AI ecommerce picks in Finland: 36 answers, 6 models.

When a Finnish buyer asks an AI model “what is the best ecommerce platform for a small business,” the answer is already decided by four names. Share of voice in AI search is the percentage of AI answers that mention your brand at all, and in this market it is now measurable for the price of a coffee. I ran 6 models against 6 Finnish buying questions in June 2026 and counted every platform named across 36 answers. The result is not a vibe. It is a table.

This matters because the funnel moved. A growing share of “which tool should I buy” research happens inside ChatGPT, Gemini, and Perplexity before anyone opens a Google results page. If the model never names you, you are not on the shortlist — and you will never see it in your analytics, because there was no click to measure. So I measured the answers directly.

How I measured it

I used the DataForSEO AI Optimization API to send identical prompts to 6 models with web search enabled, then parsed every response for platform names and cited sources.

The 6 models span four providers and two cost tiers: OpenAI’s gpt-5.5 and gpt-5-nano, Anthropic’s claude-opus-4-8 and claude-haiku-4-5, Google’s gemini-3.1-pro, and Perplexity’s sonar-pro. The 6 prompts are the questions a real Finnish founder types, in Finnish: “Mikä on paras verkkokauppa-alusta suomalaiselle pienyritykselle?” (what is the best ecommerce platform for a Finnish small business), a budget variant, a startup variant, a head-to-head (“MyCashflow vs Shopify vs Vilkas”), and a B2B variant. Six models times six prompts is 36 answers.

For each answer I recorded three things: which platforms were named, which platform appeared first (a proxy for the model’s top pick), and which web pages the model cited as sources. The entire run cost $2.56. That number matters on its own — measuring your AI share of voice is no longer a six-figure brand-tracking contract. It is a script you can run on a Tuesday.

The six prompts, sent verbatim to every model:

1. “Mikä on paras verkkokauppa-alusta suomalaiselle pienyritykselle? Suosittele 3–5 vaihtoehtoa ja perustele.”
2. “Suosittele verkkokauppa-alustaa Suomessa toimivalle verkkokaupalle. Anna konkreettiset vaihtoehdot.”
3. “Paras edullinen verkkokauppa-alusta suomalaiselle aloittavalle yrittäjälle?”
4. “Mikä verkkokauppa-alusta kannattaa valita kun perustan verkkokaupan Suomessa? Vertaile vaihtoehtoja.”
5. “MyCashflow vs Shopify vs Vilkas — mikä näistä sopii parhaiten suomalaiselle verkkokaupalle?”
6. “Mikä verkkokauppa-alusta sopii suomalaiselle B2B-verkkokaupalle? Suosittele ja perustele.”

The set covers generic best-of, budget, startup, general comparison, head-to-head between the three known domestic leaders, and B2B. In Finnish throughout — because that is how Finnish buyers actually type the question, and because language choice changes which retrieval set the model pulls from.

One caveat worth stating up front: web search makes these answers non-deterministic, and model versions move. This is a snapshot, not a constant. The point is the method and the gap it exposes, not a leaderboard frozen in amber.

Four platforms own the answer

Across 36 answers, four platforms appear in three out of four responses or better. Everyone else is a rounding error.

The headline is the second row. MyCashflow, a Finnish platform, is the clear number two — ahead of WooCommerce and Vilkas, behind only Shopify. When I tracked which platform each model named first, the same picture held: Shopify led with 15 first-mentions, MyCashflow took 12, WooCommerce 6, Magento 3. So MyCashflow is not just present. It is the default domestic recommendation an AI gives a Finnish buyer.

That cuts against the usual assumption that global platforms with enormous English-language footprints drown out local players in AI answers. In this market they don’t. A domestic platform that has invested in Finnish-language content and gets cited by Finnish comparison sites can hold the number-two slot against Shopify’s gravity. The corollary is harsher: if you are platform number five through ten, you are effectively invisible. Squarespace, Clover Shop, and PrestaShop combined appear in fewer answers than Finqu alone.

The same question, different answers per model

Here is the finding that should change how you think about AI visibility: there is no single “AI answer.” There are six of them, and they disagree.

