Salesforce's 2026 Holiday Forecast: 20% of Traffic From AI Agents
Salesforce’s 2026 holiday forecast puts a number on the shift: 20 percent of holiday ecommerce traffic will originate from AI chat agents. That bucket includes consumer-facing bots handling live queries, autonomous agents running backend tasks, and competitor scrapers feeding algorithmic price matching.
This is a vendor forecast from a company that sells agent software, so read the 20 percent with that in mind. The report is useful because it breaks AI traffic into metrics you can check separately against your own analytics.
What the 20 percent actually contains
Salesforce is counting three things together: human visits referred from ChatGPT or Perplexity, sessions produced by a brand’s own on-site agent, and requests from price-comparison programs.
The third group is not people. Folding machine requests into a traffic share inflates the headline. Split the forecast before acting on it: how much human traffic arrives from AI platforms, how many sessions your own agent handles, and how much is bots.
Three supporting figures frame the shift. Engagement on brand-owned properties fell 7 percent year over year. Consumers using traditional search engines and marketplaces to help with shopping each dropped 15 percent. New channels, meaning AI assistants plus social, rose 38 percent.
Brand agents are the actual new front door
The section worth acting on is the performance gap between retailers that run their own agents and those that do not.
Retailers using branded AI shopper agents grew holiday sales 6.2 percent in 2025. Those without grew 3.9 percent. The report projects one in three ecommerce sites will have a site-specific shopper agent live by Cyber Week 2026.
Consumer data lines up. Half of shoppers report using an AI assistant at some point in the buying journey, up 67 percent year over year, and 74 percent say they trust product recommendations from AI chat. On-site, 41 percent say a brand agent answering their questions makes them much more confident about a purchase, another 36 percent somewhat more confident, and 5 percent do not trust the answers at all.
Watch the sampling. The 6.2 versus 3.9 comparison sits inside Salesforce’s own customer base and does not adjust for category mix or brand size. It shows that brands running agents grew faster. It does not mean adding an agent adds 2.3 points of growth.
What SEO now involves
Classic SEO made pages crawlable and rankable. Both still matter and neither is sufficient. Three tasks follow from the report’s data.
Structure product data so agents can read it directly. The report’s recommendation is to treat AI platforms as critical data syndication partners and invest in high-velocity product feeds, structured APIs, and product graphs. A feed is not the same artifact as schema markup. Schema talks to search engines; the feed talks to agents.
Let a brand agent answer questions and route to product pages. On-site behavior splits into 37 percent semantic queries, 26 percent generative tasks such as comparing products or summarizing reviews, and 27 percent traditional search. All three paths need to end at a product page.
State shipping terms where a machine can find them. Free shipping ranked as the top reason shoppers pick a merchant across every income level in the report. Retailers are expected to spend an additional $3 billion subsidizing it this season, a 7 percent year-over-year cost increase. When an agent answers whether you ship free, a clear policy gets you recommended more often.
A checklist before Cyber Week
Start with brand-owned engagement. If it is down year over year, determine whether that is normal variance or AI channels diverting traffic. Google Search Console’s AI reporting and your site search logs will point one way or the other.
Then audit feed completeness. Price, stock, size, and delivery window are the fields agents need. A product missing them gets skipped, which means it does not exist in AI recommendations.
Then check whether your site can answer questions at all. Without a brand agent, you still need a search or support entry point that handles questions like whether a size will fit or when an order arrives, and that links answers to product pages.
Finally, confirm where your free-shipping threshold and return policy are published. Buried in an FAQ, they are hard for an agent to retrieve and hard to quote in a conversation.
The conversion evidence from Prime Day, where AI-referred shoppers converted 40 percent better, is covered in a separate analysis. The four checks above are what to run before the holiday season.
Do not budget off a forecast
The 20 percent, the one in three, and the ninefold growth figures are projections, not results. Salesforce’s inputs come from its own Shopping Index and consumer surveys, with a sample weighted toward merchants using Salesforce products.
Treat the report as an index of things to check. Every projection in it maps to a measurable signal. Treating it as a reason to move a fifth of next year’s budget into AI channels goes too far. Verify direction in your own GA4 and Search Console data before deciding how much to spend.
Related Articles
SEO for a New Trilingual Site: What We Did in the First Month (and What We're Still Measuring)
A first-month SEO log for a small site in English, Simplified Chinese and Traditional Chinese. Google started showing it in search; Bing had read the sitemap and still hadn't crawled the homepage. What we set up on day one, what the early data showed, why Google shows a globe instead of our favicon, where the first links came from, and what we chose not to do.
Amazon Seller Assistant in Claude: A Setup Guide for US Sellers
Amazon announced the Selling Partner plugin at Accelerate on September 23, 2026, bringing Seller Assistant into Amazon Quick and Claude. What it can read and change, the three always-on workflows to set up first, and the limits: no ads, one account at a time, US stores only.
AI Visibility Platforms in 2026: Auto-Fix Agents vs Read-Only Trackers
Old GEO tools tell you where you rank in AI answers. New platforms like Lantern go a step further and edit your catalog and product pages for you. Auto-remediation sounds like a time-saver, but handing rewrite rights to an AI has real trade-offs. Here is how the agent platforms stack up against read-only trackers.