AI Agents Are Becoming the New “Shoppers” — What Meta’s Muse Traffic Surge Means for Your SEO, Security, and Revenue
Meta’s Muse and OpenAI agents are now driving the bulk of observed AI-agent browsing activity—mostly hitting product and search pages, not checkout. That changes what you should measure, how you should protect your site, and how to win visibility when agents—not humans—become the first touchpoint.
AI agents are no longer a sci-fi edge case. They’re quietly becoming a new class of “visitor” that can browse your catalog, compare options, and influence what gets recommended—often without behaving like a normal human session.
A recent report cited by Search Engine Journal describes a sharp rise in agentic traffic observed across sites and apps protected by HUMAN Security, with Meta’s Muse and OpenAI-related agents representing the majority of observed requests in that dataset. More importantly, the browsing patterns skew heavily toward product and search pages—and much less toward checkout.
That’s the tell.
Agents are building “shortlists,” not completing purchases (yet). Which means the next competitive advantage isn’t just ranking for keywords. It’s becoming the easiest business for an agent to evaluate confidently and recommend without getting blocked, confused, or sent down a dead end.
In this editorial, I’ll break down what changed, why it matters for SMEs, ecommerce, and publishers, what can go wrong (security and measurement included), and a practical plan to make your site agent-ready—without handing your business over to bots. I’ll also show where AYSA fits as an execution system that monitors, prepares, asks for approval, and ships the changes that help you show up in AI-driven journeys.
Concise summary

- Agent traffic is surging in at least one large security provider’s view of the web, and two companies’ agents dominate the observed requests in that dataset.
- Most agent requests hit product and search pages, not checkout—so “visibility” is shifting earlier in the buying journey.
- Identification and verification are messy: depending on how agents identify themselves (or don’t), your bot protection may misclassify traffic and break experiences for legitimate automation.
- Robots.txt alone isn’t the control surface for many modern agents; security, verification, and analytics instrumentation matter more than ever.
- Winners will build agent-friendly pages (clear offers, shipping/returns, availability, Structured data, canonical hygiene) and measure agent impact separately from human sessions.
Key takeaways for business owners

- Don’t panic-block “unknown” agents sitewide; you can lose visibility where it matters (product discovery) while still protecting sensitive endpoints.
- Assume your analytics can’t reliably separate agents from humans out of the box; you’ll need server-side logs, edge/WAF signals, and deliberate tagging strategies.
- Optimize the pages agents actually visit: category/search pages, product detail pages (PDPs), and core policy pages that answer pre-purchase questions.
- Shift from ‘rankings’ to ‘recommendability’: clarity, trust, and machine-readable specifics beat vague marketing copy.
Table of contents

- What changed: Meta’s Muse surge is a signal, not trivia
- The new reality: agents browse more than they buy (for now)
- Standalone agent browsers are losing to embedded agents
- Identification is breaking: why “it looks like your activity” changes everything
- Security and SEO are now one operational system
- Analytics: why your dashboards will lie to you about “users”
- Practical SEO/AEO/GEO implications: optimize for the agent’s shortlist
- Technical checklist: make key pages accessible and verifiable
- Content checklist: reduce uncertainty, increase machine confidence
- A concrete SME scenario: the agent-first holiday spike that breaks your reporting
- What agencies should rethink (and what to sell instead)
- Where AYSA fits: monitored, approved execution for agent-ready SEO
- What to do next (action list)
- Sources and further reading
What changed: Meta’s Muse surge is a signal, not trivia
The headline point from the HUMAN Security dataset described by Search Engine Journal is concentration: a small number of agent providers appear to drive the bulk of observed agent requests in that month’s reporting window. In their view, Meta’s Muse grew quickly and became the largest single source of agent requests among the agents they tracked, with OpenAI-related agents also contributing meaningfully.
You don’t need to treat any one vendor’s dataset as “the whole internet” to take the lesson seriously. Even with sampling bias (HUMAN’s view is limited to sites and apps they protect), the directional message is clear:
- Agent traffic is accelerating.
- Agent traffic is concentrating in a few ecosystems.
- Agent browsing behavior looks unlike human browsing—and that creates operational side effects.
This is the kind of shift that initially looks like an infrastructure/security problem (“why are there so many requests?”) but quickly becomes a revenue and brand problem (“why are we not being recommended?”).
Historically, “search” was the discovery layer. Then marketplaces, social platforms, and retail media became discovery layers. Now we’re adding a new layer: agents that visit your site to decide if you belong in a shortlist. And the shortlist is the new shelf.
