, not buy. The browsing phase is agents building recommendation databases and price comparison indices for future purchase decisions.

What are AI agents doing when they browse product pages?

Agents parse structured data, compare prices across competitors, evaluate product attributes, and build recommendation lists. They read the DOM and product feeds, not your marketing copy or design. From January to August 2025, approximately 87% of all pages browsed by agents were product-related.

How much more do AI bots crawl compared to Google?

AI bots crawl retail sites 198 times more per visit than Google. They traverse product pages at a speed no human matches, reading dozens of product pages in seconds. This intensity means your server load from AI agents may already exceed your Googlebot load without you knowing.

Why do 80% of AI agents not declare themselves when visiting websites?

Because there is no incentive to identify during the discovery phase. Protocols for identification exist only at the payment layer. An agent works best for its user when it looks like a regular visitor during browsing, so most present standard browser user-agent strings instead of identifying as bots.

How can you detect AI agent traffic on a WooCommerce store?

Look for request patterns like rapid page traversal without scrolling or mouse movement, header signatures including specific user-agent strings, and session shapes that go product-to-product with no homepage or category browsing. Server-side tracking captures these signals that client-side analytics miss.

How much did bot traffic grow in 2025?

Bot traffic rose 5.4x in 2025 and 80% of websites are exposed to agent spoofing. AI agents now represent a significant and growing share of ecommerce site traffic, and their activity is concentrated almost entirely on product pages rather than content or navigation pages.

What percentage of agent-browsed pages are product pages?

From January to August 2025, approximately 87% of all pages browsed by agents were related to products. Agents are heavily focused on the commerce layer — product data, pricing, availability, specifications — not blog content or marketing pages.

Will the 98% browsing to 2% buying ratio change?

Yes. As payment protocols mature and more agents gain purchasing capability, the ratio is shifting. The browsing agents of today are the buying agents of tomorrow, and stores that structure their data now will be the ones agents buy from when the purchasing phase scales.

References

  1. MetaRouter / HUMAN Security (2025). Agentic Commerce Trends and Statistics. Source
  2. commercetools (2026). Agentic Commerce Stats: Enterprise Guide. Source
  3. Omniconvert (2026). Agentic Commerce Statistics. Source