Every retailer plans for the holiday spike. This year, the spike has a second layer that retail and ecommerce website leads should understand.
Adobe expects U.S. online holiday sales to reach $275.1 billion, and it expects AI-referred traffic to retail sites to grow 130% year over year, with the biggest jumps on Thanksgiving and Black Friday. The RealReal has just rolled out its AI shopping agent to all 45 million members. And retailers still don’t agree on whether these agents are welcome at all. Amazon sued Perplexity last year to stop its browser agent from making purchases on shoppers’ behalf.
Whatever side a retailer lands on, the infrastructure question is pretty much the same. Peak season now brings two kinds of visitors, people and the software shopping for them, and the two don’t behave the same way.
AI Agents Don’t Browse Like People
When a living, breathing human lands on a product page, they scroll, and maybe open two more tabs. An AI agent comparing options for a shopper can request dozens of product and category pages in seconds, across several stores at once. That agent doesn’t wait patiently either. Agents that fetch pages on demand work with short timeouts, and a page that loads slowly or not at all is simply skipped.
That traffic gets piled on top of the human peak, not instead of it. A site that’s been scaled for last year’s busiest day may not have the headroom it needs the next time around.
Decide Your Bot Policy Before The Spike
Not all automated traffic is the same. Search and citation crawlers help shoppers find you. An agent fetching a page because a shopper asked it a question is, in effect, the shopper. Other bots collect website content to train AI models without directing customers back to the site. Treating all automated traffic the same can mean blocking potential shoppers while allowing bots that offer little value to consume website resources.
Make the decision deliberately, write it into your rules, and remember that a robots.txt file is a request, not a lock. Bots that ignore it need to be handled at the firewall or CDN. We’ve written up how to block or welcome bots and AI crawlers, and most of it applies on any host.
Check That Your Firewall Isn’t Turning Away Buyers
Bot protection is tuned to stop scrapers and credential stuffing, which is the right priority most of the year. But the same rules can challenge or block an agent that is trying to load a product page for a real shopper. Rate limits written for scrapers can cut off legitimate comparison traffic at exactly the moment it matters.
Before your next peak, review firewall and bot-management logs for blocked requests on product, search and checkout pages. If legitimate agents are being challenged, adjust the rules now, while there’s time to test the change.
Speed For Machines, Freshness For Prices
For agents, server response time matters more than how fast a page looks finished in a browser. Caching product and category pages at the edge is the most effective way to keep response times low under load.
The catch is that agents will also quote prices and stock to shoppers. A cached page with yesterday’s price can turn a sale into a complaint. Short cache lifetimes for price and stock data, and purging the cache automatically when they change, let you keep both speed and accuracy.
Load-test the real peak. Most load tests hit the homepage. Real holiday traffic hits search, category filters, inventory lookups and checkout, which are the parts that can’t be served from cache. Test those, at a volume that includes automated traffic, and include the third-party scripts and payment services your checkout depends on. One slow external script can hold up an otherwise healthy page.
Watch the right signals during the rush. During peak days, separate automated and human traffic in your analytics and logs so one doesn’t hide problems in the other. Set alerts for checkout errors, sudden spikes in blocked requests and server response time, not only on whether the site is up. And make sure someone who can change a firewall rule or purge a cache is on call through the busiest days.
No need to pick a side yet. The debate over whether AI agents should be allowed to shop on people’s behalf won’t be settled this season. Retailers don’t need to settle it either. What they need is a deliberate decision about which traffic they want, infrastructure that reflects that decision, and enough headroom to handle both kinds of visitors at once.
The retailers who struggle this year won’t be the ones who chose the wrong side. They’ll be the ones who never noticed the second wave of traffic until it arrived.
Nickola Naous
is co-founder and CTO of Flashcloud, a web hosting company. He has nearly two decades of experience building and running hosting infrastructure, and writes about performance, security and how websites handle AI-driven traffic.





