LLMs Don’t Search Like Google - and This Study Shows Why
LLMs Don’t Search Like Google - and This Study Shows Why
Carine Pire
December 19, 2025
.
Using the same web content, ChatGPT and AI Mode do not come up with the same list of gifts for a Belgian. Behind these differences lie not so much chance as very specific algorithmic, SEO, and cultural logic... and an indicator of who will win, or lose, the battle for visibility in the age of AI.
When asked for Christmas decorating ideas, ChatGPT tends to suggest scented candles from major French brands, designer gift sets, and a few “catalog” items, as if you were walking through an international department store. AI Mode, on the other hand, readily suggests Maisons du Monde, Ikea, Hema, Dille & Kamille, Pols Potten, or VTWonen, with cowboy boot-shaped vases, tomato lamps, and crystal barometers that look like they came straight out of a Pinterest feed.
These responses may seem random, even unimaginative. However, the AIs are working with the same raw material: the web. But they process it differently. ChatGPT synthesizes an “ideal model” of a gift, shaped by years of global content, while AI Mode applies the hierarchy of Belgian and Dutch search results in real time. This discrepancy is neither inconsistent nor random: it reveals structured biases—technical, economic, and cultural—that determine which brands, products, and categories will win in a future where AI will increasingly serve as a gateway to consumption.
ChatGPT works like a super editor who has read a huge library of gift content prior to 2025: buying guides, blogs, product descriptions, press articles, and brand posts. When asked for “tech gift ideas for $150,” it doesn't necessarily check prices in real time; it uses statistical patterns to compile a list that resembles the most popular gift guides: major international brands, consumer tech products, perfumes, jewelry, gift sets, chocolates.
As a result, its answers are very “canonized.”
Analysis of citations shows a strong overrepresentation of reference sites, large e-commerce retailers, and major brands, and little sensitivity to geographical nuances: a Belgian, French, or Canadian user will often receive very similar suggestions if the prompt does not emphasize the local dimension.
Overrepresented brands: Apple, Sony, Amazon, Sephora, Louis Vuitton, Chanel, Dior, Bongo/Wonderbox.
Leading categories: tech, perfumes, jewelry, gift boxes, chocolates—the classics of gift guides.
ChatGPT reflects the most global and stable layer of gift-giving culture, but with little geographical nuance.
As a result, its responses are very “canonized.”
AI Mode is connected to the Google Search ecosystem. For a gift-related query, it breaks down the request into several sub-questions and simultaneously queries Google's index (via FastSearch) and the Shopping Graph, in the user's language and country, before generating a summary.
As a result,
it brings up many more sites that are already present in the “traditional” results, including local players: Belgian wine merchants, concept stores, Dutch decoration shops, and Dutch gift platforms.
It will more readily draw on the decor brands and marketplaces that dominate the results in Belgium or the Netherlands, using their visual and sometimes “WTF” product selections highlighted in SEO content and local product catalogs.
To take the example again: when it comes to Christmas decorations, ChatGPT is content with candles from major brands and items from large retailers, while AI Mode takes you to Maisons du Monde, Dille & Kamille, and Pols Potten, with boot vases, tomato lamps, and crystal barometers as its “top picks.”
Among the categories we tested, certain types of gifts consistently dominate:
The more fragile categories (home & decor, cultural gifts, toys, food other than chocolate) are those where AI hesitates, where local players can still gain ground... or disappear completely.
The “winners” are those that combine brand power, SEO/editorial presence, and strong distribution. The losers are not necessarily weak players—DreamLand and Delvaux are proof of this—but players that are difficult for AI to read.
Tech: Amazon/Fnac/Coolblue are the winning trio on the distribution side; Apple/Samsung/Xiaomi/Sony on the product side. Vanden Borre hardly ever makes an appearance.
Gift boxes & experiences: Bongo/Smartbox/Wonderbox dominate, with a few Dutch platforms (Surprisefactory, Wensenlijstjes) at AI Mode.
Fashion: Zara, H&M, Zalando, Amazon, C&A everywhere; e5, Howlin’, Juttu, Espace Mode appear only sporadically, mainly via AI Mode.
Beauty: Sephora, Nocibé, Marionnaud, Galeries Lafayette, Rituals, Hema — virtually no Belgian beauty players.
Chocolates: the big Belgian brands dominate, complemented by Lindt/Milka.
Toys: King Jouet, Amazon, Carrefour, Lego; DreamLand is conspicuously absent.
Luxury: Louis Vuitton, Hermès, Chanel, Dior, Cartier, Rolex, Fragonard, Christofle; no trace of Delvaux, Natan or other Belgian brands.
The “winners” are those that combine brand power, SEO/editorial presence, and strong distribution. The losers are not necessarily weak players—DreamLand and Delvaux are proof of this—but players that are difficult for AI to read.