Finqu is the cleanest example. It was named in 8 of the 12 answers from Claude (claude-opus-4-8) and Perplexity (sonar-pro) — and in 0 of the 12 answers from OpenAI’s gpt-5.5 and gpt-5-nano. Not one. The two OpenAI models defaulted to the four biggest names and stopped. Claude and Perplexity reached further down the market and surfaced Finqu and Clover Shop as real candidates.

This is the practical reason share of voice has to be measured per model, not as one number. A platform can be a strong recommendation on Perplexity and a non-entity on ChatGPT, and the buyer’s choice of tool decides which reality they see. If your customers skew toward one assistant, that is the answer set you are actually competing in. Averaging across models hides the gap that matters.

It also tells you something about retrieval. The models that surfaced smaller platforms are the ones leaning harder on live web results for niche, non-English queries. The ones that played it safe fell back on the most prominent names — which, for an under-the-radar brand, is the worst-case behavior. Closing that gap is an exercise in getting cited by the sources those models retrieve, not in publishing more pages on your own site.

Who AI actually cites

This is the part that turns the study from a scoreboard into a playbook. Every answer came with citations — the web pages the model pulled from. Tally them across 36 answers and a small, knowable set of sources runs the entire category.

Read that table the way a strategist should. The single most-cited source across the whole study is one comparison site, tyokaluvertailu.fi, with 59 citations. A solo blog, sampsavainio.fi, was cited 20 times — more than Vilkas’s own website. A payments provider, Paytrail, gets cited 9 times for a question that was never about payments, because its compatibility pages list every platform and AI treats that list as ground truth.

The lesson is the lesson of generative engine optimization generally: AI answers are downstream of a thin layer of trusted intermediaries. In Finnish ecommerce, that layer is maybe six comparison sites and a couple of independent blogs. You do not influence the answer by shouting louder on your own domain. You influence it by being present, accurate, and favorably described on the pages AI already retrieves.

For a platform, that means three concrete moves. Get listed and correctly specced on every comparison site the models cite. Earn mentions in the independent blogs that punch above their weight — sampsavainio.fi did more for platform visibility here than most brand budgets do. And make sure your own pages state the facts AI needs in plain, extractable language, because your own site is still cited (MyCashflow’s was, 22 times) when it answers the question directly.

What this means if you sell into the market

If you run marketing for one of these platforms, the strategic picture is now legible instead of anecdotal.

You can see your real position. Not “we think we show up in ChatGPT sometimes,” but “we are named in 33% of answers, zero of them on OpenAI, and the gap is Finqu’s missing presence on the sources gpt-5.5 retrieves.” That is a brief, not a feeling.

You can see the model split and prioritize. If your buyers research on Perplexity, you are in better shape than your ChatGPT share of voice suggests, and vice versa. The fix is provider-specific.

You can see the leverage points by name. The citation table is a target list. Six sources, ranked by how often AI trusts them. Working that list is cheaper and faster than trying to move your own domain’s authority, and it compounds, because once a comparison site describes you accurately, every model that retrieves it inherits the description.

And if you are not a platform but a B2B company in any vertical, the method transfers without modification. Swap the platform names for your competitors, swap the Finnish buying questions for your market’s, and run the same 36 answers. The same three outputs fall out: your share of voice, your model-by-model gaps, and the named sources you need to win. This is the same discipline as classic entity-based SEO — make the model confident about who you are and what you do — applied to answer engines instead of the blue links.

How to run this for your own market

The build is short. Pick 6 to 10 prompts that mirror how buyers actually phrase the “what should I buy” question — including a head-to-head between you and your two main rivals, because comparison prompts surface the sharpest data. Pick a model set that covers the assistants your buyers use, and include at least one budget tier, because the cheap models behave differently and many users are on them. Enable web search; without it you are measuring training data, not the live answer.

Then parse every response for two things: brand mentions and cited URLs. Mentions give you share of voice. Citations give you the target list. Re-run it monthly, because web-search answers drift and model versions ship, and a single snapshot can mislead. I keep the raw responses so I can diff them over time and catch a competitor’s new citation before it calcifies into the default answer.

The whole apparatus cost $2.56 to run once. The reason most teams have no idea what AI says about them is not cost. It is that they are still measuring clicks, and AI answers increasingly do not produce one. Measure the answer instead. Then go change it.

The four platforms that own this answer did not win it by accident, and the smaller ones losing it are not losing on product. They are losing on presence in a thin layer of sources that, until you measure it, you cannot even see.