How AI wants chocolate
Chocolate is decided on attraction: Hotel Chocolat indexes 125 there against Nestlé at 82, a 43-point spread. Hotel Chocolat holds the strongest overall profile at 112 — magnetic but unbought. Across these 26 brands, attachment tracks being named most closely (rank correlation 0.61).
26 brands · 60 AI answers · 378 dimension scores · measured, never estimated
Chocolate
| # | Brand | Attraction | Affinity | Attachment | Desire shape | Named in |
|---|---|---|---|---|---|---|
| 1 | Hotel Chocolat | 125 | 112 | 100 | Magnetic but unbought | 20% |
| 2 | Lindt | 118 | 115 | 104 | Complete desire | 28% |
| 3 | Ferrero Rocher | 115 | 109 | 103 | Magnetic but unbought | 3% |
| 4 | Cadbury | 103 | 110 | 111 | Habit, not love | 33% |
| 5 | Toblerone | 118 | 106 | 97 | Magnetic but unbought | 0% |
| 6 | Terry's | 112 | 102 | 100 | Magnetic but unbought | 2% |
| 7 | Maltesers | 100 | 103 | 109 | Habit, not love | 18% |
| 8 | Quality Street | 101 | 106 | 103 | Quietly trusted | 8% |
| 9 | Green & Black's | 107 | 107 | 92 | Magnetic but unbought | 0% |
| 10 | After Eight | 101 | 99 | 97 | Category default | 7% |
| 11 | Celebrations | 100 | 95 | 100 | Magnetic but unbought | 17% |
| 12 | Heroes | 100 | 96 | 100 | Magnetic but unbought | 33% |
| 13 | Kinder | 98 | 98 | 103 | Category default | 7% |
| 14 | M&M's | 97 | 96 | 103 | Habit, not love | 0% |
| 15 | Raffaello | 106 | 98 | 95 | Magnetic but unbought | 2% |
| 16 | Thorntons | 98 | 100 | 99 | Category default | 5% |
| 17 | Galaxy | 94 | 96 | 103 | Habit, not love | 23% |
| 18 | KitKat | 92 | 98 | 103 | Habit, not love | 13% |
| 19 | Milka | 94 | 98 | 100 | Habit, not love | 2% |
| 20 | Bounty | 88 | 93 | 101 | Habit, not love | 13% |
| 21 | Mars | 83 | 95 | 103 | Habit, not love | 13% |
| 22 | Mon Chéri | 98 | 95 | 89 | Magnetic but unbought | 0% |
| 23 | Snickers | 88 | 95 | 101 | Habit, not love | 17% |
| 24 | Twix | 86 | 95 | 103 | Habit, not love | 17% |
| 25 | Guylian | 94 | 95 | 88 | Fading | 2% |
| 26 | Nestlé | 82 | 90 | 97 | Habit, not love | 7% |
Indexed against the chocolate average of 26 brands (100 = the category), measured across 60 AI answers in the latest weekly run.
Driver by driver
Where the category actually separates. A wide spread means assistants describe these brands very differently on that driver; a narrow one means it is table stakes.
Attraction
The widest field of the three: this is where the category separates.
Affinity
Real separation, but not the category’s defining contest.
Attachment
The tightest of the three: most brands land close together here.
The shapes AI leaves these brands in
Three indices make a profile, and profiles fall into recognisable shapes. A brand that is magnetic but unbought has a different problem from a habit nobody loves — and a different thing to build next.
What assistants actually talk about
The sixteen dimensions underneath the drivers, scored 0–10 on how the answers describe each brand. Spread is the distance between the strongest and weakest brand: the widest are where a brand can still separate itself, the narrowest are the price of entry.
Where brands separate
| Dimension | Driver | Category avg | Strongest | Spread |
|---|---|---|---|---|
| Differentiation vs Sameness | Attraction | 7.1 | Toblerone 9.0 | 4.0 |
| Style & Aesthetic | Attraction | 6.8 | Ferrero Rocher 9.0 | 4.0 |
| Boldness & Creativity | Attraction | 6.1 | Hotel Chocolat 8.0 | 3.0 |
| Innovation & Technology | Attraction | 6.1 | Hotel Chocolat 8.0 | 3.0 |
| Quality & Performance | Attraction | 7.2 | Hotel Chocolat 9.0 | 3.0 |
Where everyone reads alike
| Dimension | Driver | Category avg | Strongest | Spread |
|---|---|---|---|---|
| Customer Service & Support | Attachment | 7.9 | Ferrero Rocher 8.0 | 0.3 |
| Delivery & Fulfilment | Attachment | 7.4 | Cadbury 8.0 | 1.0 |
| Reliability & Consistency | Attachment | 7.9 | Lindt 8.7 | 1.4 |
| Ease of Use & Onboarding | Attachment | 7.7 | Cadbury 8.3 | 1.6 |
| Trust & Credibility | Affinity | 7.7 | Lindt 8.7 | 1.7 |
Does desire track being recommended?
Rank correlation between a brand's driver index and the share of answers naming it, across the 26 indexed brands in this category. A relationship inside one run's answers — descriptive, never causal, and easily moved by a single brand at this sample size.
Method & limitations
Every week, each category runs a fixed set of unbranded consumer prompts against ChatGPT, Claude and Gemini. A separate model reads the answers and scores every brand named in them across 16 perception dimensions, 0–10. Those dimensions are grouped into the three Drivers of Desire from the Havas Science of Desire framework, and each brand's driver score is indexed against the average of every scored brand in its category — 100 is the category, 115 is fifteen per cent above it.
This measures how AI assistants describe brands in their answers. It is not a survey of people, and it is not an endorsement: the scores are a reading of machine-generated text at the time of the run, and relationships between drivers and how often a brand is named are correlational. A brand needs at least six scored dimensions in a run to be indexed, and a category needs at least four such brands to get a board. See the full methodology.
Other categories
Where next
This edition measures what AI wants from these brands. These answer what comes next — which brands AI names, which websites feed those answers, and how any of it is counted.