If you’re an SME, this is uncomfortable but also an opportunity. Big brands are often slow to fix messy catalogs, inconsistent policies, or content that’s optimized for internal stakeholders instead of buyers. Agents punish confusion.
The new reality: agents browse more than they buy (for now)
In the HUMAN dataset summarized by Search Engine Journal, the bulk of agent requests went to product and search pages, with a much smaller share reaching checkout pages.
This matters because it tells us where competitive pressure will show up first:
- Category pages and internal search results become “AI-readable catalogs.”
- Product detail pages become “evaluation documents.”
- Policy pages (shipping, returns, warranty, contact) become “risk reducers.”
- Checkout becomes relevant later—once agents are allowed (and trusted) to transact more directly.
Agents are doing what good shoppers do at the top of the funnel: scan, compare, filter, validate. But unlike humans, agents can do it at scale, with weird navigation patterns, high request rates, and low tolerance for ambiguity.
If your product page requires three popups, a location gate, a chat overlay, and a cookie wall before the price is visible, an agent may still get through—but the cost is friction and misinterpretation. And in a shortlist world, friction is disqualification.
Standalone agent browsers are losing to embedded agents
One of the more strategic insights in the SEJ summary is that “standalone agent browsers” appear to be losing share to agents embedded in platforms people already use. In the HUMAN view, standalone agent browsers that were prominent earlier declined in relative share over time, while agent activity consolidated around ecosystems.
As a business operator, you should care because it affects distribution and dependency:
- Embedded agents inherit user trust (and default usage) from the parent platform.
- They may present themselves differently to websites, including how they identify requests, authenticate, or appear in logs.
- They can change quickly (UA strings, signature schemes, browsing behavior) without giving you time to adjust.
That combination makes it hard to treat “AI agents” as a single category. Operationally, you’ll manage them like you manage payment methods: the top few drive most outcomes, but edge cases still break things.
Identification is breaking: why “it looks like your activity” changes everything
A key line in the SEJ summary is that Meta’s Muse browsing “will appear as your activity” to the sites it visits (cited in the context of Meta’s engineering communication as referenced by SEJ). Separately, the SEJ piece discusses Web Bot Auth signatures, key directories, and the reality that some portion of requests may arrive unsigned—meaning the receiving site can’t verify identity via signature.
Here’s the business translation:
- Traditional bot management assumes bots identify themselves (user-agent, IP ranges, signatures) and behave consistently.
- Modern agents may blur Attribution (appearing as user activity), may change identification mechanisms, or may partially sign requests.
- So your controls move upstream: edge security policy, rate limiting, endpoint-level rules, and “allow/deny by behavior,” not just by name.
This is where many teams get trapped. They want an “AI agents allowlist,” but the reality is conditional:
- Allow browsing of public product content? Probably yes.
- Allow scraping of inventory endpoints at scale? Maybe, with rate limits.
- Allow checkout automation? Not without deliberate design.
- Allow account access, coupon abuse, card testing vectors? Absolutely not.
And the identification nuance matters because if your security provider expects signed requests from one directory but the agent rotates to another, you can accidentally treat legitimate agent requests as anonymous automation—then block them—then wonder why your brand disappears from AI-generated recommendations.
To be clear: you should not blindly trust any request. But you also shouldn’t blindly block what you don’t recognize, especially on the pages that shape the shortlist.
Security and SEO are now one operational system
For years, SEO teams and security teams have lived in different worlds. SEO wanted Crawling and accessibility. Security wanted control and reduction of abuse. Agentic browsing collapses that separation.
What’s happening now looks like this:
- Agents request pages at volume.
- Security tools detect patterns that resemble automation.
- Security blocks, challenges, or rate-limits.
- Agents fail to evaluate pages (or get partial content).
- Your brand becomes “less recommendable” in AI journeys.
The bad version of this story is a silent failure: no 500 errors, no obvious outage, just fewer recommendations and fewer qualified visits over time.
The good version is coordinated governance. I recommend a simple operating rule:
Security owns protection; marketing/SEO owns discoverability; both co-own “agent access.”
In practice, this means:
- Define which endpoints are “public evaluation surfaces” (category/search/PDP/policy pages).
- Define which endpoints are “transaction surfaces” (cart/checkout/account/payment APIs).
- Set different policies for each surface (rate limits, challenges, signature verification, caching rules).
- Monitor changes in agent behavior and adjust quickly.
If you’re an SME without a security team, you still have this responsibility—because your hosting/WAF settings are security settings. And they can quietly rewrite your growth curve.
