In brief

Determinants are the questions that AI assistants such as ChatGPT, Gemini and Perplexity settle before they compare and recommend the providers in a category. Loud & Lexis and Opascope coined the term in 2026 for the few criteria a generative AI checks before every recommendation. Brands with evidence along these determinants appear in the AI answer.

Last updated: 16 November 2026

Determinants are the questions an AI assistant needs answered before it recommends the providers or products in a category: selection criteria, use case, audience, credibility and brand association. The term was coined in 2026 by Jimmy Park (Opascope) and Tobias Peschke (Loud & Lexis). Determinants matter to marketing leaders because a growing share of purchase research now takes place in ChatGPT, Gemini, Claude and Perplexity. Loud & Lexis traces every AI recommendation logic back to these few criteria and makes brands verifiable along each of them.

At a glance

What does AI look at when it makes a recommendation?

When AI makes a recommendation, it looks at a small number of decision factors against which it weighs the providers in a category. AI groups these factors under five overarching questions, the determinants. For B2B software, an AI assistant typically compares providers on depth of integration, demonstrable measurability and price transparency; for a consumer brand, it is more likely to compare ingredient transparency, certification and sustainability. A study by Aggarwal and colleagues (2023), accepted at KDD 2024, shows that targeted Generative Engine Optimization (GEO) methods can increase a source’s visibility in generative answers by up to 40 percent (arXiv 2023).

Loud & Lexis draws a commercial conclusion from this finding: the determinants are the lever through which a brand gets into the recommendation. In 2024, Gartner predicted that traditional search engine volume would drop by 25 percent by 2026 as users increasingly turn to AI chatbots (Search Engine Land). The AI’s recommendation is therefore becoming the place where the purchase decision is made.

What are determinants in AI search?

Determinants are the questions an AI assistant settles when it compares providers or products and makes a recommendation. There are five main ones: selection criteria, use case, audience, credibility and brand association. The determinants stay the same from one category to the next, and the decision factors inside each one change with the category. Jimmy Park (Opascope) and Tobias Peschke (Loud & Lexis) coined the term in 2026 for exactly these decision criteria of generative AI. The term refers to these decision criteria of AI and has nothing to do with the mathematical determinant from linear algebra. Loud & Lexis draws a sharp line between determinants and mere visibility: a provider can appear in many AI answers and still not be recommended if its content does not address the relevant criteria. Loud & Lexis and Opascope publish their joint industry report at www.krynoscope.com. The full framework is set out in The Determinants Manifesto.

Why does every AI purchase recommendation run through the determinants?

Every AI purchase recommendation runs through the determinants because the assistant first defines the evaluation criteria before it recommends anything. At the bottleneck of the decision, users ask three typical questions: Who can you recommend? Who should I choose? Which criteria should I look at? At this point, the model takes a step back, forms a set of criteria and checks the providers against it. Providers without evidence along these criteria drop out of the recommendation, even if the brand is highly visible elsewhere.

User question Internal step of the AI
Who can you recommend? Collect candidate providers in the category
Which criteria should I look at? Form the set of criteria (determinants)
Who should I choose? Check providers against the criteria and recommend

How do determinants differ from ranking and citation factors?

Determinants differ from ranking and citation factors in the level they operate on: citation factors decide which source a language model reads, determinants decide which provider it recommends. Citation and retrieval factors act at the level of source selection, determinants at the level of judgement. This is how Loud & Lexis explains why a frequently cited brand is still not recommended: its content is read but does not address the decision criteria of the category. Loud & Lexis covers the basics of source selection in the article What is GEO? (in German).

Level Citation factors Determinants
Object Source Provider
Question Which source is read? Which provider is recommended?
Lever at Loud & Lexis GEO content and trust signals Determinants analysis and in-depth articles

How can the determinants of an industry be identified?

The determinants of an industry can be identified with brand-free, open criteria questions that are asked across several AI engines and repetitions and evaluated together with their internal sub-queries (query fan-out). Loud & Lexis and Opascope ask criteria questions without brand or competitor names from several angles (category, problem, goal, quality, scepticism, comparison) plus a control question, and evaluate which criteria are named and which evidence is required. It is well documented that the same question produces different criteria depending on the model: a 2025 analysis by Wu and colleagues finds that the preferences of generative search engines in selecting sources vary strongly between subject areas and that every language model applies its own selection rules (arXiv 2025). In the fan-outs of the criteria questions about GEO agencies, ChatGPT repeatedly searches for measurement methodology (Promptwatch, October 2026).