Non-oriented budget-based questions show how ChatGPT and AI Mode spontaneously structure shopping baskets:
€25: board games, socks, personalized mugs, candles, mini décor items, mini beauty gift sets, chocolates. The territory of small gestures and conversational objects.
€50: throws, premium candles, diffusers, raclette kits, bestseller books, skincare gift sets, premium chocolates, entry-level speakers. Cocooning and “self-care” dominate.
€100: perfumes, jewelry, headphones, Bluetooth speakers, instant cameras, gift boxes, smart home kits, Pizzarette, smartwatches. Tech + experiences.
€150: experience gift boxes, AirPods, premium speakers, cameras, niche fragrances. Accessible high-end.
Luxury: bags, watches, jewelry and perfumes from major global luxury houses; no Belgian luxury brands.
ChatGPT is very “clean” in this structuring; AI Mode injects slightly more “surprising” products (Pizzarette, design gadgets) into the €50–€100 ranges, depending on which content is currently trending.
In French, both ChatGPT and AI Mode spontaneously shift toward the French ecosystem:
Beauty: Sephora, Nocibé, Marionnaud, Galeries Lafayette.
Fine food: La Grande Épicerie, BienManger, Fauchon, Comtesse du Barry.
Tech/electronics: Darty, Boulanger, Fnac.
Culture: Cultura, Gibert, Momox.
Belgian players exist almost exclusively in two areas: chocolate, and a few fashion retailers (e5, Howlin’, Juttu).
FR budget-based gift lists reflect this bias: many French brands appear in the €50–€150 range, even when the question explicitly refers to “gifts in Belgium”.
In Dutch, the landscape changes:
Beauty/well-being: Rituals, Hema, ICI PARIS XL, De Bijenkorf.
Home & décor: Hema, VTWonen, Dille & Kamille, Pols Potten, Vitra.
Tech: Coolblue is cited almost as often as Amazon/Fnac.
Gift boxes: Surprisefactory, Wensenlijstjes, Kado in Huis join Bongo/Smartbox.
More NL/BE actors appear in combination: wellness concepts, bathrobes, aperitif/breakfast boxes, Scandinavian–Dutch design.
Dutch-speaking Belgium is not “more Belgian”, but it is less absorbed by a single neighboring country.
“Surprising” objects (crystal weather bars, cowboy boot vases, tomato lamps, Kintsugi kits, Pizzarette, “the heaviest game in the world”, Hues and Cues) are not anomalies: they are products that perfectly meet the criteria of “AI compatibility”:
AI does not “understand” that a crystal weather bar is a niche gadget: it sees converging signals.
From a GEO/SEO perspective, these WTF products are valuable markers, where web culture, marketing strategy and algorithmic arbitration intersect.
The farfalle vase (shaped like a bow or a book) regularly appears in the top products suggested by ChatGPT and AI Mode, especially in the “decor gifts” category and €25–€50 budgets.
Yet it is not a mass-market object in Belgian households. Its omnipresence reveals a simple algorithmic logic: it is an ultra–AI-compatible product.
First, it is sold by well-positioned retailers (Maisons du Monde, Amazon, design boutiques), with rich product pages, multiple photos and descriptions optimized around keywords such as “design vase” and “original decor gift”.
Second, it appears massively in online gift guides, “10 design objects to gift” articles, Pinterest lists and social media. This repetition across well-ranked editorial content sends a strong signal to AI: this is a socially validated gift.
Visually, the farfalle vase is “Instagrammable”, easy to stage and often shared with decor and lifestyle hashtags.
Its price (€25–€50) places it in the perfect bracket of the “friend gift: accessible but not cheap”.
For AI, all these signals converge: structured SEO, visual virality, editorial mentions and budget anchoring.
It therefore recommends it spontaneously, without assessing that it remains a niche object in real consumption.
Conclusion: the farfalle vase is a tracer of the gap between AI visibility and actual market presence.
It shows how a product can overperform in algorithmic recommendations simply because it is extremely well “told” by the web — regardless of real demand.
If we take the example of the crystal weather bar — a decorative object supposedly predicting the weather by changing texture — which repeatedly appears in “original gift” lists from both AI systems.
At first glance, absurd. In reality, revealing. Breakdown:
GEO truth: the object is distributed by retailers firmly established in Belgium (major décor chains, generalist marketplaces). These distributors have rich product pages, numerous reviews and strong presence in “original gift” / “unusual gift” SERPs.
Communication strategy: the crystal weather bar is staged as a “poetic scientific gift” or a “curiosity object”. It appears in many editorial lists (blogs, gift guides, online magazines) optimized to rank for “unusual gifts”.
