Primary Bias
| Category | AI model phenomenon |
|---|---|
| Introduced | 2025-2026 |
| Attributed to | Dan Petrovic / DEJAN entity home · about/bio |
| Attribution status | Explicit creator concept |
| Attribution confidence | High |
| Original source | dejan.ai |
| Related concepts | |
| Selection Rate Selection Rate Optimization Algorithmic Genericization Entity Home | |
Primary Bias is a ai model phenomenon used in AI SEO. A model's prior association/pre-retrieval preference regarding a brand/entity before current retrieval evidence. The term or framework is attributed to Dan Petrovic / DEJAN and is documented in 2025-2026.[1]
AI SEO Wiki records the attribution as Explicit creator concept, with a confidence rating of High. This rating concerns the documentary evidence for the term's origin; it does not imply that the underlying idea is unique to one practitioner or that the terminology has become an industry standard.[2]
Definition
A model's prior association/pre-retrieval preference regarding a brand/entity before current retrieval evidence. In practical usage, the concept is most useful when its measurement unit, observation window, source set, or decision context is stated explicitly. This is particularly important in AI-mediated search, where answers can vary between models, prompts, retrieval systems, and repeated runs.
Origin and attribution
Primary Bias is attributed in the available documentary record to Dan Petrovic / DEJAN. The registry dates the term to 2025-2026 and classifies the evidence as Explicit creator concept.[1]
Useful pair with Selection Rate. This qualification is retained because AI-search terminology often spreads rapidly through conference talks, social posts, vendor material, and practitioner communities before a stable publication record exists.
An additional source provides independent or supporting evidence for the terminology or attribution.[2]
Role in AI SEO
Primary Bias names an observed or proposed pattern in AI-mediated search. Naming the pattern makes it easier to distinguish a repeatable change in system behaviour from ordinary volatility in a single prompt or model run.
The concept is most useful diagnostically: practitioners can test whether the pattern appears across prompt sets, model versions, source corpora, and observation periods before treating it as a causal explanation.
Application and interpretation
The concept is applied by first identifying the observable behaviour it describes, then testing that behaviour across a sufficiently broad set of prompts or sources. It should be used as a model for investigation rather than as a substitute for direct evidence from the relevant search or AI system.
Related concepts
| Concept | Relationship |
|---|---|
| Selection Rate | How often a model selects a brand/source from the candidate set it could use when generating an answer. |
| Selection Rate Optimization | Optimizing the probability that an AI model selects a brand/source from its candidate set. |
| Algorithmic Genericization | Machines collapsing/substituting a brand or entity into generic concepts when identity signals are weak or inconsistent. |
| Entity Home | The canonical page/source a search or AI system can use as the authoritative home for an entity's facts. |
Limitations and attribution notes
The attribution evidence for Primary Bias is comparatively strong, but the existence of a documented term origin does not establish that every underlying mechanism was first discovered by the named creator. Similar ideas can arise independently under different terminology.
The field itself changes rapidly. Model behaviour, retrieval systems, citation interfaces, and measurement tooling can change without the terminology changing, so operational claims should be re-tested against current systems.
See also
References
- Primary/original source for Primary Bias. dejan.ai. https://dejan.ai/blog/primary-bias/.
- Creator/entity home. dejan.ai. https://dejan.ai/.
- Creator biography/about page. dejan.ai. https://dejan.ai/about-us/.