Search is moving from a personalization model to a customization model. The two terms are often used interchangeably, but they describe opposite mechanics. The difference is the locus of control: personalization is a change the system applies to results without the user's input, while customization is a change the user selects.

Personalization (system-controlled, implicit). The engine infers intent from signals the user never sets: location, search history, device, and prior behavior. It differentiates results silently. When a million people run the same query, outputs diverge across users, and no user chose that divergence.

Customization (user-controlled, explicit). The engine serves one shared baseline result set, then exposes controls the user sets directly: which sources to prefer, which channels to include or exclude, and whether results are regionalized. Divergence is user-directed and opt-in.

The single distinction that separates the two models is choice. Personalization removes it. Customization grants it.

Key Facts

Organic CTR, informational queries: down 61% in 2026
AI Overviews trigger rate: 59% of informational queries vs 19% of commercial queries
Preferred Source label: roughly 2x click-through when applied in Top Stories or AI Overviews
Author credentials on top-ranking pages: 73%, up from 58% before the 2026 updates
2026 Google updates: February Discover, March Spam, April Core

Customization Controls in Practice

Preferred Sources is the clearest customization control. A signed-in user designates specific brands as preferred. Designated results then carry a "Preferred" label in Top Stories and AI Overviews and are roughly twice as likely to be clicked when that label is shown. The mechanic runs in three stages:

Selection: the user marks the brand as preferred.
Prioritization: the system applies the "Preferred" label to that brand's results.
Authority: later AI summaries draw on the brand as a primary source.

Ranking position is no longer the controlling variable. Inclusion in the user's selected set is.

The Reading Layer: Agentic Retrieval

Modern search runs on Gemini Flash, a low-latency model used for agentic retrieval, meaning AI that completes tasks autonomously. In many queries the consumer of a page is an agent assembling an answer, not a human. Measured behavior of that layer:

Agent parse time is roughly 400 milliseconds per page. Human skim time is 4 to 8 minutes.
Agent visits register as a bounce with zero session duration, so they are absent from standard analytics.
Agents tend to truncate or skip pages above roughly 15,000 tokens for core or "Quick Start" pages, and 30,000 tokens for individual pages. A token is approximately one chunk of text a model reads at once.
Markdown parses with lower overhead than heavy HTML or JavaScript.
An llms.txt file at the site root functions as a sitemap for agents. A robots.txt that blocks AI crawlers removes the brand from AI search.

Trust Signals Under the 2026 Updates

To suppress AI-generated filler, the ranking system now weights original reporting and identified human authorship. This parallels LinkedIn's 2025 algorithm change, which increased the reach of first-person, experience-based posts. Three weighted signals:

Information gain: original data or a viewpoint not already present in the top 10 results.
Highly cited labels: the citation is assigned to the first publisher of a fact, even over a higher-authority domain.
Perspectives carousel: a UI slot that surfaces human discussion from sources such as Reddit and TikTok.

Measurement

With informational CTR down 61%, total traffic is no longer a meaningful metric. Two measures replace it:

Cost avoidance: the value of AI citations benchmarked against paid equivalents. A cited appearance in an AI Overview for a keyword at a $15.00 CPC, across 1,000 impressions, equals $15,000 in avoided paid-search cost.
Assisted brand lift: an AI mention frequently triggers a follow-on branded search, so growth in branded search volume serves as a proxy for AI-citation success.
ROI formula: ROI = (Revenue - Cost) / Cost, where Revenue is counted as the customer lifetime value of leads that organic search influenced, including those whose final click came through another channel.

Subjective

The operative implication is a change in target metric. Personalization rewarded matching inferred intent. Customization rewards two things a brand can act on directly: being selected as a preferred source, and being parsable by retrieval agents. Ranking position is downstream of both. In this model, source worthiness, not rank, is the variable that determines visibility.