AI Visibility (prompt tracking)

AI Visibility asks ChatGPT, Claude, Gemini and Perplexity the questions your buyers ask, on a schedule, and records what they answer. For every answer it notes whether your brand was named, where it came in the list, which competitors were named, and which pages the answer cited. Over time that gives you a trend you can act on.

Open it from a project's sidebar under My Site → AI Visibility, or use the MCP tools from an agent.

How it works

1. Set up tracking. Enter your brand name, plus any other names that should count (product names, an old name). Your project's domain always counts. Pick the models to ask and how often: daily, weekly, monthly, or only when you run it. You can also choose the country the models search from.
2. Add prompts, or let PandaCrawl suggest them. Press Suggest prompts for candidates you can tick, each shown with why it is suggested (see Suggested prompts). Or write the questions the way a buyer would type them into an assistant, for example "What is the best vitamin C serum for aging skin?". A project can track up to 50.
3. Runs. Each run sends every prompt to every chosen model with web search on. Scheduled runs start on their own; Run now starts one immediately after showing the estimated cost.
4. Reading the answers. Each answer is checked against your brand and every competitor in the project's competitor list (the same list your agents fill in through project context). Names match as whole words, ignoring case, so "Acme" does not match "Acmecorp". A name that is also an everyday word (say, a brand called "Gap") will also match that word, so use the most distinctive name the brand goes by.

Suggested prompts

The prompt list decides whether the numbers mean anything, so suggestions come from evidence first, strongest first:

1. AI answers. Questions ChatGPT already answers about your first two competitors, from DataForSEO's LLM Mentions data, with how often people ask them. These are the prompts your competitors own. About $0.20 of DataForSEO per suggestion; you can switch it off.
2. Your searches. Question-like searches from your Search Console data (last 3 months, your brand's own name excluded), rewritten as the question a buyer would ask an assistant. Shown with their impressions.
3. Written by AI. Questions PandaCrawl's own model writes from your project description, positioning and competitors, to cover what the data does not. This runs on PandaCrawl's model, not your keys.

Nothing is tracked until you tick it, because every prompt is asked on every run. The top ten are ticked for review. If a source is missing (no competitors, Search Console not connected), the list says what to add for better suggestions.

What the numbers mean

Prompts that no model answers with your name are content gaps. The usual fixes are a page that answers the question directly, and getting mentioned on the third-party sources the models already cite for it.

Click any cell in the prompts table to read the full answer, the brands it named in order, and its sources.

Cost

Each answer is one DataForSEO LLM Responses call with web search on. On a bring-your-own-key workspace it is billed to your own DataForSEO account; on a hosted plan it draws from your credits. What one answer cost on real runs:

A run costs prompts × the sum of the chosen models' costs. All four models come to about $0.15 per prompt per run, so 10 prompts cost about $1.50 per run: about $6 a month weekly, or $45 a month daily. Perplexity alone is a quarter of the price of ChatGPT or Claude.

The setup form shows each model's price and what your prompts will cost per run and per month on the schedule you pick, before you save. Estimates use typical prices until your first run, then your project's own average for each model.

Runs stop straight away with a clear message if your DataForSEO key is missing or your balance runs out, rather than failing every prompt. When one model fails, the prompts table shows why (hover failed), and the others still count.

Copilot and Bing AI citations

Bing Webmaster Tools' AI Performance report counts how often Microsoft Copilot and Bing's AI answers cite your pages. Bing has no API for it yet, so PandaCrawl reads the exports: open AI Performance → Copilot citations in your project and upload the CSV files.

1. In Bing Webmaster Tools, open AI Performance for your site and pick the date range.
2. Export the chart, the Grounding queries table and the Pages table as CSV.
3. Upload all three together. Upload again whenever you like: days that overlap are updated, not counted twice, so the trend keeps growing.

The page shows citations for the last 28 days against the 28 before, a daily chart, your most-cited pages, and the grounding queries: the searches Copilot ran to find sources (not what people typed), with each query's intent, topic and your share, your percentage of all the citations Copilot gave for that query. Bing samples this data and reports no clicks. Uploading and reading it is free.

MCP tools

A typical agent flow: setup_prompt_tracking → suggest_tracked_prompts → add_tracked_prompts with the chosen ones → get_ai_visibility for the estimate → confirm with the user → run_prompt_tracking → poll get_ai_visibility until latestRun.status is completed.