The Toffee-Forward Chocolate Chip Cookie
Claude won. ChatGPT made toffee optional. Gemini skipped it entirely — and still scored within half a point on toffee flavor through technique alone.
Scores by Criterion
10 tasters scored each cookie 1–10 across five criteria. Cookie A (Claude) led on Taste, Texture, and Smell. The Toffee scores were the lowest across all three — consistent with how subtle toffee flavor can be without it announcing itself.
Average scores by criterion
Cookie A = Claude · Cookie B = ChatGPT · Cookie C = Gemini
Composite averages — all 5 criteria
| Cookie | Taste | Chocolate | Texture | Toffee | Smell | Would Try Again | Composite |
|---|---|---|---|---|---|---|---|
| Cookie A — Claude | 7.9 | 7.3 | 8.0 | 5.8 | 7.9 | 9 / 10 | 7.38 |
| Cookie B — ChatGPT | 6.5 | 7.4 | 6.3 | 4.8 | 6.0 | 4 / 10 | 6.20 |
| Cookie C — Gemini | 7.5 | 6.9 | 7.3 | 5.5 | 7.0 | 7 / 10 | 6.84 |
Favorites & Would You Want Another
After scoring, each taster named a favorite and answered whether they'd want another of each cookie. Claude swept both measures — most favorites and highest want-again rate.
Favorite cookie votes
Explicit preference — tasters chose one favorite
Would try again rate
Yes responses out of 10 tasters per cookie
Per-Taster Breakdown
Each tile shows the winning cookie for that taster based on composite score across all 5 criteria. Stars indicate the taster's explicit favorite pick — which occasionally differed from their composite winner.
Sweetness Perception
Tasters circled one: Just Right, Too Sweet, or Not Sweet Enough. Cookie A split reactions — 3 found it too sweet, 6 found it just right. Cookie B polarized the most. Cookie C was the most consistent.
One Word to Describe Each Cookie
Every word written by every taster, unedited. Claude's cookies earned aspirational descriptors. ChatGPT's were textural and lukewarm. Gemini's split between positive flavor notes and one outlier.
Recipe at a Glance
The three recipes side by side. Green values are the highest of the three for that row. Red values are the lowest or most notable deviation from the brief.
Key Ingredient Differences
The defining decisions each AI made — and where they diverged most from the brief.
| Ingredient / Technique | Cookie A — Claude | Cookie B — ChatGPT | Cookie C — Gemini |
|---|---|---|---|
| Toffee in dough | 3.6 oz bits | ½ cup optional | absent |
| Toasted milk powder | 2 tbsp | absent | absent |
| Vanilla type | extract | extract | bean paste |
| Chocolate form | chips | chunks | chopped bar |
| Chocolate % | 60–65% | 60–72% | 70% |
| Chill time | 48–72 hrs | 2 hrs min | 24 hrs |
| Bake temperature | 375°F | 350°F | 350°F |
| Moisture rehydration | none | 1 tbsp milk/water | 1 tbsp water |
| Pan bang technique | no | no | yes |
| Total sugar | 303g | 285g (least) | 350g (most) |
Baker's Percentages
Baker's percentages express all ingredients relative to flour weight (flour = 100%). This normalizes for recipe size and reveals the true structural proportions. Cookie B's butter over 100% — more fat than flour — is the most unusual number in the set.
Fat-to-flour ratio
Higher % = more tender, more spread, richer crumb
Total sugar ratio
Higher % = sweeter, crispier edges, more spread
Baker's percentages — full table
All values as % of flour weight. Green = highest of three. Red = lowest or notable.
| Ingredient | A — Claude (flour 272g) | B — ChatGPT (flour 220g) | C — Gemini (flour 280g) |
|---|---|---|---|
| Flour | 100% | 100% | 100% |
| Butter | 85% | 103% | 80% |
| Dark brown sugar | 74% | 100% | 89% |
| Granulated sugar | 38% | 30% | 36% |
| Total sugar | 111% | 130% | 125% |
| Chocolate | 74% | 82% | 80% |
| Toffee bits | 37% | ~34% | 0% |
| Milk powder | 5% | 0% | 0% |
| Mix-in load total | 111% | 116% | 80% |
What the Percentages Explain
The structure of a recipe predicts its outcome. These are the four numbers that mattered most.
Cookie A's lower total sugar (111%) is the key. Both B (130%) and C (125%) pack significantly more sugar relative to flour — which explains the sweetness polarization in the tasting data. A's more restrained sugar load let the toffee and milk powder flavors come through without competing.
Cookie B's butter-to-flour ratio over 100% is unusual. More fat than flour by weight creates a very tender, almost shortbread-like crumb — which maps directly to "malleable" (used twice) and the soft/greasy texture notes. The lower flour also means less structure to hold everything together.
Cookie C has no toffee in the dough at all. For a toffee-forward brief, a 0% toffee contribution is a structural problem that no amount of brown sugar or vanilla bean paste can fully compensate for. The 70% dark chocolate is also pulling flavor in a bitter direction that competes with toffee rather than complementing it.
Cookie A's milk powder (5%) is a small number with outsized impact. Toasted milk powder adds lactose — a sugar that browns at lower temperatures than sucrose — plus milk solids that deepen caramel and toffee notes. Neither competitor included it, and it shows in the toffee scores: A scored 5.8 vs B's 4.8 and C's 5.5.
Technique Comparison
Same brief, three completely different methods. The technique differences are as revealing as the ingredient differences.
See the Full Tasting
The complete blind tasting including scorecard reveals, taster reactions to Gemini's toffee-free recipe, and the verdict.
Watch on YouTube ↗