What it actually means when AI reads your shop instead of sending you visitors
There's a specific kind of meeting happening in eCommerce companies right now.
Someone pulls up Search Console. Average position: unchanged. Impressions: flat, maybe slightly up. Then they open Analytics, and sessions are down double digits. Nobody did anything wrong. No migration, no penalty, no redesign. The numbers just stopped agreeing with each other.
The usual response is to go hunting for the mistake. A broken canonical, a slow LCP, a botched hreflang. Sometimes there is one. Increasingly there isn't. What changed is not your shop. It's what happens between the search and the visit.
Most of what hits your server isn't a customer
Start with who is actually knocking.
As of mid-2026, Cloudflare Radar puts bot traffic at 57.5% of all web requests, against 42.5% human. Imperva's read on 2025 landed in the same place at 53%.
So more than half of everything your infrastructure serves is a machine. That part isn't new. Search engines have crawled forever. What's new is the mix: roughly 20% of verified bot traffic now comes from AI crawlers, and about half of those requests exist to collect training data rather than answer a live question.
This matters because of the exchange rate. Googlebot has always been a reasonable trading partner, crawling a lot and sending a lot back, at roughly 5 crawls per referred visitor. The AI crawlers are a different deal:
PerplexityBot: around 111 crawls per referral
GPTBot: around 1,276 crawls per referral
ClaudeBot: around 11,000 crawls per referral, down from roughly 24,000 earlier in the year
Read those as business numbers, not technical ones. Your catalogue is being read thousands of times to produce a single visit. The reading isn't the problem. The ratio is.
The click stopped being the default outcome
Here is where the flat-rankings mystery resolves.
About 65% of Google searches now end without a click at all. On mobile it's 77%. Not "the user clicked a competitor." Nobody clicked anything, because the answer was already on the results page.
When an AI Overview appears, organic CTR falls by roughly 18% on average. For the position-one result specifically, it drops from about 31.7% to 19.8%, a decline of 37.5% for the exact same ranking you held last year. One analysis puts CTR with an AI Overview present at 0.61%, against 1.62% without.
Category-level numbers make it concrete. Organic click share for headphones fell from 73% to 50%. Jeans went from 73% to 56%. Greeting cards, 88% to 75%.
Nothing inside your shop moved. The surface in front of your shop moved.
That distinction decides your budget. "We lost traffic" and "we lost clicks per impression" are different problems with different fixes. If impressions held and CTR collapsed, you don't have an SEO problem in the traditional sense. You have a distribution problem, and no amount of on-page work will close it.
Some of the missing visitors are sitting in "direct"
Before writing those visitors off, check where they went.
Most AI assistants pass no referrer, or pass one your analytics doesn't recognise. The result is that a real share of AI-driven visits lands in your direct traffic bucket, mixed in with people typing your URL from memory. If direct has quietly grown while organic fell, that isn't a rounding error. That's a new channel arriving unlabelled.
It's worth finding, because the quality is unusually good. Adobe measured 393% year-over-year growth in AI-referred US retail traffic in Q1 2026, converting about 42% better than other channels. The zero-click research points the same direction, with surviving clicks converting around 23% higher, because the person arriving has already read the summary and decided.
Fewer visits, warmer visits. That changes what a traffic decline is actually worth, but only if you can see it.
The trade nobody wants to name
So block the crawlers. Two lines in robots.txt, a rule at the CDN, done. Origin load drops. Bandwidth drops. The freeloading stops.
It also removes you from the answers.
That's the whole trade, and it is a real trade rather than a trick question. The same systems consuming your bandwidth are the ones deciding whether your shop gets mentioned when a customer asks which retailer to buy from. Brands cited in AI Overviews see roughly 35% higher organic CTR even on the clicks that still happen. Being inside the answer carries value that never shows up in a crawler log.
The mistake is making this decision once, globally, for every bot. It isn't one decision. It's a list:
Training crawlers take your content and return nothing measurable this quarter. Blocking them is defensible, as long as you accept you are also opting out of the next model's picture of your catalogue.
Answer crawlers fetch a page because a real person is asking a question right now. Blocking those costs you the citation and the visit that follows it.
Impostors are the easy ones. A surprising share of traffic claiming to be Googlebot fails a reverse DNS check. That traffic is pure cost with no upside, and it's the first thing to cut.
Most shops can't tell these apart today, which is why the conversation stalls at "block everything" or "block nothing."
What to actually do next
Four steps, in order, none of them glamorous.
Read your own server logs. Not analytics. Logs. Analytics only sees visitors who run JavaScript, which means it is blind to the majority of what your server handles. Verify crawlers by reverse DNS rather than trusting the user agent string.
Split the loss. Pull impressions and CTR by query cluster, year over year. Flat impressions with falling CTR points at AI answers. Falling impressions points at something you did. Those get different owners and different budgets.
Audit direct traffic for the AI visits already arriving. Tag them, then look at their conversion rate before you decide what this channel is worth.
Decide bot by bot, write the decision into robots.txt, and enforce it at the CDN where it can actually be measured. Then measure it.
None of this stops AI from reading your shop. That was never on the table. What it does is let you answer the question your leadership is already asking, which is not "why did traffic drop" but "is this costing us money, and which part."
Those are answerable. You just can't answer them from a rankings dashboard.
0 comments
No comments yet.