Skip to content

Barometer · October 2026

46 Belgian online shops, seen by the AIs.

On 9 October 2026, I asked seven AIs where to buy what these shops sell, in French and in Dutch. The brand never appears in the question.

Get my result

By ordering, you accept the Snapshot terms.

  • 46independent Belgian shops
  • 9,568answers requested
  • 2languages: French and Dutch

The figures

ChatGPT never cites 41% of the shops.

  • 41%never cited by ChatGPT (19 of 46)
  • 41%cited in one language only, or in one mode only (19 of 46)
  • 17%cited everywhere (8 of 46)

9% of the shops (4 of 46) get no citation at all, across the 7 AIs, in both languages and both modes.

In their place

The AIs never say that they do not know. They give another address.

Share of the top 3 lists given in place of the Belgian shops1,187 lists read
Bol.com11%131 of 1,187
Zalando7%
Decathlon6%
Douglas4%
Dreambaby4%

Two languages

French and Dutch are two different measurements.

Across the 7 AIs, a Belgian shop is cited in 21% of the answers in French and in 17% in Dutch. One shop goes from 75% in French to 49% in Dutch.

A Belgian shop, across the 7 AIs9 October 2026
French21%
Dutch17%
Dutch, with web search14%one answer in seven

The good news

Eight shops are cited everywhere.

Brouwland is the most cited: 86% of the answers from the 7 AIs name it.

  • Brouwland
  • Foodbag
  • Luminette
  • Dierendonck
  • Tierenteyn-Verlent
  • Baobab Collection
  • Impermo

Start

Start with one measurement.

$250, once. Two markets, measured now and measured again in 30 days. If the report teaches you nothing, I refund it.

By ordering, you accept the Snapshot terms.

See a real report
Johnny Lagneau, founder of VectorGapI run the measurement and I present the report to you myself.

Belgian e-commerce · October 2026

Barometer: 46 Belgian online shops and the AIs, October 2026

When a Belgian buyer asks ChatGPT a question, 41% of the shops (19 of 46) are never cited. ChatGPT sends the buyer to Bol.com, Zalando or Decathlon instead.

My name is Johnny Lagneau, founder of VectorGap. Since September, I have spent my days asking the AIs the same question in several languages and several cities. This month, I took 46 independent Belgian online shops that sell in French and in Dutch, and I looked at what the AIs say about them. Here is what I found.

What I measured

For each shop, I asked 8 typical buyer questions (“Where can I buy a cargo bike in Belgium?”, “Which Belgian chocolate brand makes a good gift?”, that kind of question). I asked them in French, then in Dutch. I asked them to 7 models: ChatGPT, Gemini, Claude, Perplexity, Grok, Mistral and Muse Spark (Meta). Twice each: once without web search, once with web search on (Perplexity with web search only).

The brand never appears in the question, so the answers are not steered. I simply count whether the model cites it on its own.

In total: 46 shops × 2 languages × 8 questions × 13 series (7 models with web search, 6 without web search), which makes 9,568 expected answers and a little over 9,000 read. All measurements date from 9 October 2026.

The figures

41%

41% of the shops (19 of 46): ChatGPT never cites them. Not in French, not in Dutch, not without web search, not with web search.

of 46 shops

41%

41% (19 of 46): ChatGPT cites them in one language only, or in one mode only.

of 46 shops

17%

17% (8 of 46): ChatGPT cites them everywhere.

of 46 shops

How the shops split according to ChatGPT's answers
CategoryShopsShare
41% of the shops (19 of 46): ChatGPT never cites them. Not in French, not in Dutch, not without web search, not with web search.19 of 4641.3%
41% (19 of 46): ChatGPT cites them in one language only, or in one mode only.19 of 4641.3%
17% (8 of 46): ChatGPT cites them everywhere.8 of 4617.4%
Each square is one shop. The colours match the categories in the table.
  • Across the 7 models together, a Belgian shop is cited in 21% of answers in French and 17% in Dutch. In Dutch with web search, 14%: one answer in seven.
  • 9% of the shops (4 of 46): zero citations, across the 7 models, in both languages and both modes, about 200 answers each. I do not name them.
  • Bol.com appears in 11% of the top 3 lists the AIs give in place of the Belgian shops (131 of 1,187). Zalando in 7%, Decathlon in 6%, Douglas and Dreambaby in 4%.
  • The gap between the two languages: one shop is cited in 75% of answers in French and 49% in Dutch. Three other shops go from 26 to 32% in one language to under 10% in the other. Only one goes the other way (cited more in Dutch).

Who the AIs name in their place

The AIs do not answer “I do not know this shop”. They give another address. In Dutch, Bol.com is in 33% of ChatGPT's top 3 lists (15 of 46), and in first place in 13% (6 of 46). In French: Zalando, then Bol.com, Decathlon, Dreambaby and Farmaline. The big Belgian brands sold through retailers (watches, beer, children's fashion) are hit hardest: the AI answers “where to buy” with a shop, not with the brand.

The good news

Eight shops (17%) are cited by ChatGPT in both languages and both modes: Brouwland, Foodbag, Luminette, Dierendonck, Tierenteyn-Verlent, Baobab Collection, Impermo, and an eighth that I do not name. Brouwland is the most cited of all: 86% of the answers from the 7 models name it. If you are on this list, well done. Share it.

What it means for a shop owner

  • The AIs never say “unknown”. They send the buyer elsewhere, to Bol.com, Zalando or Decathlon. Your customers leave without you seeing it.
  • French and Dutch are two different measurements. Being cited in three answers out of four in French guarantees nothing in Dutch.
  • A model's memory and its web search give different answers. Some shops are known to the models and disappear as soon as web search turns on; others do the opposite. What web search reads can change as soon as your pages change. A model's memory waits for its next version.

Method and limits

  • 46 shops measured out of the 48 prepared: two were blocked before the measurement.
  • One pass per question. The measured margin is 6 cells out of 48: a gap of 1 or 2 answers out of 8 can come from chance. A single measurement is a snapshot. I will run the same measurement on the same shops in November.
  • Mistral answered 66 to 74% of the questions (the provider's rate limit). The six other models are between 98 and 100%. A missing answer counts as “not received”, never as 0.
  • Perplexity searches the web even when I ask it to answer without searching. I do not count it in the series without web search.
  • The two modes did not run on the same versions of the models. So I compare the two languages with confidence, and the two modes with caution.
  • The names cited in place of the shops are not yet checked one by one (a site that responds).
  • Every figure carries an audit identifier. I can show the exact answer behind each line.

To go further

If your shop is among the 46 and you want your detailed result (the 7 models, both languages, both modes), write to me: gap@vectorgap.ai. I will send you the link the same day. If you are not among the 46, I can add your shop to the November measurement.

Get my resultJoin the November measurement

Johnny Lagneau, founder of VectorGap

18 October 2026

You are named here and want a correction or a removal? Write to gap@vectorgap.ai.