Compare AI citation exports
Check whether an AI visibility report connects each cited page to the prompt, engine and run.
One question on two engines
On Nov 19, 2026, we sampled this question through ChatGPT web search and Perplexity Sonar API.
Which AI search visibility tools should a US agency compare for 10 clients, with separate workspaces and citation exports?
| Engine surface | Source hostnames | Source pages |
|---|---|---|
| ChatGPT web search | 6 | 8 |
| Perplexity Sonar API | 10 | 10 |
The provider records shared 1 source page by hostname and path after removing query strings, fragments and common URL prefixes. This sample records provider-reported sources from two answers. A new run may return different results.
semrush.com/blog/ai-visibility-tracking-tools-for-agencies
Which columns to request
Require a run ID, exact prompt, timestamp with timezone, model or engine surface, language, market and execution status. Keep the response text or a durable link to it. A source list should point back to that same run and preserve the original URL. Keep a second field for your chosen normalization rule.
Decide how to distinguish a named brand, a page cited in the answer and a page retrieved during search. Export a period that includes a known failed run or an empty answer. Compare attempted runs with completed runs before calculating rates.
How the documented exports differ
Promptwatch lists response text, model, brand-mention and citation URL fields for its response export. Its documentation sets a 10,000-row cap and lists plan restrictions. Peec documents exact prompt text in filtered chat exports and a separate URL view with country and model filters. OtterlyAI lists prompt and citation data exports. Ask each vendor for sample files from the tier under consideration.
- Promptwatch fields and export limits
- Peec chat export and URL detail changes
- OtterlyAI reporting and export features