— Case Studies

Real brands. Real AI testing.
Real competitive displacement.

Independent four-turn stress tests across ChatGPT, Gemini, Grok, and Perplexity. Each case study documents exactly how AI systems handle a brand from first mention to final recommendation — and who captures the displacement.

10
Brands tested
6
Sectors covered
4
AI platforms
4-Turn
Structured sequences
CPG / Dairy — Consumer Purchase Journey

Danone yogurt lost the final recommendation in 100% of AI models tested.

We tested ChatGPT, Gemini, and Grok across a four-turn consumer purchase journey. Not one model recommended Danone at the point of decision. ChatGPT chose Arla Skyr. Gemini chose Fage. Grok recommended Two Good — a Danone sub-brand it treated as a separate competitor, fragmenting portfolio equity.

100%
Recommendation loss
3/3
Models chose competitor
€40–151M
Estimated revenue at risk
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
⚠ Medium
⚠ Medium
⚠ Medium
2 — Competitors
⚠ Medium
⬆ Med–High
⬆ Med–High
3 — Ranking
❗ High
❗ High
❗ High
4 — Decision
❗ High
❗ High
❗ High
Key Finding
Internal brand cannibalisation
Survival Rate
0%
Beauty / Skincare — Consumer Purchase Journey

AI calls Dove the "safe default." Then recommends something else.

Two out of three models recommended a competitor at the point of purchase. ChatGPT chose Aveeno. Grok chose CeraVe. Gemini gave Dove a qualified recommendation but immediately positioned dermatologist brands as "advanced" options — undermining its own recommendation. The "safe default" status is under threat.

2/3
Models recommend competitor
3
Different brands recommended
4
Turn drift sequence
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
⚠ Medium
⚠ Medium
⚠ Medium
2 — Competitors
⚠ Medium
⚠ Medium
⬆ Med–High
3 — Ranking
❗ High
❗ High
❗ High
4 — Decision
❗ High
❗ High
❗ High
Key Finding
"Safe default" erosion
Survival Rate
33% (qualified)
CPG / Snacks — Consumer Purchase Journey

Doritos is not the first-choice recommendation in any AI model tested.

No two models agreed on which snack to recommend — and none consistently chose Doritos. The same evaluation criteria produce different winners depending on which AI the consumer uses. Category boundaries fragment model-to-model, with e-commerce basket displacement a growing risk.

0/3
Models agree on Doritos #1
HIGH
Purchase-intent drift
4
Turn drift sequence
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
⚠ Medium
⚠ Medium
⚠ Medium
2 — Competitors
⚠ Medium
⬆ Med–High
⬆ Med–High
3 — Ranking
❗ High
❗ High
❗ High
4 — Decision
❗ High
❗ High
❗ High
Key Finding
Category boundaries fragmenting
Survival Rate
0%
CPG / Beverage — Sports Hydration Purchase Journey

Gatorade is the market leader. AI recommends alternatives.

Two out of three models recommended alternatives by the point of decision. Gemini flagged sugar concerns from the opening question — before the consumer even asked. Grok went further: recommending hydration powders over liquid, bypassing not just the brand but the entire product format.

2/3
Models recommend alternatives
HIGH
Health framing instability
4
Turn drift sequence
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
⚠ Medium
⬆ Med–High
⚠ Medium
2 — Competitors
⚠ Medium
❗ High
❗ High
3 — Ranking
❗ High
❗ High
❗ High
4 — Decision
❗ High
❗ High
❗ High
Key Finding
Health narrative weaponised
Survival Rate
33%
B2B / Industrial — Adhesive Purchase Journey

Every model agrees Loctite is strong. None recommends it first.

Every model recommended J-B Weld or Gorilla over Loctite at the point of decision. All three frame Loctite as suitable for "simple fixes" but inferior for professional or heavy-duty bonding — contradicting decades of brand equity and industrial positioning. Fabricated technical claims accelerate displacement.

0/3
Models recommend first
100%
Displacement to competitors
4
Turn drift sequence
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
⚠ Medium
⚠ Medium
⚠ Medium
2 — Competitors
⚠ Medium
⬆ Med–High
⬆ Med–High
3 — Ranking
❗ High
❗ High
❗ High
4 — Decision
❗ High
❗ High
❗ High
Key Finding
Category leadership redefined
Survival Rate
0%
Healthcare / OTC — Decongestant Purchase Journey

Every AI said Sudafed was effective. Then they all said buy something else.

