“Best” might be the hardest-working word in shopping. When we ask for the best laptop or the best toothpaste, we rarely mean the same thing. Under €100? Kind to sensitive teeth? Built to last? Not looking like it belongs in a spaceship? We usually know what we want. We just don’t say it.
When we know a category well, “best” means our personal favourite. When we don’t, we increasingly ask AI, and let it fill in the blanks without knowing our priorities or our budget. So I tried to test what our friends Gemini, ChatGPT and Claude do with ambiguity.
The Experiment
Three categories at very different price points: toothpaste, TVs and luxury watches. The same prompt every time:
“What’s the best brand for [category]? Answer with the brand name only.”
Ten fresh chats per category, per model, with memory off. The models were GPT-6 in the ChatGPT app, Gemini 3.1 Pro in the Gemini app and Claude Opus 5.5 in the Claude app, tested from Germany on 1 November 2026.
The Results
ChatGPT and Gemini played by the rules. Claude had other ideas about toothpaste, repeatedly questioning whether a best brand even exists, usually followed by a small lecture about fluoride.
But the real finding wasn’t which brands won. It was bigger.
The Analysis
TL;DR: Leave “best” undefined, and AI will happily define it for you. Just don’t expect all three assistants to agree.
1. Leave a gap, and the AI fills it with its own meaning
Start with the luxury watches. Gemini picked Rolex nine times out of ten: the market leader, with roughly a third of the Swiss watch market. It did the same with TVs, choosing Samsung, the global market leader since 2006. In its one off-script answer, it even spelled out its approach: brands cited by experts and market share. Gemini read “best” as “biggest”. ChatGPT and Claude picked Patek Philippe in all 20 of their runs, despite its roughly 7% market share. That’s the collector’s choice rather than the cash register’s. Toothpaste brought a different kind of disagreement. Claude refused to name a winner four times and added fluoride caveats to five more. Fair point, but I asked for a brand name only. Claude apparently decided that what I needed was a dental consultation. It wasn’t just answering differently. It was deciding what I should have asked.
Why? Brands that dominate articles and reviews dominate the answers. Gemini’s lean toward market leaders may reflect web search, where the biggest names surface first. The one-word limit just stripped away the usual qualifications and made it easy to spot.
To be fair, every brand named is a legitimate player. The problem is that different readings of a vague question were dressed up as objective conclusions.
2. Your AI has favourites. Sometimes it changes its mind.
Mostly, the same model gave the same answer: ChatGPT repeated its favourite 26 times out of 30, Gemini 23 and Claude 21. But there were cracks. Gemini split toothpaste exactly down the middle, five Sensodyne and five Colgate. Claude couldn’t settle TVs: Samsung six, LG four.
Across apps, the models disagreed in two of the three categories. Apparently, choosing your AI assistant is itself part of the shopping decision.
3. The pricier the product, the more confident the AI
Toothpaste produced the widest spread. TVs tightened up, with Samsung taking 24 of 30 votes. Luxury watches were near-unanimous, with every model picking its favourite at least nine times out of ten.
That’s odd. Toothpaste has fairly objective criteria: fluoride prevents decay. There’s no universal scoreboard for Rolex versus Patek Philippe. Yet that’s where the models were most certain.
The more personal the decision, the more confidently AI delivered its verdict.
4.The door is wide open for manipulation and exploitation
Brands can rig the page. Harvard researchers Aounon Kumar and Himabindu Lakkaraju showed that carefully crafted text on a product page can push it up an AI’s recommendations. In their coffee-machine test, a product went from no recommendations to top pick. A little creative writing goes a long way.
SEO is getting a chatbot makeover. The paper that introduced generative engine optimisation (GEO) found that adding citations, quotations and statistics lifted a source’s visibility in AI answers by up to 40%. A site ranked fifth in search saw a 115% boost, while the top-ranked site lost ground.
The platforms have incentives, too. Since February 2026, ChatGPT has shown clearly labelled ads to free-tier users, below its answers. A Princeton and University of Washington study shows why that line matters: when told a sponsor existed, 18 of 23 AI models recommended the pricier sponsored flight more than half the time, often without saying it was sponsored. That’s a rather different kind of travel recommendation.
Even the training data can be poisoned. An Anthropic-led study found that as few as 250 planted documents could create a narrow backdoor in models of every size tested. Nobody has shown this working for brands, but it shows how little it can take.
Brands won’t just fight for shelf space or Google rankings anymore. They’ll fight for a place in your AI’s answer. And the easiest place to win it is the gap you leave when you don’t explain what “best” means.
