A guide on a Semrush test site earned AI citations across 50 tracked prompts — and zero Google clicks. That single finding, surfaced by the first refresh of a custom AI visibility dashboard, shows exactly why the company now recommends pairing its AI Visibility Toolkit data with Google Search Console in one reporting view.

Semrush's walkthrough, published on its own site, describes a dashboard built from four components: the AI Visibility Toolkit, Google Search Console, Google Sheets, and a paid Claude plan. The author reports the first build took about 90 minutes, with weekly updates running 20 minutes each.

What does the dashboard actually measure?

The setup places two datasets side by side: how visible a brand is in AI answers, and how the same pages perform in Google. Specifically, it tracks:

• AI visibility rate — the share of tracked prompts where the brand appears
• Mentions and cited pages from Semrush's prompt tracking
• Generative AI impressions from GSC's Generative AI performance report
• Clicks, impressions, CTR, and average position from standard GSC reports
• SEO and answer engine optimization (AEO) opportunities generated from the combined data

The dashboard runs as a Claude artifact that pulls live data from a Google Sheet via the Google Drive connector. Two constraints apply: artifacts calling connectors require a paid Claude plan (Pro, Max, Team, or Enterprise), and the dashboard cannot be shared by public link. For stakeholder reporting, Semrush suggests asking Claude to export the report as an HTML file.

How many prompts do you need?

The author tracked 30 prompts, split into five groups of six: customer problems, product or category discovery, comparisons, product selection, and branded questions. Monitored daily, 30 prompts produce roughly 200 observations per week — enough, per Semrush, for a stable site-wide visibility rate. The entry-level AI Visibility Toolkit plan covers only 25 prompts, so 30 requires a higher tier or an add-on.

Group-level numbers need careful reading. With six prompts per group, a single prompt changing shifts that group's rate by 17 percentage points. Semrush recommends expanding each group to 15 or more prompts for dependable segment reporting, and reporting both total and non-branded visibility separately, since branded prompts almost always surface the brand and inflate the overall rate.

Prompt Tracking supports ChatGPT, Google AI Mode, and Gemini. Neither prompt-level performance nor cited-page data is available through the Semrush API yet; both reach Google Sheets via manual export or scheduled email delivery.

What did the test site's first refresh reveal?

Three findings went straight onto the content plan:

• Cited but not clicked: an executor checklist cited for 50 prompts with zero Google clicks — a Google-side problem, not an AI problem, per the analysis. The recommended fix: find a realistic keyword, refresh the page around it, and build links.
• Competitors cited where the brand isn't: rival domains led with citations for seven prompts each, defining a list of comparison and category content to build.
• Zero visibility across an entire prompt group: all six customer-problem prompts showed no presence, which Semrush reads as missing content rather than one weak page — those prompts became briefs for new top-of-funnel pieces.

One structural point stands out for teams tracking ChatGPT specifically: a page can rank well in Google and still never appear in ChatGPT answers, because ChatGPT draws from its own web index. Bing and Copilot coverage requires adding Semrush's Bing Position Tracking export as an extra data source.

What are the GSC limitations?

Google's Generative AI performance report, available to every property but populated only once pages accumulate enough AI Overviews and AI Mode impressions, currently exposes impressions only — no queries, clicks, or CTR, and no prompt-to-click attribution. Properties with empty reports should fall back on standard Search results exports. Semrush advises ending every export window 2-3 days before the current date, since the newest GSC data is incomplete.

Weekly updates follow a fixed routine: export fresh prompt tracking data, cited pages, and GSC data; append rows with snapshot dates rather than replacing them; and note content changes so movements can be traced to causes. URL normalization matters — GSC can report the same page twice with and without a trailing slash, and Semrush flags URL variants as the first suspect whenever page-level numbers look implausibly low.

As prompt sets stabilize over months and Generative AI reporting in GSC matures, the dashboard's comparison history — especially topics improving in both AI answers and Google search — becomes the clearest evidence of which page structures actually earn citations.