Analytics: why your dashboards will lie to you about “users”
Search Engine Journal’s summary notes that HUMAN reported analytics tools on most sites can’t tell AI agents from human visitors. That aligns with what many operators experience in practice: standard analytics implementations were designed for browsers operated by humans, not for delegated browsing through an agent layer.
The implications are not academic. Here are a few ways agent traffic can distort decision-making:
1) Conversion rate panic
If agents hit product pages heavily without buying, your Conversion rate can drop even if your human demand is stable. Teams then “fix” the wrong thing: they change product pages, pricing, offers, or UX based on a measurement artifact.
2) Paid media misattribution
If your attribution model or retargeting pools are polluted by agent sessions, your paid performance reports can drift. You may retarget “visitors” who are not actual prospects.
3) Infrastructure and performance noise
Agent traffic can create real load. But not all load is equal. Some agents fetch more assets, some execute scripts, some re-request URLs quickly. If you don’t separate it, you may overbuild for the wrong peak.
4) SEO reporting confusion
SEO teams often use engagement metrics as directional signals (even if imperfect). If session quality drops due to agents, you can misread content performance. In a worst case, you delete the very pages agents rely on to evaluate your offer.
What to do instead: combine client-side analytics with server logs and edge-layer bot/agent classification, then create a separate “agent traffic view.” Not every SME can do that overnight—but you can start by asking your WAF/bot protection provider what they can export and how classification works.
Practical SEO/AEO/GEO implications: optimize for the agent’s shortlist, not just rankings
“AI Search” isn’t one product. It’s a behavior shift: systems that generate answers, recommendations, and shortlists increasingly influence what humans click—and in some cases, agents may click for them.
So the goal evolves from ranking to being chosen. That’s AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) in practical terms: making your content unambiguous, structured, trustworthy, and easy to verify.
Based on the browsing patterns described in the SEJ/HUMAN summary (product and search pages dominating), here’s where to focus first.
Your product page is now a proof document
A PDP should answer the questions an agent would need to recommend you confidently:
- What is it (clear title, variant logic, compatibility)?
- What does it cost (including fees if possible)?
- Is it available (stock status, lead time)?
- How fast can I get it (shipping zones, cutoffs)?
- What if it doesn’t work (returns, warranty, support)?
- Why trust you (reviews, policies, business details)?
Humans tolerate ambiguity because they can ask questions or improvise. Agents need explicitness.
Category/search pages are the new comparison engine
If agents spend disproportionate time on product listings and search pages, those pages need to be:
- Fast and indexable.
- Not blocked by aggressive bot challenges that hide prices or product grids.
- Canonicalized correctly (to avoid fragmented evaluation across duplicates).
- Consistent in how filters change URLs (so agents don’t get trapped in infinite faceting).
Policy pages become conversion assets even when they don’t “convert”
Many SMEs hide or underinvest in shipping/returns pages. Agents may treat them as trust signals. If the agent can’t find your return window, it may choose a competitor with a clear 30-day policy—regardless of whether you rank higher.
Brand clarity: entity signals are not optional
Even without going beyond the supplied source context, one reality is obvious: agents need to associate your site with a real business. That includes consistent business info, clear contact paths, and clean site architecture.
If your site looks like a thin affiliate layer, an agent may deprioritize you. If your business details are inconsistent across pages, an agent may avoid recommending you due to uncertainty.
AYSA’s perspective here is simple: visibility is now operational. You can’t “content strategy” your way out of broken templates, messy canonicals, or blocked pages.
Technical checklist: make key pages accessible and verifiable
Let’s keep this grounded: you don’t need to redesign your entire site to prepare for agentic browsing. You need to reduce failure modes on the pages agents actually touch.
1) Separate evaluation surfaces from transaction surfaces
Work with your dev/security owner (or hosting provider) to define:
- Public evaluation surfaces: homepage, category pages, internal search pages, PDPs, shipping/returns, store locator, FAQs.
- Protected transaction surfaces: cart, checkout, payment, account, order history, coupon apply endpoints, inventory APIs.
Then apply different WAF/bot settings. Too many SMEs apply one aggressive rule everywhere and unintentionally break discovery.
2) Review bot/agent rules with current identification reality in mind
The SEJ summary discusses how signature verification can be impacted by key directory changes and unsigned requests. The takeaway is not “trust signatures” or “don’t trust signatures.” It’s: your verification assumptions can go stale quickly.
Action items:
- Ask your bot protection provider how they identify agents (UA strings, behavioral signals, signatures, integrations).
- Document what you currently allow/deny and why.
- Create a process to update verification rules when major agent providers change their identity signals.
