Module 03 · Lesson 2 of 4
Running a Repeatable Query Set
The weekly run is the least glamorous and most valuable habit in this course. Same queries, same order, same conditions, logged every time. It converts AI search from anecdote into data, and it takes under an hour for a thirty-query set.
What to record per query
| Column | What goes in it |
|---|---|
| Date and run number | When you ran it; runs numbered so runs are comparable |
| Query / prompt | The exact string, never paraphrased mid-run |
| Surface | AI Overview / AI Mode / assistant / classic only |
| Brand mentioned? | Yes / no for your brand, and how (cited link, plain mention) |
| Cited domains | Every domain credited by the answer |
| Competitors present | Which named competitors appear |
| Classic position | Your rank in traditional results, same session |
| Notes | Anything odd: different format, new surface, answer changed shape |
The distinction between a mention and a citation earns its own row. A citation is a displayed link; a plain mention is your brand named without a link. Both matter, differently: citations drive clicks and are easier to verify, mentions build the entity association you will work on in Module 5. Log both, label them separately.
Consistency rules
- Same assistant accounts, same sign-in state, week to week. Pick one state and freeze it.
- Same location settings, or honestly the same physical location if you cannot control it.
- Same browser profile, ideally one used for nothing else.
- Same order of queries; fatigue effects are real and at least they will be constant.
- Fixed day and time window; put it in the calendar as a recurring block.
You will not achieve laboratory conditions. You do not need them. You need your noise to be roughly constant so that changes in the numbers mean something.
How often, and for how long
Weekly is the working cadence for most sites. Daily burns a day a week and adds little signal; monthly is too slow to catch AI search, which moves quickly. Run for at least four weeks before drawing any conclusion: with high run-to-run variability, single-week deltas are noise until a direction repeats.
| Run | Query | Surface | Mentioned | Cited domains | Classic |
|---|---|---|---|---|---|
| 4 · 21 Oct | best ai seo course | AI Overview | No | publisher-A, publisher-B, vendor-C | Position 9 |
| 4 · 21 Oct | how to track ai citations | Assistant | Yes, no link | tool-vendor-D, community-E | Position 4 |
| 4 · 21 Oct | geo vs seo difference | Classic only | No | n/a | Position 12 |
Fake data, real shape. After four runs, row two's plain mention becomes a trend worth acting on: the entity is known, the citation is not yet earned.
- Set up the log with the eight columns above. A spreadsheet is fine.
- Run your full starred set from Workbench 3.1 once, this week, under conditions you can repeat.
- Book a recurring weekly slot for the run. Same day, same time window.
- After run four, look only for repeats: queries where the same story appeared three or four times. Those repeats are your first findings.
Why is the run, not the answer, the unit of measurement?
What is the difference between a mention and a citation?
Your AI presence doubled in week two. Cause for celebration?
Sources used in this lesson
Google Search Central: AI features and your website