Most marketers assume an AI mention is a win. Semrush’s new walkthrough shows why that assumption breaks: on the company’s test site, a single guide was cited in answers for 50 prompts and earned exactly zero clicks from Google.
That gap is the entire case for an AI visibility dashboard. You put prompt-level AI data from Semrush next to page-level Google Search Console data, then ask a simple question: are the pages AI cites actually earning traffic?
What the dashboard actually tracks
Instead of one blended visibility score, the dashboard keeps two data sources separate. It tracks AI visibility rate, mentions and cited pages from prompt tracking, generative AI impressions from Search Console, and the standard clicks, impressions, CTR and average position.
It also flags answer engine optimization opportunities: pages AI trusts but Google ignores, prompts where competitors appear instead of you, and prompt groups where you have no presence at all.
Build a prompt set that sounds like buyers
The prompt list is the foundation. Semrush recommends around 30 prompts, spread evenly across five journey stages:
- Customer problems
- Product or category discovery
- Comparisons
- Product selection
- Branded questions
Keep that prompt set unchanged for at least a month. Trend lines only mean something if the inputs stay constant. Tag each prompt by group so you can spot segment-level changes instead of reading one blended number.
Semrush Prompt Tracking supports ChatGPT, Google AI Mode and Gemini. Branded prompts will usually appear, so report the non-branded visibility rate separately to see whether AI recommends you to people who don’t know you yet.
Pull the right Google Search Console data
If your property has generative AI data, export the Generative AI performance report twice: a three-month daily trend and a 28-day pages table. Impressions are currently the only metric available there. If the report is empty, fall back to the standard Search results report and compare two matched 28-day windows ending two to three days before today, so incomplete data doesn’t undercount performance.
Connect the sheet and let Claude build the artifact
The workflow is a master Google Sheet plus a paid Claude plan. You export Semrush prompt-level and cited-page data into the sheet, add GSC exports, and connect Google Drive inside Claude. Then ask Claude to build a live dashboard that re-reads the sheet on demand. The method took about 90 minutes to set up and roughly 20 minutes per week to refresh.
Keep headers identical, stamp every export with its snapshot date, append rows rather than replacing them, and normalize URLs across tabs. URL variants with and without trailing slashes are the classic reason page-level numbers look absurdly low.
Read the gaps, then act
On the test site, the first refresh produced three actions:
- Page cited for 50 prompts, zero clicks: AI already trusts it, so the work is on the Google side—refresh around a realistic keyword and build links.
- Competitors cited where you are not: Open those prompts, read what gets cited, and create the comparison or category content that is missing.
- Whole prompt group with no presence: That signals missing top-of-funnel content, so use those prompts as content briefs.
Over time, watch for pages that rank in Google but rarely get cited—they usually need clearer structure, direct answers and current dates. The dashboard turns a fuzzy AI visibility gap into a weekly prioritization list.
Source: Semrush Blog
