98% of online stores promise you when your order will arrive. None publishes whether they keep that promise.

We read, page by page, what 250 active online stores selling in Spain, across twelve categories, publish about themselves. Almost all of them tell you how long your order will take. None publishes any data that lets anyone check whether they deliver on it: not you, and not an AI assistant buying for you.

1. They all promise; none of them proves it

98% of stores say how long their orders take: “24/48 h”, “2 to 4 working days”, “delivery between 23 and 26 October”.

What none of them publishes is verifiable data on what happens next: what share of their orders arrives, or how long it really takes to arrive. Only 5 of the 250 give a figure of their own (“90% of orders arrive within 48 hours”), but without saying what period it covers or how to check it.

2. One in three doesn't say when you get your money back

Around two in three stores say how many days it takes them to refund the money for a return. The rest don't say so on any of their pages: not under returns, not in the terms and conditions, not in the FAQ. Anyone returning an order to those stores has no idea how long they will have to wait.

Of those that do say, half give 14 days, the legal maximum for refunding the money when the buyer withdraws from the purchase.

3. Three in ten don't say how you get it back

Around seven in ten say they refund the money using the same method it was paid with: to the same card, to the same account. The rest don't say, or only say so for specific cases, such as a cancelled order or a faulty product.

4. Who they are, yes; what they do, no

93% publish their tax ID (CIF or NIF), as the law requires of any website that sells. And almost half show reviews or a rating from their customers.

In other words: most tell you who they are, and many show you what some customers think. What happened to all their orders — whether they arrived, when, whether the money was refunded — none of them shows.

5. When an assistant does the buying

More and more purchases start with a question to an AI assistant: “find me some trainers that will arrive before Friday”. In Spain, most people already use AI when shopping, but hardly anyone lets it have the final say.

Source: 6th Veepee-IESE Study “El futuro del e-commerce en España” (The future of e-commerce in Spain), with Kantar: 805 interviews with online shoppers aged 16 to 75, presented on 19 May 2026.

Every month, more people reach stores from an AI assistant

Adobe measures more than a trillion visits to online stores in the United States. Since it started counting them, in October 2024, visits that come from an AI assistant's answer have grown by 1,324%, and in August 2026 they were still more than double the figure a year earlier. They are no longer just browsing: that month they were 48% more likely to end in a purchase than visits arriving by other routes.

Visits to US online stores that come from an AI assistant, month by month: how much they have grown since October 2024. Values read from the chart in the Adobe Digital Insights report of June 2026; they match what Adobe publishes in writing (+1,324% since October 2024 in May 2026; +269% in one year in March 2026). August 2026 and the purchase figure: Adobe's holiday season forecast. These are visits, not purchases made by an agent.

What an assistant receives from each store

To recommend a store, an assistant can only use what the store publishes about itself. In the public formats through which OpenAI, Google and Shopify receive store data, everything to do with trust is declared by the store itself.

That is why, to an assistant, the store that always delivers on time and the one that never ships say exactly the same thing. Until that is published, it will be comparing promises, and anyone can write a promise.

Where does an assistant already buy for you?

In the United States, several assistants already pay within the conversation. In Spain, for now, they search and compare, and the purchase is completed on the store's website: trust is decided at the moment of choosing.

And where they already pay, the data they receive from each store is the same: what the store declares.

The figures

250 active online stores selling in Spain, October 2026. “≈”: figure corrected with the control review; details in the table.

How it was done

• Written criteria, set before starting. What an active store is and what counts as “yes” for each question, with examples of what doesn't count: the days the buyer has to return an item are not the refund time; “fast shipping” with no figure is not a delivery time; a number of orders sold is not delivery data.
• Which stores. Small and medium-sized independent stores in twelve categories (fashion, footwear, jewellery, cosmetics, home, sport, pets, food, baby, technology and gifts), found through searches, excluding marketplaces and large chains. They were reviewed in random order until we had 250 active stores: ones that load, sell in euros to Spain and have products that can be bought. Along the way, 20 that did not qualify were discarded: closed, in liquidation, behind a password, not selling online, or blocking the review. It is not a sample of all stores in Spain.
• Page-by-page reading. For each store we read what a buyer would see: the home page, a product page, the legal notice, the contact page, the terms and conditions, the privacy policy, returns, shipping and the FAQ. Without logging in or placing orders.
• Every “yes”, with its evidence. Every positive answer stores the page and the exact sentence that supports it. We checked automatically that all 897 sentences were on their page: 896 appear; the missing one is in a PDF document that could not be read automatically.
• The same criteria for all. Where two reviews settled a borderline case differently, a rule was written and applied to every store. Stores whose pages only open with JavaScript were reviewed with a browser.
• Control review. 30 randomly chosen stores were reviewed again separately, without seeing the first answers and with a browser that runs the full pages. The two reviews agreed on 29 or 30 of 30 for what supports the main conclusions (delivery time, tax ID and delivery data). Where they disagreed, the first review's errors were always of the same kind: stores that did publish something that had been missed. None counted as published something that wasn't there. That is why the measured figures are a minimum, and where the difference matters the corrected figure is given.
• Human check. A person opened the pages of those 30 stores one by one and checked each answer with its evidence in view: all 180 were confirmed (six questions per store).
• Date. Pages reviewed between October 21, 2026 and October 22, 2026. We will repeat it every six months with the same method.

What this study does not say

• Anything about any particular store. We do not name any and we will not: store-by-store data is not published.
• That a store that doesn't state its refund time takes longer. It is about what is published, not about what is done.
• That a store with reviews is better or worse. Reviews are useful for what they are useful for. What is missing is something else: facts that can be checked.

Why we did it

Buying online today means trusting a promise. 98% of stores tell you when your order will arrive, and none lets you check whether they keep their word. Not you, and not the assistant buying for you.

holdmark exists to fill that gap: so that before you pay you can know whether a store delivers its orders and refunds money. Not because the store says so, but because its carrier and its payment processor record it. The rules we use to calculate every figure are public: how it is calculated.

And it is a bet on what is coming. Today AI assistants compare promises because they have nothing else. We believe that, once they can read checked facts, stores that show they deliver will start with an advantage over those that only promise it, and that this will weigh on the decision to buy.

How to cite it

You can use the figures in this study by citing the source:

holdmark (2026). What an online store publishes about itself. Study of 250 active online stores selling in Spain, October 2026. https://holdmark.org/estudio/