If you don’t have a bot protection vendor, your CDN/WAF likely has bot features—use them intentionally.
3) Rate limits over blanket blocks
For evaluation pages, prefer:
- Reasonable rate limits per IP/session.
- Cache rules that reduce origin load.
- Graceful degradation (serve simplified HTML when needed).
Blanket blocks are seductive—and often counterproductive.
4) Canonicals, faceting, and internal search hygiene
Agents browsing search and product pages will hit:
- Filter permutations
- Sort options
- Pagination
- Internal search queries
If those create infinite URL spaces or inconsistent canonical tags, agents can waste cycles and form a low-confidence view of your catalog.
Technical basics that matter more in an agent world:
- Correct canonical tags on variants and filtered pages.
- Robust handling of parameter URLs.
- Consistent 200/404 behavior (no soft-404 confusion).
- Fast server response times and stable rendering.
5) Structured data as a reliability layer (not a hack)
I’m not going to invent claims about which exact schemas a given agent uses. But broadly, structured data helps machines interpret products, prices, availability, and policies when page layouts change.
For ecommerce, that typically means validating product structured data and keeping it consistent with visible content. If your markup says “in stock” but your page says “backorder,” you create machine uncertainty.
AYSA can help by monitoring templates and preparing schema and on-page fixes for approval—without you needing to manually audit hundreds of pages every month.
Content checklist: reduce uncertainty, increase machine confidence
Agentic evaluation rewards clarity and penalizes marketing fog. Here’s what I would fix first for most SMEs.
1) Shipping and returns: make them explicit and easy to extract
- Put shipping timelines on PDPs (not only in the footer).
- State return window and conditions clearly.
- Make policy pages accessible (not blocked behind interstitials).
2) Price transparency: avoid “surprise math”
If shipping or fees radically change the total cost, the agent may treat your offer as uncertain. Even if you can’t show final totals early, you can provide ranges, calculators, and clear thresholds.
3) FAQs that answer real pre-purchase questions
FAQs aren’t filler; they’re the quickest path to resolving uncertainty for both humans and agents. Keep them grounded:
- Compatibility
- Sizing
- Materials
- Setup
- Warranty/support
4) Comparison tables and “why choose this” sections
Agents compare. Give them structured comparisons:
- Model differences
- Use-case guidance
- Tradeoffs (honestly stated)
Honesty wins here. Agents are trained to recognize sales language; they prefer verifiable specifics.
5) Trust signals that don’t look like trust theater
- Clear business identity (address, contact, support hours)
- Real reviews (with moderation and transparency)
- Accurate policy content
If you’re a local business (clinic, hotel, home services), the “who are you and where are you located” layer becomes even more important. Agents won’t recommend a business they can’t confidently describe.
For broader context on SEO and AI search, Search Engine Journal tracks developments in AI Search and SEO. Use those as directional reading, but build your strategy around your own data and your own technical reality.
A concrete SME scenario: the agent-first holiday spike that breaks your reporting
Let’s make this painfully practical.
Scenario: You run a mid-size ecommerce brand selling premium home fitness equipment. In late October and November, you normally see rising demand, higher spend, and higher conversion rates.
This year, your analytics look weird:
- Sessions are up sharply.
- Product pageviews are up even more.
- Add-to-cart is flat.
- Conversion rate drops.
- Customer support gets more “Do you ship to X?” questions than usual.
Your team debates:
- “Our offer got worse.”
- “Our ads are bringing junk traffic.”
- “The site is slow.”
- “SEO traffic is low quality.”
But what’s actually happening is more subtle:
- Agents are crawling your product and search pages to build shortlists for users who are asking AI systems what to buy.
- Your PDPs are missing clear shipping timelines, so the agent can’t confidently recommend you.
- Your WAF occasionally challenges the agent on category pages, so it gets inconsistent price visibility.
- Meanwhile, your analytics counts agent browsing as “sessions,” dragging down conversion rate.
The fix is not one silver bullet. It’s a coordinated set of changes:
- Create a separate view of traffic using edge/WAF classification where possible.
- Make shipping/returns information explicit on PDPs and policy pages.
- Ensure category pages and PDPs are accessible without heavy interstitials.
- Apply stricter controls to checkout and account endpoints, not the entire site.
This is exactly where execution systems matter. Because the business risk isn’t “we don’t understand agents.” The business risk is “we make the wrong decision quickly.”
What agencies should rethink (and what to sell instead)
If you run an agency, agentic browsing is going to disrupt your deliverables.
