How big is the problem?

When Metricus runs AI visibility reports across brands, we find that 72% have at least one factual error in AI-generated responses. These are not subtle interpretive differences — they are demonstrably wrong facts: incorrect founding dates, wrong pricing, discontinued products listed as current, features attributed to the wrong product tier, and fabricated company details that never existed.

What we found: The error rate increases with brand complexity. Companies with multiple product lines, frequent pricing changes, or recent rebrands have the highest hallucination rates. The most common errors involve pricing (cited from stale review sites) and feature descriptions (conflated across product tiers).

The 5 most common AI errors about brands

  • Wrong pricing: AI cites prices from 12–24 months ago, typically sourced from cached G2 or Capterra listings. This is the most common error we find.
  • Feature conflation: AI merges features from different tiers or product lines into a single description, misrepresenting what each plan includes.
  • Fabricated details: AI invents founding dates, employee counts, or headquarters locations that have no basis in any indexed source. These are pure hallucinations.
  • Outdated information: AI recommends discontinued products, cites old partnerships, or describes deprecated features as current.
  • Competitive misattribution: AI attributes a competitor’s feature or product to your brand, or vice versa, typically sourced from comparison articles.

What we also found is that error types cluster by platform. ChatGPT tends toward feature conflation and fabricated details because it relies more heavily on training data. Perplexity, which uses more real-time web search, tends toward outdated pricing errors because it pulls from whatever stale sources rank well. Gemini shows higher rates of competitive misattribution, possibly due to how it synthesizes information from multiple comparison articles. Understanding which platform produces which errors is valuable because it indicates which source types are feeding the misinformation.

The business impact of AI hallucinations

What we found when measuring the downstream effects of AI hallucinations is that the damage extends beyond inaccuracy. When AI fabricates a founding date or employee count, it undermines trust with buyers who fact-check. When AI conflates pricing tiers, it creates mismatched expectations that sales teams must correct. When AI cites a discontinued product, it generates leads for something you no longer sell. Each type of error has a different cost profile, but all share one characteristic: they are invisible to you unless you actively audit what AI says about your brand.

The compounding effect is particularly concerning. What we found is that AI hallucinations about a brand tend to persist across model updates unless actively corrected. A fabricated detail embedded in one model’s training data can propagate to newer models, to AI-generated content on third-party sites, and from there back into future training data. Without intervention, hallucinations become self-reinforcing.

Which brands are most at risk

What we found is that certain brand characteristics correlate with higher hallucination rates. Brands that have undergone recent name changes, mergers, or acquisitions see the highest error rates because AI’s training data contains both old and new information. Brands with complex pricing structures (usage-based, tiered, or custom enterprise pricing) are more frequently misrepresented because AI models simplify variable pricing into static numbers. Brands with limited web presence outside their own domain have higher fabrication rates because AI fills information gaps with plausible-sounding but invented details.

Why hallucinations happen to brands

What we found is that AI hallucinations about brands stem from three sources: conflicting information across indexed sources (your pricing page says one thing, a review site says another, and AI picks the wrong one), information gaps that the model fills with plausible-sounding fabrications, and stale training data that reflects your brand as it was months or years ago rather than as it is today. The more inconsistent your brand information is across the web, the more likely AI is to get something wrong.

Understanding what AI gets wrong about your brand and which sources are feeding the errors is the prerequisite to any correction strategy. A Metricus AI visibility report includes a factual accuracy check that flags every error across all major AI platforms and traces each to its source.

Last updated: April 2026