Campaign India: Hallucinated Brand Capabilities in AI Search

Campaign India reported on 21 November 2025 on Pulp Strategy’s GEO Benchmark Index 2025 whitepaper about brands in AI-driven search. Its headline finding was hallucinated capabilities: AI answers crediting brands with products, partners, or features they do not have. The report also covers brands missing from AI answers, mistaken identities, negative sentiment, and the warning that CMOs who ignore GEO lose control of their brand narrative.

Invented capabilities take several forms

Campaign India gives concrete examples of what the research found AI answers getting wrong: product lines that do not exist, incorrect founding dates, partnerships that were never formed, and a competitor’s feature presented as the brand’s own. A buyer comparing vendors in ChatGPT, Gemini, Claude, or Perplexity has little reason to question any of them.

The whitepaper found hallucinated capabilities for 52 percent of brands.

NeuroRank classifies gaps in an AI answer through ORHL: Omitted, Replaced, Hallucinated, Zero Leads. Each of these examples is the Hallucinated condition.

Brands drop out where purchase intent is highest

The report found that 68 percent of brands failed to appear for one or more core commercial prompts. The absence was most marked in mid- and lower-funnel questions, the ones built around a brand’s unique selling points and reasons to believe. Strong traditional search rankings did not guarantee that AI models recalled the brand.

It also found inconsistent naming and recognition across models for 88 percent of brands, and negative sentiment bias in AI-generated summaries for 90 percent of consumer brands.

Campaign India sets this against the rise of AI-driven search, from Search GPT to Perplexity, and cites Adobe data showing steep growth in AI-referred traffic to US retail sites, with those visitors staying longer and spending more per visit.

The causes sit in how brands publish

According to the whitepaper, the gaps come from how brand information is published, independent of market performance. The causes: content in formats a machine struggles to read, PDFs and pages without structure, and thin trust signals in the knowledge graphs models draw on. References to missing structured data and schema appeared more than 600 times across the audits, and the word hallucination more than 200 times, roughly three times per report. Machine-readable content standards became the most frequent recommendation.

The report analyzed 70 full-spectrum LLM visibility audits across more than 350 brands, 408,000 prompt simulations, and more than 60 CMO interviews. Each audit covered a content inventory, hallucination and gap checks, top prompts, competitor battlecards, and visibility recommendations.

Narrative control passes to the algorithm

The report closes on a warning for marketing leaders. CMOs who disregard these signals risk handing their brand narrative to algorithms that take no account of long-term strategy. Those who adopt GEO, in Pulp Strategy’s words, can steer the conversation, protect their reputation, and capture high-value intent at the source.

Hallucinated is a quality flag

In NeuroRank’s method, Hallucinated is a flag that attaches to any answer naming a brand when the answer contains a factually incorrect statement about it. A brand can be recommended and hallucinated in the same response. NeuroRank’s Hallucination Score reports the share of brand-naming responses that contain such an error, and lower is better.

Where the research stands now

Pulp Strategy’s 2025 study has since been followed by NeuroRank’s AI Visibility Index 2026, the current research program.