AI Buyer Behavior and Structural Bias
A substantial body of controlled research has tested how frontier AI models behave when placed in the role of autonomous buyers — making product selections, comparing options, and evaluating landing page content. The consistent finding is that AI buyers are economically rational in aggregate but exhibit systematic, non-human biases in how they process information.
Position effects are structural, not visual
In controlled e-commerce experiments, AI agents exhibit strong position bias: items placed in certain structural positions within a product listing receive disproportionate selection probability. This bias persists even when product information is presented as plain text with no images or layout. It is not a visual processing artifact — it is how language models weight information in ordered sequences. On landing pages, this means content that appears earlier in the document's linear structure carries more weight in AI buyer reasoning, independent of visual hierarchy. Headlines and above-the-fold copy are not just attention-getters; they anchor the entire evaluation frame.
Framing and anchoring effects are measurable and exploitable
Replication studies of classic behavioral economics experiments confirm that AI agents exhibit the same framing effects documented in human consumer research: identical choices framed as losses ('avoid losing $10') produce different responses than the same choices framed as gains ('save $10'). Status quo framing — presenting one option as the established default — measurably increases its selection frequency, replicating the same status quo bias observed in human decision-making. Anchoring is also measurable: AI agents shown a reference price systematically weight that anchor in their valuation judgments, with anchor effects comparable in direction to human studies (though somewhat attenuated in magnitude). These effects are not incidental — they are encoded in how language models represent value from training on human-generated text.
Credibility signals vs. self-assertion: a persistent divide
Controlled experiments consistently find that AI buyers heavily weight third-party credibility signals — editorial curation, press mentions, analyst endorsements, verified reviews — and discount self-promotional language, superlatives, and uncorroborated claims. The effect sizes are substantial: editorially endorsed products see selection probability increases of 14–43% in some categories, while 'sponsored' or promotional tags produce decreases of 8–13%. This maps directly to landing page copy: claims backed by named sources, specific numbers, or third-party validation outperform equivalent claims that are self-asserted.
Choice homogeneity: AI buyers collapse to consensus
Human consumers distribute purchase intent across a range of options. AI buyers, evaluated at scale, show significantly higher consensus: demand collapses onto one or two dominant products, with the long tail of alternatives receiving near-zero selection probability. This is not a problem to eliminate — it is a signal to interpret. When eLLMo's panel of personas reaches strong consensus around a specific friction point, that consensus is a reliable indicator of a real conversion barrier. When the panel diverges, the insight is about segment-specific positioning. Consensus and divergence are both informative.
Model version shifts are distributional, not incremental
When a frontier model releases a new version, the behavioral profile of AI buyers changes in ways that cannot be predicted by extrapolation from the prior version. In documented cases, position preferences inverted (bottom-row bias became top-row bias after a model update), and price elasticity coefficients shifted significantly. For a simulation product, this creates a requirement: every simulation run must be tagged with a specific model version, and version upgrades must be treated as a change in the measurement instrument, not just an improvement to it.
Methodology note
Built on the research. Designed for decisions.
eLLMo simulation surfaces ranked friction patterns across calibrated buyer personas — specific findings, traceable to buyer segments, actionable on the same day. The methodology is grounded in peer-reviewed research on AI agent behavior and OCEAN psychometrics. The output is a prioritized list of what to fix before your campaign launches — and why it matters for each buyer type.