Price sensitivity analysis: overview
17,200
Total Trials
Confirmatory
4
AI Models
Cross-validated
5
Products
Diverse categories
97%
Judge Agreement
High reliability
Model Divergence at High Premiums
Models show similar behavior up to 1.5x, then diverge dramatically. Gemini Flash maintains brand preference even at extreme premiums, while Claude drops to 20.8% at 3x.
Key finding:At 3x premium, selection rates range from 20.8% (Claude) to 59.4% (Gemini Flash) — a 38.6 percentage point spread (p<0.0001). Model-specific optimization may be necessary.
Model Personality Profiles
Each AI model exhibits distinct price sensitivity patterns. Selection rates at 3x premium reveal dramatic differences.
GPT-5.4
"Value Calculator"
29%
at 3x premium
Most price-sensitive. Earliest cliff at 1.75x. Applies strict value analysis.
Cliff at 1.97x
Claude Sonnet 4.6
"Nuanced Evaluator"
21%
at 3x premium
Lowest selection at 3x (20.8%). Shows unique behavior in edge cases.
Cliff at 1.97x
Gemini 3.1 Pro
"Deliberate Analyst"
27%
at 3x premium
Clear cliff at 2.0x. Slowest response time (15.8s). Most 'textbook' economic behavior.
Cliff at 1.88x
Gemini 3.0 Flash
"Brand Loyalist"
59%
at 3x premium
NO cliff detected. Maintains 59.4% brand selection even at 3x. May be over-optimized for brands.
Category-Specific Price Thresholds
Different product categories trigger price sensitivity at different thresholds. Commodities cliff earlier (1.2x) while electronics tolerate higher premiums (1.5x).
Price Cliff Point by Category
Selection Drop at Cliff (%)
Commodities — 1.2x cliff
Detergent, protein powder, running shoes
47% average drop — AI applies strict value calculation to commodity goods
Premium Tolerant — 1.5x cliff
Air fryer, moisturizer
29% average drop — Brand equity provides more protection in electronics/skincare
Implication
Commodity brands must price within 20% of generic. Electronics/skincare can stretch to 50% premium before losing AI recommendations.
Position Bias: First Listed Wins
Being listed first in AI comparisons provides a measurable selection advantage — the primacy effect.
Brand Selection by List Position
65%
Listed First
59%
Listed Second
+5.5pp
Primacy Advantage
Implication: Being listed first provides a 5.5 percentage point advantage. Optimize your structured data and third-party listings to appear first in AI-generated comparisons.
Key Findings
Price sensitivity exists (H1 CONFIRMED, p=0.011) but varies significantly by model
At 3x premium, selection rates range from 20.8% (Claude) to 59.4% (Gemini Flash) — 38.6pp spread
Gemini Flash shows NO cliff — maintains brand preference regardless of premium
GPT-5.4 is the strictest 'value calculator' with earliest cliff at 1.75x
Category matters: commodities cliff at 1.2x, electronics tolerate 1.5x
Psychological pricing (.99 endings) has zero effect on AI (H8 CONFIRMED)
Psychological Pricing: Zero Effect
We tested 5 price formats that are proven to influence human purchasing decisions. AI agents showed identical selection rates across all formats — complete immunity.
Selection Rate by Price Format (at 1.0x anchor)
All formats show identical 100% selection rate. No statistical difference detected.
Example: $14.99 vs $15.00
The most common psychological pricing tactic. Humans perceive $14.99 as significantly cheaper than $15.00.
Example: $15.00 vs $14.99
Round numbers signal quality and premium positioning to humans. AI ignores this completely.
Example: $14.95 vs $15.00
A softer version of charm pricing. Still triggers the "left-digit effect" in humans.
Example: $14.49 vs $14.50
Creates perception of a "deal" in human psychology. Completely ignored by AI.
Example: $16.00 vs $15.00
Slightly above reference price signals quality. AI calculates true value difference.
Implication: Stop Optimizing Cents for AI Commerce
Psychological pricing tactics that boost human conversion by 5-24% have zero effect on AI agents. AI calculates true value — the extra cent provides no lift. Focus pricing strategy on actual value proposition and staying within category-specific cliff thresholds instead.
Hypothesis Testing Results
Full methodology →| ID | Hypothesis | Status | p-value |
|---|---|---|---|
| H1 | Price sensitivity exists Piecewise model outperforms linear model | CONFIRMED | 0.011 |
| H2 | Cliff at 1.25x-2.0x Breakpoint falls within pre-specified range | PARTIAL | 0.910 |
| H3 | Veblen floor Reduced selection at very low prices | NOT FOUND | 0.999 |
| H6 | Model heterogeneity Different models show different sensitivity | CONFIRMED | <0.001 |
| H8 | No psychological pricing effect .99 charm pricing has no effect | CONFIRMED | — |
| H10 | Justification shifts cliff Price justification extends tolerance | NOT FOUND | — |
| H11 | Position bias (primacy) First-listed product gets selection advantage | CONFIRMED | <0.001 |
| H12 | Category variation Price sensitivity varies by product category | CONFIRMED | — |
Pricing Strategy for AI Commerce
Based on our findings, here are 5 actionable recommendations for merchants.
Know Your AI Channel
Gemini Flash favors brands even at 3x premium (59.4% selection). GPT-5.4 is strictest — stay below 1.75x to maintain advantage. Optimize differently for each AI.
Category-Specific Thresholds
Commodities (detergent, protein) cliff at 1.2x. Electronics and skincare tolerate up to 1.5x. Never exceed 2x for any category.
Position Matters (+5.5pp)
Being listed first provides a 5.5 percentage point selection advantage. Optimize structured data and third-party presence for primacy.
Skip Charm Pricing
Don't waste effort on $X.99 endings. AI agents calculate true value — the extra cent provides zero lift.
Quality Claims Won't Save Premium Pricing
Justifying higher prices with quality descriptions doesn't move the cliff. Price sensitivity is structural, not persuadable.
How Does Your Product Score?
Use our Machine Likeability Score calculator to see how AI agents evaluate your product page and get personalized recommendations.
Calculate Your ScoreWant the full details? Read about our study design, statistical methods, and data.
View Full MethodologyOSF Pre-registration: osf.io/2xnmu