Track Gemini visibility with a monthly routine

#01What goes in the prompt set?

Keep prompts in natural language, as buyers type them. See Building a prompt set and Building your first prompt set.

#02How do you sample each surface?

The grounded API response lists the search queries the model ran (the google_search_call step) and each cited URL (url_citation annotations), which is a useful view of fan-out. Grounding on Gemini 3 and later models includes 5,000 free search queries a month, then $14 per 1,000 queries. One prompt can trigger several queries, so estimate before you schedule runs. For automation, see Automate Gemini marketing workflows with n8n.

A minimal logging call:

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" -H "Content-Type: application/json" \
  -H "Api-Revision: 2026-05-20" \
  -d '{"model":"gemini-3.8-flash",
       "input":"best AP automation software for mid-sized companies",
       "tools":[{"type":"google_search"}]}'

#03How do you score answers?

Metrics per surface:

• Mention rate = prompts mentioned / prompts run.
• Link rate = prompts linked / prompts run.
• Share of voice = your mentions / all mentions of you and Competitors A-D.
• Accuracy rate = correct descriptions / mentions.

Run API prompts three times and treat a mention as stable at two or more of three runs. See Statistical confidence in AI visibility tracking.

#04What's the monthly calendar?

#05What goes in the one-page report?

Our view: label each number as measured, sampled or estimated. Leadership trusts the report more when the uncertainty is visible, and it stops month-to-month noise in sampled data from being read as a trend.

#06Worked example: Ledgerline

Ledgerline is a fictional accounts-payable software company. Details are illustrative.

Walkthrough

1. The SEO lead ran 32 prompts on each surface and 96 API runs (32 x 3). She estimated grounding queries in advance to stay within the free monthly allowance.
2. The league table showed a review site and Competitor A's integration hub as the most-cited domains.
3. Accuracy problems in the Gemini app traced back to an outdated pricing description on a review site.
4. Search Console showed Generative AI impressions up 18% and GA4 showed 410 Gemini sessions, both reported as measured.
5. Five actions went to owners: update the review-site profile, publish two integration pages, refresh the benchmark, fix product markup, and pitch one cited publisher.

#07Common mistakes

• Changing the prompt set every month.
• Treating API results as the consumer app.
• One run per prompt.
• Reporting only your own mention rate.
• Long action lists that nobody finishes.

#08Try it yourself

Write 25 prompts across the five stages. Run ten of them today in the Gemini app, AI Mode and Google Search, and score them with the fields above. Note the three most-cited domains you don't own.

#09Quick quiz

1. Why run API prompts three times?

Answer: Answers vary, so a repeated mention is more reliable than a single run.

2. What does a grounded API response give you?

Answer: The search queries the model ran and the URLs it cited.

3. How should report numbers be labelled?

Answer: As measured, sampled or estimated.

#10What should you check before the next lesson?

• The prompt set is locked for the quarter.
• The scoring sheet and league table are set up.
• The first one-page report has five owned actions.

Previous: Search Console AI reports for AI Mode and AI Overviews. Or return to the Gemini for Marketers track. Related: Perplexity monthly routine and governance, Tracking competitors in AI search, Monthly AI visibility report template.

#11Common questions

How many prompts should I track? Start with 25-40 buyer prompts grouped by stage and topic. Test each in the Gemini app, AI Mode and Google Search, and log the same prompts monthly.

Can I track AI Overviews impressions in Search Console? Yes. The Generative AI performance report shows impressions for AI Overviews and AI Mode combined, by page, country, device and date.

Why do my results change every time? AI answers vary by model, location, personalisation and time. Track patterns across a fixed prompt set and several runs rather than single answers.