This publication is produced with support from Docket, which sells software in this category.
Google AI Overviews cited marketingagent.com on zero of twenty questions in our September 17, 2026 study. We used the publication's previously published questions about marketing agents and AI-search measurement, retaining two completed answers per question.
Those forty answers came from one signed-in, personalized browser profile, with English results and a US location. This is a snapshot of that context and question set, not an estimate of what every buyer sees. Study record
For this study, answer share means the percentage of our fixed questions on which an engine cites marketingagent.com in at least one of two passes. It measures citation coverage across questions.
| Engine | Completed answers retained | Verified citation observations | Result over the fixed twenty questions |
|---|---|---|---|
| Google AI Overviews | 40 of 40 | 40 negatives | 0 of 20 questions, or 0% |
| ChatGPT | 25 of 40 | 17 negatives; 8 unresolved | Unavailable |
| Perplexity | 1 of 40 | 1 negative; next question gated | Unavailable |
The observation ledger includes every planned cell. Missing or unresolved observations leave an empty score; the two incomplete engines receive no percentage.
The answer about answer share changed its denominator
The final question asked what answer share is and how to calculate it. Google's first retained answer divided answers naming a brand by prompts tested; its second used competitor mentions as the denominator, then offered prompt coverage as an alternative. Both outputs are retained under question 20 in the Google capture archive.
That difference changes what a number means. Question coverage can rise because a site appears on more questions, while a share calculated across brand mentions can also move because other brands receive more or fewer mentions.
Neither calculation, by itself, says whether a mention includes a source link. Our study requires a qualifying link to this publication in the composed answer; an unlinked name or retrieval-panel result does not count.
A question scores one if either pass contains that citation, regardless of repeated links within an answer. Both passes must be complete and verifiable before it scores zero.
The denominator remains the same twenty questions: 0 ÷ 20 × 100 for Google in this run. With citation presence zero, position is not applicable and framing is absent.
What does the Google zero establish?
The site was absent from answer citations in both retained passes for every question. We expanded each Overview and its related sources, saved the answer text and link destinations, and checked for marketingagent.com links.
The result does not explain the absence. We did not test crawling, indexing, content quality or Google's reasons for choosing a source, so the zero alone cannot tell an editor which page to rewrite.
The two passes came from the same profile on the same evening and may include cached responses. Forty saved answers should not be interpreted as forty independent observations of buyer behavior.
One capture interruption on the final question required another search. The archive preserves that unscored interruption beside the completed capture.
Why are the other engine results unavailable?
ChatGPT's twenty-five retained answers include seventeen verified negatives and eight with unresolved citation coverage: grouped sources were not fully preserved. A visible lead link is insufficient to certify absence from the other sources.
We stopped collection following a request-limit notice, leaving fifteen planned prompts unsubmitted. Five answers already in progress were recovered without regeneration; the method records that recovery.
These were signed-in Temporary chats with default automatic search and account personalization available. Fresh conversations did not create an anonymous control, and the raw personalized text is excluded from the public supplement.
Perplexity returned one full answer, then a sign-up gate on the next question. We stopped there, leaving thirty-eight further prompts unsubmitted.
What carries forward into the next run?
The full method records session settings, scoring and every missing observation. Our earlier GEO measurement article explains the broader rationale.
That earlier method called for logged-out sessions in a fresh profile; this run used an existing browser, with Google and ChatGPT signed in. It establishes a baseline for the observed context, with no trend comparison against September 7, whose records did not retain enough evidence to validate their scores.
Our editorial judgment is that a future anonymous sample should be tracked separately from this existing-profile sample. Otherwise, a change in browser context could be mistaken for a change in the publication's visibility.
To audit this result, follow question 20's two captures into the ledger. A future zero can be as interpretable as an increase when the evidence is complete and collection conditions are comparable; explaining the cause of either result requires a separate test.