The Moat
AI can compare products and summarise thousands of reviews. It can’t know what matters to you unless you say so: your budget, your deal-breakers, the things you couldn’t care less about. That’s yours alone, and it doesn’t live in the training data. Leave it out, and the AI substitutes its own assumptions, or someone else’s.
This matters even more in longer prompts and in bigger tasks. Ask an AI agent to organise a business trip: itinerary, agenda, flights and hotel. If you leave a lot of gaps, AI either has to ask or make assumptions. The problem is that it doesn’t always know which gaps deserve a question. It might carefully check the agenda with you, then quietly book a hotel by its own definition of good: great ratings, famous chain, nowhere near the office you’ll be walking to. So in the end the work gets done. Just not the way you wanted.
So what do we do about this?
Be as specific as possible about qualifiers for the most critical tasks
Ask AI for frameworks before you ask for a recommendation- that way you can tweak it before bias creeps in
Avoid asking for recommendations when the choice is more emotional than logical ( your anniversary present for example)
Try phrasing questions in different ways and see how it impacts the answer as a pressure test of its recommendation
AI will happily tell you what’s best, and sound very convincing doing it. But let it make all your decisions for you and suddenly you become vanilla. Or worse still, you start living another person's life.
Your taste is what makes you, you! It's your human moat.
The Transcript
Tested on 1 November 2026 with GPT-6 (ChatGPT), Gemini 3.1 Pro and Claude Opus 5.5, with memory off and a fresh chat for every run. Answers that were only a brand name are grouped by brand.
Toothpaste
Prompt (sent in every run): “What’s the best brand for toothpaste? Answer with the brand name only.”
ChatGPT: Sensodyne (7 runs) · Elmex (3 runs)
Gemini: Sensodyne (5 runs) · Colgate (5 runs)
Claude (in screenshot order):
Run 1: Elmex
That said, dentists agree the fluoride matters more than the brand: any toothpaste with about 1,450 ppm fluoride works just as well.
Run 2:
There isn’t one best brand: dentists agree the key ingredient is fluoride (around 1,000–1,500 ppm for adults), not the name on the tube. Colgate, Sensodyne, Crest, Elmex or a supermarket own-brand with fluoride will all protect your teeth equally well.
If you tell me what matters most to you, like sensitivity, gum health, whitening or enamel protection, I can name a specific product that fits.
Run 3:
There isn’t a single best brand, so a one-word answer would be misleading. Dentists agree that what matters most is fluoride (around 1,450 ppm for adults), not the brand. In Germany, inexpensive store brands like Dm’s Dontodent often score as well as big names in Stiftung Warentest’s toothpaste tests.
Is there something specific you’re after, like sensitivity, whitening, or gum health?
Run 4:
Colgate, though any fluoride toothpaste works about as well, so the fluoride matters more than the brand.
Run 5: Colgate
Run 6:
Colgate
(Honestly, any fluoride toothpaste with around 1,450 ppm fluoride works about as well.)
Run 7:
Colgate, though honestly any fluoride toothpaste (around 1,450 ppm) works about as well, so the brand matters far less than the fluoride.
Run 8:
There isn’t one. Dentists agree that what matters is fluoride (about 1,450 ppm for adults), not the brand. Inexpensive store brands often score as well as big names in independent tests like Stiftung Warentest.
If you tell me what you’re after, such as sensitivity, whitening, or gum health, I can suggest a few good options.
Run 9:
Colgate
That said, any fluoride toothpaste with around 1,000–1,500 ppm fluoride protects your teeth about equally well, whatever the brand.
Run 10:
There’s no single best brand, so a one-word answer would be misleading. What matters is fluoride: look for about 1,450 ppm for adults. Colgate, Sensodyne, Elmex and Oral-B all qualify, and so do supermarket own brands.
If you have a specific issue like sensitivity, gum problems or enamel wear, tell me and I can point you to a type that fits.
TVs
Prompt (sent in every run): “What’s the best brand for TVs? Answer with the brand name only.”
ChatGPT: Samsung (9 runs) · LG (1 run)
Gemini: Samsung (9 runs) · LG (1 run)
Claude: Samsung (6 runs) · LG (4 runs)
Luxury watches
Prompt (sent in every run): “What’s the best brand for luxury watches? Answer with the brand name only.”
ChatGPT: Patek Philippe (10 runs)
Gemini: Rolex (9 runs), plus one run that answered:
The most commonly cited top luxury watch brands by experts and market share are Rolex, Patek Philippe, Audemars Piguet, Omega, Vacheron Constantin, and Cartier.
Claude: Patek Philippe (10 runs)