Zero brands named, but the sources are still there
Two earlier posts on this blog measured null responses: organic, buyer-intent prompts where an AI engine names none of the six brands a panel tracks, the target plus five named competitors. The question those posts did not answer is what the engine was doing instead. Did it fail to retrieve anything and fall back on general knowledge, or did it retrieve real pages and simply choose not to name any of the tracked brands from what it read? The citation field on every scored response answers that directly, and the answer is not what a "failed to find anything" story would predict.
Null responses cite sources almost as often as normal responses do
We re-scored every real, search-grounded organic response across our 270 live audit panels (project management, customer support, and HR software; ChatGPT with web search, Gemini with grounding, Perplexity Sonar, and Claude with web search) using CitedWell's production scoring logic, and this time compared the citation field between the two groups: responses that named zero tracked brands (null) and responses that named at least one.
| Response type | Count | Carried real citations | Avg citations per response |
|---|---|---|---|
| Null (zero brands named) | 488 | 95.7% (467/488) | 7.0 |
| Non-null (at least one brand named) | 6,126 | 95.8% (5,866/6,126) | 8.9 |
Those two citation-presence rates are within a tenth of a point of each other. A null response is not, in the large majority of cases, an engine coming up empty. It is an engine that read seven real source pages on average, the same category of pages a normal response reads, and still did not put any of the panel's six brand names into its answer. Retrieval worked. Naming did not follow.
One engine breaks the pattern
The overall number hides a real split once you group null responses by engine. Claude and Perplexity produce the large majority of null responses in this dataset (263 and 179 of the 488), and on both engines the citation rate on those nulls is close to universal.
| Engine | Null responses | Carried real citations |
|---|---|---|
| Perplexity | 179 | 100.0% (179/179) |
| Claude | 263 | 99.2% (261/263) |
| Gemini | 41 | 65.9% (27/41) |
| ChatGPT | 5 | too small to report |
Claude and Perplexity's null responses read like fully-sourced answers that happen not to mention a tracked brand. Gemini is different: about a third of its null responses carry no traceable citation at all (after excluding Gemini's opaque vertexaisearch redirect links, which an earlier post on this blog covered separately). ChatGPT's null count is too thin here to report a rate, five responses out of 973 organic ChatGPT responses in the whole dataset. For Claude and Perplexity specifically, a null result is almost never a retrieval gap. For roughly a third of Gemini's nulls, it might be.
What the citations actually point to
Most of the null-with-citations responses in this dataset come from project management software's "for agencies" qualifier, the same prompt an earlier post identified as the single biggest driver of Claude's and Perplexity's null rate. Pulling the citation URLs from those specific responses shows exactly what the engines read before answering.
Why this changes what a null result means
A null score on its own looks like a blank: the AI did not talk about your category, or could not find anything relevant to say. That is not what the citation data shows. In 95.7% of null responses, the engine read real, specific, topical source pages, close to as many as it reads on a normal response, and then wrote an answer using a different set of brand names than the six a panel is built to track. That is a content and entity problem: the pages the engine retrieved for that specific query do not talk about your brand, or talk about a narrower competitive set your panel was not built to catch. It is not a signal that the engine failed to find anything to say.
The one exception worth flagging separately is Gemini. When roughly a third of its null responses carry no traceable citation, some of that group really may be Gemini answering from general training knowledge rather than a specific retrieved page, which is a different failure mode than "read the wrong pages" and calls for checking whether the query is grounding at all before assuming a content fix will move the needle.
Find out whether your own null results are a content gap or a retrieval gap, broken down by engine.
Get an AI Visibility Audit, $490Methodology
Data drawn from 270 live, search-grounded audit panels (project management, customer support, and HR software brands), each run across four AI engines: ChatGPT with web search, Gemini with grounding, Perplexity Sonar, and Claude with web search. We re-scored every successful (non-error) organic (non-branded) response using CitedWell's production scoring logic (engine/scorer.ts) this session, 6,614 responses total, and cross-referenced each response's targetBrandMentioned/competitorsMentioned result against its citedSources field. A response counted as null when it named zero of the panel's six tracked brand names. Citation counts exclude Gemini's opaque vertexaisearch.cloud.google.com redirect links, which do not resolve to an auditable domain, matching the exclusion CitedWell's production scorer already applies when ranking top cited sources. No development-rail or fixture data is included; all responses came from live engine calls. Data collected June-July 2026.