Skip to main content
Start with a Diagnostic

Hendricks Research

Research for the AI Search Era.

What this section is

Practical, source-supported research on how people search, how brands enter consideration, how AI-mediated discovery changes the buying journey, and how organizations can measure the commercial result.

Start with a Search Intelligence DiagnosticExplore the Latest Research

Hendricks observes four systems: Google AI Overviews, ChatGPT, Perplexity, and Gemini.

Published

The latest study.

  • The Answer Index
  • Edition 1, September 2026
  • Package version v2026.09.1
  • Quarterly

Four Engines, 480 Questions, 16,069 Citations. ChatGPT Cited Nothing on Six of Ten Question Types.

A research experiment. Hendricks sent the same 480 questions across eight industries to four AI answer engines and recorded 16,069 citations across 7,775 domains. ChatGPT cited a source in 72 of 480 answers, 69 of them on two of the ten question types, and no engine agreed with another engine anywhere near as much as it agreed with itself 38 minutes later.

AI-Mediated Search. Research experiment. Observation. The Answer Index is published quarterly.

By , Search Intelligence Engineer, HendricksPublished Updated

Data through

Take the report and the data with you

Edition
Edition 1, September 2026
Package version
v2026.09.1
Cadence
Quarterly
Data DOI
10.5281/zenodo.22242103
Latest version
10.5281/zenodo.22242102
Related solution
Selection Intelligence

Supporting studies

The studies behind it.

Each study names its relation to the flagship in the flagship’s own words.

  1. Measurement and Attribution. Research experiment. Observation.

    Published
    Updated
    Data through

    Three Engines, 17 Questions, 449 Citations, and Not One Source Cited by All Three. ChatGPT Cited Nothing on 15 of 17.

    A research experiment. In one archived run, 17 questions across three AI engines filled 449 citation slots with 319 distinct domains, no domain was cited by all three engines, and ChatGPT cited no source at all on 15 of the 17 questions it answered.

    Relation to the flagship: The 17-question run this corpus scales by a factor of 28, and the page whose zero-shared-domains reading the larger corpus explains rather than refutes.

  2. AI-Mediated Search. Research experiment. Observation.

    Published
    Updated
    Data through

    An AI Engine Cited a Hendricks Article the Firm Had Deleted. The Next Run Returned Zero Citations.

    A research experiment. Hendricks measured its own citation presence twice. Run 1 returned one citation, to an article the firm had deleted. Run 2 returned none across 51 cells.

    Relation to the flagship: The same instrument, read for whether one brand appeared in the answers at all.

  3. Measurement and Attribution. Research experiment. Observation.

    Published
    Updated
    Data through

    Two Runs, Seven Hours Apart, Agreed on Whether Each Answer Cited Anyone. They Disagreed on Which Sources.

    A research experiment. The same 17 questions, three AI systems, two runs seven hours apart. Whether an engine cited anyone matched on 51 of 51 cells. Which sources it cited had a mean overlap of 0.68.

    Relation to the flagship: The first reading of answer stability, on 20 cells. This corpus repeats the design at 441 cells.

  4. AI-Mediated Search. Research experiment. Observation.

    Published
    Updated
    Data through

    One Run Cited 247 Domains Across 308 Citation Slots, 86 Percent of Them Once. There Was No Incumbent to Displace.

    A research experiment. In one archived run, 247 distinct domains filled 308 citation slots across 20 AI answers, 86 percent of those domains were cited exactly once, and Perplexity and ChatGPT shared a single source.

    Relation to the flagship: The structure of a citation set on an earlier run, and the first appearance of the no-incumbent finding this corpus confirms at scale.

Foundations

Start with the category pages.

The vocabulary the studies are written in, defined on pages of their own.

Definition

What Is Search Intelligence Engineering?

The category, defined before anything is sold under it.

Definition

What Is Selection Intelligence?

The measures a consideration baseline reports, defined before they are reported.

Definition

The AI Selection Problem

Why being discovered and being chosen came apart.

Definition

Search Presence Engineering

The seven engineering layers, and what each one is accountable for.

Coverage

What this section covers.

Hendricks Research supports the firm’s professional methodology and category education.

Eight categories

  • Search Intelligence Engineering
  • The AI Selection Problem
  • Selection Intelligence
  • Search Demand
  • AI-Mediated Search
  • Search Presence Engineering
  • Measurement and Attribution
  • Data and Systems

Publication Standards

What every research page carries.

A study that cannot be checked is an assertion. Every substantial page in this section publishes the material a reader needs in order to check it.

  1. A direct executive answer
  2. Key findings
  3. Definitions
  4. Data or primary evidence
  5. Methodology
  6. Sample and date range
  7. Assumptions
  8. Limitations
  9. Author
  10. Published date
  11. Meaningful updated date
  12. Data-through date
  13. Source list
  14. Corrections link
  15. Related solution
  16. Related research

Read in order

Read the method before the finding, and the limitations before either.

Establish a baseline through the DiagnosticReview the Demand-to-Selection methodology

Absence is not yet a diagnosis. A single answer screen is one observation under one set of conditions.