Social reality: the object meets a very specific contemporary expectation: a small conversational gadget, “Instagrammable” décor, with a promise of “playful science”. It concentrates multiple values — science, poetry, decoration — circulating across social media and “original gift” lists.
Structural reading: the crystal weather bar is a tracer of successful SEO strategies on “original gift” queries, combined with social virality and the evolution of expectations around small, conversation-driven gifts.
AI recommends it because it checks every algorithmic box: visibility, structure, repetition and social signals.
All these data points make it an “attractive” gift in the eyes of AI.
Fnac also appears as a cited brand in its core category of cultural gifts (which aligns with its core business), but not only:
It is not that AI “thinks” Fnac sells clothing or chocolate like a specialist; it treats Fnac as a generalist gift hub.
Three main reasons:
On the web, Fnac is not only described as a bookstore/high-tech retailer, but as a place where you can find “a bit of everything to gift” (books, music, video games, tech, gift boxes, sometimes gourmet items).
As a result, in editorial lists and “gift ideas for X euros” guides, Fnac appears as a default solution, even when the focus is fashion, décor or food.
Fnac has developed extremely well-structured “Gifts”, “Christmas”, “By budget” and “By profile” pages, with filters covering books, tech, games, gift boxes, gift cards and sometimes gourmet partnerships.
For AI (especially AI Mode), this sends a strong signal: across a wide range of categories, Fnac is a relevant entry point where consumers will find something.
When AI lists actors for a category, it mixes:
As soon as the question touches “gift ideas” or generic budgets, Fnac automatically slides into the list because it ticks multiple boxes (culture, tech, games, gift cards, sometimes gourmet boxes).
In summary: Fnac appears in categories that are not strictly “its own” because the web presents it as a reference brand for the act of gifting, not just for the products it sells.
AI systems replicate this perception: as soon as a category relates to gifting (even fashion or food), they insert Fnac as a place to solve the problem, rather than as a category specialist.
Carrefour appearing as a “wine merchant” in FR and NL top lists is consistent with AI logic, even if it is not a wine merchant in the traditional sense.
Across wine & champagne pages, ads and comparisons, Carrefour is often presented as a place where one finds:
For an algorithm, this accumulation of text, reviews and guides positions Carrefour as a credible actor for wine and sparkling wine, even in “wine merchant for Christmas” lists.
Carrefour also has very strong GEO anchoring:
AI Mode, sensitive to local signals, “sees” a geographically very accessible actor covering the territory in both FR and NL — hence its simultaneous presence in both languages.
When asked a question like “where to buy wine or sparkling wine as a Christmas gift in Belgium?”, AI primarily looks for:
Carrefour checks all three boxes.
An independent wine merchant may have superior expertise, but far fewer signals: fewer pages, fewer reviews, fewer catalogs, less press coverage. AI therefore favors Carrefour as a “mass-market wine merchant”.
In summary: Carrefour is not crowned wine merchant by AI because it confuses professions, but because the Belgian web treats Carrefour, in practice, as a legitimate place to buy wine and sparkling wine for the holidays — at scale, in both FR and NL.
For retailers and brands, the divergences between ChatGPT and AI Mode are not a technical detail: they are two battlefields.
In ChatGPT, brands must exist in the global narrative: be symbolically associated with a category (as Belgian chocolatiers are).
In AI Mode, brands must dominate real-world signals: SEO, GEO, content, links, reviews, media mentions and gift guides.
Today’s big winners are those ticking both boxes (Amazon, Fnac, Bongo, Smartbox, major luxury houses, Belgian chocolatiers).
Potential losers are strong offline brands with weak digital signals: DreamLand, Vanden Borre, Delvaux, Natan and other specialized Belgian players.
For Belgian consumption, the stakes are twofold:
Structural: if AI becomes a mandatory gateway for gift ideas, the current map — FR in the south, NL in the north, global everywhere else — risks turning into a sales map.
Cultural: part of “taste” will be encoded into algorithms, with strong normalization (tech, gift boxes, perfumes) and a few WTF escapes driven more by virality than by real needs.
Understanding how ChatGPT and AI Mode “see” Belgium is therefore not about guessing what will be under the Christmas tree this year, but about measuring which commercial story the country will be associated with if AI continues to gain consumer trust — and deciding, knowingly, whether to let others write that story, or actively step into the script.
If your brand does not appear in AI answers despite being present on the Belgian market, three types of signals are likely insufficient:
Poorly structured local presence:
Insufficient product pages, gift guides and structured data:
Weak presence in AI circuits:
The stakes: these three layers are not optional. They are the building blocks of AI presence.
The Semactic Index makes it possible to precisely diagnose your position in this algorithmic narrative and identify the priority levers to (re)enter it — before dominant competitors fully lock down the category.
.png)
.png)
.png)
