Most year-in-review pieces about ecommerce are assembled from other people’s press releases. This one is not. Everything below is either an observation from a dataset of 153,515 verified online stores or an argument built on top of it, and where something is an opinion it is labelled as one. No surveys, no vendor forecasts, no borrowed statistics.
The audience here is people who sell to merchants — agencies, freelancers, founders of tools and services. What follows is a picture of the market you are actually prospecting into, which turns out to be meaningfully different from the market the discourse describes.
The platform map is not the one you think it is
Sixty distinct ecommerce platforms are represented across the dataset. But the distribution is savagely concentrated: WooCommerce accounts for 58,549 stores and Shopify for 50,002, which is roughly seven stores in ten between just two platforms. PrestaShop follows at 16,501, JouwWeb at 15,477, then a long drop to Odoo at 4,510, Squarespace at 2,447 and Magento at 1,761. Those seven platforms together cover about ninety-seven percent of everything observed. The remaining fifty-three platforms share what is left.
Two things follow from that shape. First, the English-language conversation is disproportionately about Shopify, which is about a third of the market — a large third, but a third. If your positioning is “Shopify agency”, you have voluntarily excluded two thirds of the stores that exist. WooCommerce is the larger population and is served by a noticeably thinner supply of specialists at the professional end.
Second, the regional platforms are a real and badly covered opportunity. JouwWeb at 15,477 stores is bigger than Odoo, Squarespace and Magento combined, and it is essentially invisible in the international agency market. A platform that dominates one language market and nowhere else produces a segment where you can be the obvious specialist within weeks rather than years. That is not a growth hack, it is just where the underserved demand physically is.
AI visibility is the genuinely new layer this year
If one thing separates 2026 prospecting from 2024 prospecting, it is that AI visibility became measurable at population scale. Across 118,363 stores there are now two signals that did not exist as routine fields before: how often a store is mentioned in AI-generated answers, and how many of its pages get cited as sources. Roughly three quarters of the dataset carries them.
What makes this genuinely novel rather than a rebrand of existing SEO metrics is that the two do not move together. A store can hold a solid organic keyword footprint and register essentially nothing on mentions and cited pages. Another can be mentioned repeatedly while its classic rankings are unremarkable. These are different distribution systems with different inputs, and treating one as a proxy for the other is the most common analytical error being made right now.
For anyone selling services, this creates a diagnostic that did not exist eighteen months ago and that almost no merchant can run for themselves. Two segments fall straight out of it. Stores with real organic traffic and zero AI presence have a specific, nameable gap. Stores with AI mentions but no cited pages are being talked about without being sourced, which is a different problem again. Both are conversations a merchant has never had, which is exactly what you want from an opening line.
The honest caveat: this layer is young. What it measures is real, but the relationship between these signals and revenue is still being established, and anyone telling you they have a proven conversion model for AI visibility is ahead of the evidence. Sell it as a diagnostic and a positioning risk, not as a guaranteed revenue lever.
Traffic stopped being a single number
Real worldwide traffic data covers 118,324 stores, and organic keyword counts cover 80,593. But the more useful structure is the history: each store carries a trend of roughly 170 monthly buckets of organic traffic and organic positions. That is over a decade of monthly observations, which changes what a traffic figure means.
A single monthly number tells you size. A sequence tells you direction, and direction is what predicts whether someone will take a meeting. A store doing five thousand visits a month on a curve that has halved since last year has a problem they can feel, a budget conversation already happening internally, and a reason to answer an email. A store doing twenty thousand on a flat line has none of those things. Prospecting on absolute size is how you end up pitching the companies least motivated to change.
The practical version: build your filters on deltas rather than levels. Percentage change over three, six and twelve months. Stores that crossed below their own two-year average. Stores whose organic positions improved while traffic fell, which is the fingerprint of a search results page that changed shape underneath them. Every one of those is a specific, verifiable observation you can put in the first line of an email, and specific verifiable observations are the entire remaining edge in outbound.
Local presence is still only partly mapped
Of the 153,515 stores, 44,132 have a matched Google Business profile — a little under thirty percent. That number is worth sitting with, because it splits the market into two segments with completely different sales conversations.
For the matched set you get a rich block: place ID, name, category, rating, review count, address, city, postal code, coordinates, whether the profile is claimed, and total photos. That is enough to build an audit before any contact. An unclaimed profile is a straightforward opening. A rating that has drifted below the segment norm with a low photo count is another. A high review count with a mediocre rating is a third, and a completely different pitch from a profile with twelve reviews and no activity.
The unmatched seventy percent is more interesting than it looks. Some of those stores are genuinely pure-play online with no physical footprint, and for them local is irrelevant. But a real portion are businesses with a physical presence and no properly linked profile — which is not a service opportunity so much as a whole missing channel. Distinguishing the two requires judgement rather than a filter, and judgement is where the margin is.
Prospecting became a query, and that changed who wins
The operational shift of the last two years is that list-building stopped being work. Between a REST API and an MCP server available on every plan from forty-nine euros a month, an agent can size a market, build a filtered list, pull each store’s catalog, traffic history, keyword footprint, AI signals and local profile, and draft outreach that references all of it — in one session, without a human touching a spreadsheet.
The consequence is that the scarce skill moved. Finding stores is no longer an advantage; anyone can do it and everyone will. What is scarce is judgement about which slice to attack, which signal actually implies a purchase, and what to build for that slice that a competitor cannot copy in an afternoon. The people struggling in 2026 are not the ones with worse tools. They are the ones who mistook the tool for the strategy.
Two guardrails, learned the hard way. Count before you search — sizing a filter is cheap and prevents an entire campaign built on a segment of forty stores. And force every personalized line to cite the field it came from, because an agent left unsupervised will happily write flattering sentences that are not true, and one hallucinated compliment in a first email destroys the credibility the data was supposed to buy.
What we do not have, said plainly
A state-of-the-market piece from a data vendor is worth exactly as much as its willingness to describe its own limits, so here are ours. There is no advertising data in this dataset. No ad creatives, no ad libraries, no estimated ad spend, no competitor campaign intelligence. If your offer depends on knowing what a merchant is running on paid channels, you need a different source, and any tool that implies otherwise about us is describing something that does not exist.
Coverage is also uneven by design. Traffic exists on 118,324 stores, keywords on 80,593, AI signals on 118,363, Google profiles on 44,132 — out of 153,515 total. A filter that requires all four simultaneously returns a much smaller universe than the headline number suggests. That is not a defect, it is what honest coverage looks like, but you should plan campaigns against the intersection you actually need rather than the total.
The reason to say this out loud is commercial, not moral. Offers built on data that turns out not to exist fail in delivery, in front of a client, after the money has changed hands. Knowing precisely where the map ends is what lets you sell confidently everywhere it does not.
What to do with the rest of the year
If you take one thing from all of this: pick a cell and own it. One platform, one country, one catalog size band. Size it before you commit. Build one artifact you can produce repeatably for every store in that cell — an audit, a report, a comparison against the segment. Lead with the signal nobody else is showing them, which right now is the AI-visibility layer, because it is the only field in the stack that most merchants have never seen a number for.
Then run the boring loop. Filter, artifact, outreach, deliver, measure, adjust the filter. The compounding does not come from any single clever move; it comes from running the same loop forty times against a segment small enough that you eventually know it better than anyone selling into it.
The market did not get harder this year. It got more legible — more measurable, more filterable, more queryable. That helps the people willing to be specific and hurts everyone still sending the same email to everyone. Which of those you are is a decision, and it is the only one that matters going into next year.