Rank tracking alone becomes less convincing
When the customer journey begins inside an AI interface, “rank #3 for keyword X” is less directly tied to revenue. Clients will still ask for it—but it won’t explain performance swings caused by recommendation systems and agents building shortlists.
Your new offer should look like “agent readiness + execution”
Agencies that win will productize:
- Agent traffic monitoring (edge signals + logs + analytics hygiene)
- Template and structured data QA
- Policy clarity and PDP improvements
- Security/SEO coordination (what to allow, where to challenge, where to block)
- Governance: who approves changes and how fast
That last point—governance—is where most agencies struggle. They can recommend changes, but they can’t ship them. Or they ship them without approval and create risk. In an agent-driven environment, both are losing strategies.
Where AYSA fits: monitored, approved execution for agent-ready SEO
AYSA is built for the messy middle between “we know what to do” and “it’s actually done.” In an agent-first discovery world, that gap is where visibility is lost.
Here’s how AYSA fits into this moment:
1) Monitor what matters continuously
Agent behavior and platform identification can change quickly. You need ongoing monitoring—not quarterly audits. Start here:
- AYSA Monitoring helps track site health signals and issues that impact discoverability.
- AI Search Visibility is the lens: what’s likely to affect recommendation and selection, not just rankings.
2) Prepare fixes tied to business outcomes
Agent readiness typically requires cross-functional updates: template tweaks, content clarity, schema validation, and internal linking adjustments. AYSA helps prepare the work so it’s reviewable and actionable.
- Explore the tooling approach: AI SEO Tools.
3) Ask for approval before execution
Security and SEO changes can break things. AYSA’s model is built around approved execution: proposed changes are reviewed, accepted, and then implemented—so SMEs and agencies can move fast without losing control.
4) Execute what’s accepted
In agentic search, speed matters. If Muse/OpenAI/Gemini-class systems shift how they browse or identify, you don’t want a six-week ticket queue. You want the changes shipped while the market is shifting.
If you’re evaluating options, start with AYSA Pricing and then browse implementation thinking in our AYSA Blog.
One note: AYSA can’t (and shouldn’t) replace your security stack. But it can ensure your site’s SEO and content execution is consistent, fast, and aligned with the way agents evaluate pages—so you don’t lose visibility due to preventable site issues.
What to do next (action list)
If you only do one thing after reading this: separate “evaluation” from “transaction” and stop treating all automation as equally harmful.
Here’s a practical next-step list most SMEs can complete in 2–4 weeks.
Week 1: Establish visibility and ownership
- Identify who owns bot/WAF settings (vendor, dev, IT, agency).
- List your critical “evaluation surfaces”: top categories, top PDP templates, shipping/returns pages.
- Pull a simple report of top requested URLs and user agents from server logs or your security provider (if available).
Week 2: Fix the biggest evaluation blockers
- Remove or reduce interstitial friction on category/PDP/policy pages (where feasible).
- Confirm canonical logic on product variants and filtered pages.
- Ensure policy pages are linked from PDPs and are crawlable.
Week 3: Improve “recommendability” on PDPs
- Add explicit shipping timeframes and return window near the buy box.
- Add a short “fits who / not for who” section to reduce returns and increase confidence.
- Validate product structured data against visible content.
Week 4: Measurement and governance
- Create a separate view or annotation process for suspected agent traffic spikes.
- Set a monthly cadence to review bot/agent rule changes with whoever controls security.
- Implement an approved execution workflow so fixes don’t stall (this is where AYSA is built to help).
Sources and further reading
- Search Engine Journal: Meta’s Muse Now Sends Most AI Agent Requests One Security Firm Sees (primary source for the reporting summarized and discussed here)
- Search Engine Journal: AI Search category (ongoing context and industry reporting)
- Search Engine Journal: SEO category (broader SEO changes that intersect with AI discovery)
- Search Engine Journal: Technical SEO (technical foundations that affect accessibility and evaluation)
AYSA next steps: If you want a system that continuously monitors your site, prepares the changes that improve AI search visibility, asks for approval, and executes what you accept, start here: AI Search Visibility, Monitoring, and AI SEO Tools. Then review Pricing for the right fit.
Author: Marius Dosinescu (AYSA.ai). This editorial uses the Search Engine Journal report linked above as research input and expands it into a practical operating guide for SMEs and growth teams.
Continue the AI search topic inside AYSA.
Use these pages to connect the article with AI SEO tools, AI visibility monitoring, AI Overviews and approved website execution.
Turn this topic into a website action plan.
Use these AYSA hubs to move from reading to technical fixes, AI visibility monitoring, research, glossary context and approval-first SEO execution.