What ChatGPT Cited: A StrokeVideo Case Study
Five questions, fifteen saved answers, and a practical way to separate brand mentions from website citations.
Five questions, fifteen saved answers, and a practical way to separate brand mentions from website citations.
StrokeVideo appeared in 8 of 15 saved ChatGPT answers, while its website was cited in 9. Those two counts describe different things: a brand can be named without a link, and a source can be linked without a product recommendation. This case shows how to read that difference before deciding what content to create.
It is a historical snapshot, not evidence that publishing a page increased visibility. We did not run a new experiment or ask the model to generate supporting results for this article.
The round started on October 26, 2026 at 06:00 UTC and finished at 06:09 UTC. It contains five English questions about whiteboard animation, with three answers per question. All 15 answers were saved successfully. The question records in the export specify English and the United States; these settings were not independently preserved as a complete historical configuration for every request.
The data came from a production monitoring round captured through the ChatGPT web-answer collection service. We used the stored answer text, mention flags and citation links. Counts follow the existing parser's definitions and were checked against the exported records; they are not an independent human rating of recommendation quality.
| Monitored question | Answers mentioning StrokeVideo | Answers citing strokevideo.com |
|---|---|---|
| Free AI whiteboard video makers with no watermark | 3 / 3 | 3 / 3 |
| Make a whiteboard explainer from a written script | 2 / 3 | 2 / 3 |
| Best AI whiteboard tools for explainer videos | 1 / 3 | 2 / 3 |
| Create an AI animated video without editing every scene manually | 0 / 3 | 0 / 3 |
| Turn a script into a hand-drawn whiteboard video automatically | 2 / 3 | 2 / 3 |
| Total | 8 / 15 | 9 / 15 |
The labels above shorten the original questions for readability. The underlying questions and three samples were kept separately. With only three answers per question, one changed answer moves a question's observed rate by roughly 33 percentage points. Treat the results as leads for inspection, not stable estimates of all buyers' experiences.
The broad “best tools” question illustrates the distinction: two answers linked to StrokeVideo, but only one was marked as mentioning the brand. A link alone therefore cannot establish that an answer recommended the product.
We do not report a separate recommendation percentage here. The historical parser retained mention and position fields, but did not supply an independently verified recommendation metric. For the definitions and denominators, see how to measure AI visibility.
Examples of links recorded in this round include StrokeVideo's whiteboard animation page, VideoScribe, and Jellypod's whiteboard software comparison. These are saved citation URLs, not endorsements or evidence that the pages remain unchanged today.
For a practical source review, classify a URL on two dimensions. First, who controls the domain: your brand, another product vendor, or a third party. Second, what kind of page it is: product page, documentation, comparison, community discussion, or another format. A competing vendor's comparison article may be third-party coverage of your brand while still belonging to a competitor. Write down that classification rule rather than hiding the ambiguity in a chart.
Count unique pages separately from citation occurrences. The same URL can appear in several answers; summing those occurrences does not tell you how many distinct sources there were. A citation also shows what the answer linked to, not the complete set of pages retrieved, the model's training data, or why the system selected a brand.
The question about making an AI animated video without editing every scene had no StrokeVideo mention or own-site citation in its three samples. Stored sources included other video products. That makes it a useful question to investigate, but it does not prove that StrokeVideo should rank for every interpretation of “AI animated video.”
Before writing, check whether the product actually meets the task. If it does, create a page that explains the workflow, inputs, output and limits, with evidence a reader can verify. If it does not, choose a narrower question rather than invent a feature. Review the current sources for gaps, not sentences to copy.
Then record the publication date and repeat the same questions with comparable settings. A later improvement would be an observation to examine alongside other changes, not automatic proof that the new page caused it. Content Agent can help draft from a gap; human review and publishing remain necessary.
No. Eight of fifteen saved answers is about 53% of this sample. It is not user reach, market share or traffic attribution.
No. The links are evidence of citations in the saved answers. They do not identify training sources or the full retrieval process.
Start with a small set of customer questions, keep the original answers and sources, and compare mentions and citations separately. Explore AI Visibility or follow the step-by-step monitoring guide.