Each export holds a different slice of your audit, and no single one holds everything. Start from what you want and the sections below point you to the right file.
I want to see the persona descriptions
Use the JSON export. It is the only export that contains the full persona descriptions, the paragraph of detail that tells the AI models who your buyer is. Every other export gives you persona names only.
This is the file to use if you want to review or check exactly what Gumshoe is telling the models about your audience.
I want the specific products the models named
Use the conversation export CSV. Its Product column holds the specific product or model named in each answer, which no other export includes.
I want the URLs where my brand was recommended
Use the JSON export. It carries the full URL attached to each brand mention, which is the same data you see in the Mention breakdown tab. The conversation export CSV gives you the brand's domain for each mention but not the full URL.
These are not the same as citations. A recommendation URL is where a model pointed when it named a brand. A citation URL is a page a model read in order to build its answer.
I want the source URLs the models cited
Use the JSON export or the sources export CSV. The JSON nests citations under each answer with the raw URL, the normalized URL, and the domain. The sources export flattens the same information into one row per citation, which is easier to work with in a spreadsheet.
The PDF shows the most-cited domains but not the individual page URLs.
I want visibility percentages, trends, or comparison to the last run
Use the PDF. It is the only export with calculated figures: your overall visibility percentage, the competitor leaderboard, the persona, topic and model breakdowns, the change since your previous run, and the trend across recent runs.
The data exports contain the underlying answers rather than the calculations, so if you want the numbers you see on screen, take the PDF.
What is in each export?
JSON export: the most complete data file. Audit and run details, all personas with their full descriptions, every prompt with its topics, and every answer with the brands mentioned, their ranks and URLs, the reasoning, and the citations.
Conversation export CSV: one row per brand mention. Prompt, model, rank, brand name and domain, the product named, the reasoning text, the persona, and the related topics.
Prompts export CSV: one row per prompt. Shows which topics apply to each prompt and your best rank for that prompt on each model. A coverage map rather than content.
Sources export CSV: one row per citation. The URL, its domain, and the model, persona, question and topics it came from.
PDF: the visual summary. Every calculated view, including the trend and the run-over-run changes.
Which export has what?
| What you want | JSON | Conversation CSV | Prompts CSV | Sources CSV | |
| Persona names | Yes | Yes | Yes | Yes | Yes |
| Persona descriptions | Yes | No | No | No | No |
| Prompt text | Yes | Yes | Yes | Yes | No |
| Answer text | Yes | Yes | No | No | No |
| Competitor brands and ranks | Yes | Yes | No | No | Summary only |
| Product or SKU named | No | Yes | No | No | No |
| Recommendation URLs | Yes | Domain only | No | No | No |
| Citation URLs | Yes | No | No | Yes | Domains only |
| Visibility percentages | No | No | No | No | Yes |
| Change since last run | No | No | No | No | Yes |
| Trend across runs | No | No | No | No | Yes |
A note on building your own reporting
No single export carries both the calculated visibility numbers and the underlying answer text. If you are building your own dashboard, take the answers from the JSON or conversation export and calculate visibility from them, or use the PDF for the finished figures.