Back to Market Signals
AI ToolsTrendinghigh impact

Perplexity Drops Brand Names in 52% of Citations as AI Search Reshapes Visibility

By InnovaAI Research1 min readSearchengineland

New research shows Perplexity omits brand names in 52% of its citations, meaning agencies can win AI search placement for clients without winning brand recognition. Separately, prompt injection attacks are exploiting help centers and multimodal inputs, creating new risk vectors for AI-assisted workflows.

Key Facts

01Perplexity omits brand names in 52% of its citations, separating content visibility from brand recognition.
02Source selection in Perplexity happens at the retrieval stream stage, before answer generation, requiring passage-level optimization.
03Prompt injection attacks now target help centers, podcasts, video, and voice agents as entry points into AI workflows.
04Forrester notes enterprise teams are deploying agentic AI to production faster than governance frameworks can keep pace.
05Classic RAG and agentic RAG have materially different failure modes, and agencies should know which one they are recommending to clients.

Why It Matters

▶A 52% brand-name omission rate means citation volume is no longer a reliable proxy for brand awareness growth in AI search channels.
▶Prompt injection via trusted content sources creates brand safety exposure in AI workflows that standard security reviews do not currently cover.
▶The fragmentation of the agent market in H1 2026 means vendor decisions made without clear criteria will require frequent and costly revision.
▶Passage-level content structure now determines AI citation outcomes more directly than overall page authority or domain signals.

Agency Actions

Audit the top 20 cited pages for each client and rewrite key passages so the brand name appears in the extractable body text, not only in headings or metadata.

medium effort

Create an inventory of every AI-assisted workflow that ingests external or user-generated content, then add input sanitization and a human review checkpoint at each one.

high effort

Before recommending any RAG-based AI tool to clients, ask vendors to specify whether the system uses classic or agentic retrieval and what monitoring exists for agentic output loops.

low effort