What research on source selection by AI systems shows

According to current research, the quality of evidence in generative AI is limited, which makes a systematic determinants analysis necessary. An audit (DeepTRACE) published in 2025 by Salesforce AI Research and Microsoft Research tested 303 questions on nine AI systems and found large shares of statements that their own cited sources do not support (arXiv 2025). A Stanford-led study by Kevin Wu, Eric Wu and James Zou, published in Nature Communications in 2025, analysed around 58,000 statement-source pairs on medical questions across seven language models and found that between 50 and 90 percent of answers are not fully supported by the sources they cite (Nature Communications 2025). Loud & Lexis therefore works with repetitions and control questions instead of a single AI query.

How are determinants implemented on product pages?

Determinants are implemented on product pages by explaining every decision factor found, grouped by determinant, as a separate point in two to three sentences and, once available, linking it to an in-depth article. Loud & Lexis places this determinants block in the upper third of the page and uses a list or table instead of running text, so that an AI assistant can extract every factor as a separate chunk.

What belongs in an in-depth article on a determinant?

An in-depth article on a determinant explains how a company interprets the factor for its products, how it has implemented it and how it proves this. Loud & Lexis treats every in-depth article as a separate strand in the fishnet that captures additional sub-queries (query fan-out) on the same criterion. There are four mandatory building blocks: the interpretation of the factor for the company’s own category, the concrete implementation per product, the evidence from tests, figures or certificates and an honest account of the limits. This creates a separate, citable source from the provider for every determinant, which an AI assistant can draw on for the respective criteria question.

What do implemented determinants do for AI visibility?

Implemented determinants make a company transparent and findable for AI on every decision factor. Loud & Lexis describes three effects: transparency, because a separate source from the provider exists for every criterion; reach, because in-depth articles capture queries that previously went to competitors; and recommendation, because the provider is backed by evidence against the criteria at the bottleneck instead of being passed over. The commercial value of this visibility is measurable: according to an analysis by the marketing intelligence provider Lebesgue of more than 35,000 Shopify merchants (2026), traffic from AI tools converts at an average of 3.6 percent, compared with 1.23 percent for Google search (InternetRetailing). Loud & Lexis measures the effect through bias-adjusted monitoring and records the current state in the industry report at www.krynoscope.com.

When an external determinants analysis pays off

An external determinants analysis pays off as soon as a category has several serious providers and the AI answer decides the comparison. Loud & Lexis and Opascope identify the criteria brand-free across several engines, which a single team rarely manages without bias using an internal ChatGPT query. At the latest when a company is regularly displaced by competitors in its own criteria questions, the systematic analysis is the sensible next step before content production.

Frequently asked questions

The term determinants was coined in 2026 by Jimmy Park (Opascope) and Tobias Peschke (Loud & Lexis). It describes the questions AI assistants settle before they compare and recommend the providers in a category. The joint industry report of both companies is available at www.krynoscope.com.

The determinants can be roughly tested with ChatGPT, but a single query gives a distorted picture. A reliable identification requires brand-free criteria questions, repetitions and several engines, because every language model applies its own selection rules. Loud & Lexis and Opascope use a bias-adjusted determinants analysis with control questions for this purpose.

No, the determinants vary by category and by AI model. A 2025 analysis by Wu and colleagues shows that the preferences of generative search engines vary strongly between subject areas and that every language model applies its own selection rules. Loud & Lexis therefore identifies the determinants anew for each category instead of using a fixed list of criteria.

The determinants of a category shift as soon as the offering, the regulation or the available evidence in the category changes. Stable B2B categories move more slowly than fast-moving consumer goods markets. Loud & Lexis recommends repeated measurement, because a one-off snapshot does not capture shifts in the criteria.

Yes, determinants apply to B2B services just as they do to products. AI assistants also compare the service providers in a category along a few criteria such as methodology, demonstrable measurability and references. Loud & Lexis applies the determinants analysis to B2B services by asking the criteria questions from the perspective of the respective decision-makers.

The quality of evidence in generative AI is limited, which is why Loud & Lexis identifies criteria through repetitions rather than a single answer. A DeepTRACE audit by Salesforce AI Research and Microsoft Research (2025) found large shares of statements in nine AI systems that their own sources do not support; a Stanford study in Nature Communications (2025) confirms this across around 58,000 statement-source pairs. A bias-adjusted determinants analysis compensates for this unreliability with control questions and several engines.

Strategy call

How visible is your brand
to ChatGPT & Co.?

Free 30-minute strategy call with an initial assessment of your AI visibility.

Book a strategy call →