The AI Selection Problem
Brands are losing control over the path between being discovered and being chosen.
The AI Selection Problem
The AI Selection Problem is the loss of control over the path between being discovered and being chosen.
Traditional search largely helped customers find pages.
AI-mediated search can perform more of the interpretation, research, comparison, and evaluation before the customer reaches a website.
Journey Comparison
The shape of the journey changed.
Traditional
- Query
- Search Results
- Website
- Conversion
AI-mediated
- Need
- Intent Interpretation
- Research
- Evaluation
- Synthesis
- Shortlist
- Choice
The Business Consequence
A company can have:
- A polished website
- Strong technical SEO
- Thousands of links
- Excellent reviews
- Paid visibility
- Recognized expertise
- A respected brand
And still be absent from the options presented during an AI-assisted buying decision.
The brand may lose before a website visit ever occurs.
Visibility Is Not Enough
Each stage has to be earned separately.
- Being seen does not guarantee being understood.
- Being understood does not guarantee relevance.
- Being relevant does not guarantee trust.
- Being trusted does not guarantee consideration.
- Being considered does not guarantee recommendation.
- Being recommended does not guarantee human selection.
The actual path is:
- Discoverable
- Understood
- Relevant
- Trusted
- Cited
- Considered
- Recommended
- Selected
- Revenue
One Observation, Not A Ranking
Why does ChatGPT recommend a competitor instead of your brand?
A competitor recommendation is one observation, not a ranking. It records that a competitor was named once, on one surface, on one date, under one set of conditions. Nothing in a single answer screen establishes that the same competitor is named on the next run, or that a shortlist formed and the brand lost a place on it.
Hendricks published a dated run of this measurement against its own brand. In run 2026-08-19-110930, 247 distinct domains filled 308 citation slots across the answers that cited anything, and 212 of those domains were cited exactly once.
A citation set distributed that way is not a standings table with a competitor placed above a brand in it. It is a wide, shallow spread in which most sources appear once and then do not appear again. That is what a reader is looking at when a competitor is named in an answer and their own brand is not.
Honest limitation
What that run does not establish.
The 2026-08-19 run covered 17 buyer questions on one date and measured 47 of its 51 cells, with nothing changed between it and the run before it and nothing held back for comparison. It records how the sources in those answers were distributed. It establishes nothing about why any source was chosen, nothing about any other brand’s category, and nothing about what a change to a website would do to a later answer.
Separating the causes behind a competitor recommendation is a different job. The What Is AI-Mediated Search page separates them in order, cheapest first. What comes before that job is the reading itself: an answer screen is one observation, and an observation is worth what its sampling is worth.
No firm controls whether an AI system recommends a brand, and Hendricks does not sell that outcome. What can be established is whether a brand enters consideration, under which customer contexts, on which observed surfaces, and on which dates. A brand that has seen a competitor recommended once knows that it happened once. The What Is Selection Intelligence page states what a baseline reports instead, and why it is reported that way.
The Intelligence Gap
Most businesses cannot answer:
- Was our brand considered?
- Was it mentioned or actually recommended?
- Which sources influenced the visible answer?
- What does the system appear to understand about us?
- Which competitors entered the shortlist?
- Under which customer contexts do we win or lose?
- Which claims have corroborating evidence?
- Where is information incomplete or contradictory?
- What should we change first?
- Did visibility affect customer behavior or pipeline?
The Hendricks Response
Four moves, in order.
- Measure demandDetermine which needs and decisions matter.
- Understand selectionObserve whether the brand enters commercially relevant consideration.
- Engineer the presenceImprove the conditions the business can control.
- Prove impactMeasure what changes and state the evidence honestly.
Sources
Sources
Last reviewed .
This page states the Hendricks position on how AI-mediated search changes buying journeys. The figures it quotes come from a dated first-party Hendricks run, published in full with its denominators and its limits on the Hendricks Selection Baseline. No third-party research, vendor study, or external finding is claimed.
This definition is applied in Selection Intelligence and the Diagnostic.
Change History
Change history
| Date | Kind | Summary |
|---|---|---|
| Not yet recorded | Publication | First publication of this page. |