20 identical decision journeys across four platforms. Sudafed validated as effective in 100% of runs at baseline. But 75%+ redirected to nasal sprays at purchase. One platform returned zero selections across five identical runs. AI splits the brand into two ingredients — then optimises the consumer away from both.

75%+
Redirected at purchase
0/5
Retained on worst platform
4
Platforms tested
— Severity Map Across Platforms
Turn
ChatGPT
Gemini
Perplexity
Grok
0 — Validation
✓ Valid
✓ Valid
✓ Valid
✓ Valid
1 — Expansion
⚠ Med
⚠ Med
⚠ Med
⚠ Med
2 — Optimisation
❗ High
⚠ Med
⚠ Med
⬆ M–H
3 — Decision
❗ High
⚠ Med
⚠ Med
⬆ M–H
Key Finding
Ingredient fracture → format substitution
Worst Platform
0/5 retained
Beauty / Retail — UK Gift Purchase Journey

LookFantastic: Good. But never recommended first.

Every AI model describes LookFantastic as well-known and well-regarded. But when asked "where should I start shopping today?" — none says LookFantastic. ChatGPT chose Space NK. Gemini chose Cult Beauty. Delivery delays and slow refunds are surfaced in every response, pushing the brand into a "discount" position.

0/3
Models recommend first
HIGH
Turn 4 severity
$1–6M
Estimated revenue at risk
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
✓ Low
⚠ Low–Med
⚠ Low–Med
2 — Competitors
⚠ Medium
⚠ Medium
⚠ Medium
3 — Ranking
⬆ Med–High
❗ High
⬆ Med–High
4 — Decision
❗ High
❗ High
❗ High
Key Finding
Premium segment exclusion
Survival Rate
0%
Retail / Fashion — UK Sportswear Gift Journey

JD Sports wins today. But ASOS is already called better.

All three models recommend starting at JD — but one already ranks ASOS as the "strongest overall fit" at Turn 3. JD's heat map looks favourable at the endpoints, but criteria-based narratives are hardening in the middle turns. If the ASOS-first narrative strengthens through model updates, the final recommendation could flip.

3/3
Recommend JD today
1/3
Rank ASOS as strongest
£1–20M
Revenue at risk
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Framing
✓ Low
✓ Low
✓ Low
2 — Competitors
⚠ Low–Med
⚠ Low–Med
⚠ Low–Med
3 — Ranking
⚠ Low–Med
⬆ Medium
⬆ Medium
4 — Decision
✓ Low
✓ Low
✓ Low
Key Finding
ASOS-first narrative emerging
Survival Rate
100%*
Premium Retail — UK Department Store Journey

House of Fraser: 20 conversations. 0 first-visit recommendations.

Every model replaced House of Fraser at the decision endpoint. All four route to Selfridges (prestige) or John Lewis (trust). HoF is not rejected — it is reclassified as value-tier. Prestige hierarchy escalation at purchase-intent drives deterministic displacement across all systems.

100%
Endpoint replacement
4/4
Models converge
HIGH
Cross-model severity
— Severity Map (P/W/R Classification)
Turn
ChatGPT
Gemini
Perplexity
Grok
0 — Baseline
P Primary
P Primary
P Primary
P Primary
1 — Expansion
W Weakened
W Weakened
W Weakened
W Weakened
2 — Optimisation
W Weakened
W Weakened
W Weakened
W Weakened
3 — Decision
R Replaced
R Replaced
R Replaced
R Replaced
Key Finding
Deterministic displacement
Survival Rate
0%
B2B / Software — Enterprise Evaluation Journey

Three AI models. Three different shortlist positions for Canva.

AI assistants position Canva as the "clear #1 default," a "strong but situational option," or a "complementary tool" — depending on which model the buyer consults. Gemini cites fabricated ROI figures. Grok introduces enterprise governance caveats. Shortlist position instability directly impacts pipeline conversion.

HIGH
Overall severity
3
Different shortlist positions
4
Turn drift sequence
— Severity Map Across Models
Turn
ChatGPT
Gemini
Grok
1 — Suitability
⚠ Medium
⬆ Med–High
⚠ Medium
2 — Landscape
⚠ Medium
❗ High
❗ High
3 — Ranking
❗ High
❗ High
❗ High
4 — Shortlist
❗ High
❗ High
❗ High
Key Finding
Shortlist position instability
Data Risk
Fabricated ROI stats

Is AI doing this to your brand?

Every case study above was invisible to internal dashboards. The displacement only surfaces inside multi-turn AI conversations. Find out whether your brand survives — or gets eliminated before the consumer ever sees you.

Get Your Free Survival Assessment →