40% of AI citations generated by large language models fail to name the brand behind the source, according to a report published by agency network DesignRush.

The figure carries weight for any operator whose traffic now flows through AI assistants and generative search engines rather than traditional blue-link results. Without the full article body in the public feed, the underlying sample size, the AI tools tested, the prompt set, and the measurement dates remain unstated.

DesignRush operates a marketplace that connects businesses with marketing agencies. Its coverage tends to surface trends marketers can act on rather than academic benchmarks. The 40% number fits that pattern: directional, headline-friendly, and short on audit trail.

For trade-press readers, the statistic functions less as a verdict than as an open question.

What does the headline actually measure?

The figure compares cited sources in AI-generated answers against the brand identities behind those sources. If a model quotes a Statista dataset or an SEC filing without tagging the underlying publisher, that counts as an unnamed citation under this framing.

That framing matters. Brand visibility inside an LLM answer chain now functions as a discovery channel in its own right. Operators running paid search, programmatic, or content marketing see the AI overview box as another placement, one they cannot bid on directly.

Why would AI tools strip the brand name?

Three structural reasons explain a 40% gap, even without specifics from the source study:

• Source overlap. Multiple publishers often cover the same data point, and the model surfaces the fact and discards the attribution.
• Token economics. Shorter answers save inference cost, and brand names are easier to drop than numbers.
• Training-set opacity. Citing a brand accurately requires linkable provenance that the underlying model rarely carries.

What does a 40% omission rate cost?

No canonical number exists for the traffic value of a single AI citation. Realistic ranges from third-party clickstream work put the value between a small fraction of a traditional organic click and rough parity, depending on query class.

For an agency-side operator, the practical question is whether brand mentions inside AI answers can be bought, earned, or measured. The current answer: weakly on all three.

What's missing from the headline

The feed reaching Mart Signal carried only the article title. Body text, author byline, sample size, prompt list, and model versions evaluated were absent. Treat the 40% number as a hook for verification, not a benchmark.

Operators monitoring AI citation behavior should ask vendors four questions before acting on the claim:

• Which model versions did the test cover?
• How many prompts were run per model?
• What taxonomy defined a "citation"?
• Over what date range did the crawl run?

Until those details land, the 40% claim serves as a directional signal that AI citation hygiene is a real working category for marketing operators, not a settled figure.