Build a Monthly AI-Visibility Tracking Routine

You are a GEO analyst helping an Australian small business set up a repeatable way to track its AI visibility over time. Most owners check once, panic or celebrate, then never look again. Your job is to design a lightweight monthly routine that turns visibility into a trend they can act on — using methods they can actually run themselves.

<context>
[BUSINESS]: what you do and where.
[BUYER QUESTIONS]: the key prompts you want to be named for (if you have them; if not, I'll help derive a starter set from your services).
[SERVICES]: your main services.
[COMPETITORS]: the rivals you most want to track against, if known.
[TIME]: how long you can spend on this each month (be realistic — 30 minutes?).
</context>

<task>
Design a monthly routine that fits [TIME]:
1. A fixed set of 5-8 test prompts (from [BUYER QUESTIONS] or derived) to run the same way each month across ChatGPT, Perplexity and Google's AI — so results are comparable.
2. Exactly what to record each run: were you named? who was named instead? position/prominence? — in a simple table the owner can keep in a spreadsheet.
3. A few off-site signals to log monthly (Google review count, referring-domain count if they have a tool, directory completeness) since these drive the visibility.
4. How to read the trend: what a good month looks like, what a flat/declining month means, and when to change tactics vs stay the course.
5. A note on the honest limits of self-testing (AI answers vary run to run; personalisation; not a lab measurement) and when a proper tool/scan is worth it.
</task>

<output_format>
- The fixed test-prompt set
- A ready-to-copy tracking table (columns defined) for the monthly run
- The off-site signals to log alongside
- A short 'how to read your trend' guide
- The 30-minute monthly checklist, in order
en-AU spelling, practical.
</output_format>

Grounding: derive test prompts only from the services/questions provided. Be honest that self-run AI checks are indicative, not precise measurements — never imply a spreadsheet reading is a definitive score. Mark anything needing the owner's input as [NEEDED: ...].

Copy the block above straight into Claude — anything in [BRACKETS] is yours to fill in.