A lab card for the AI question panel: same conditions, every wave

One reading is an anecdote. Two, taken the same way, are a trend.

Does an AI assistant name EKU when the buyer describes the problem?

Prompts locked 3 August 2026 · first wave planned for October

37prompts: 25 base, 12 Maximus

3assistants, one new account each

116runs: 111 plus 5 stability re-runs

3 h 15of focused time, in one or two days

What you are seeingA lab protocol. The value of the wave is not the number it produces; it is that wave 2, in March, can be compared with it. Anything that changes between waves (account, wording, a follow-up question) breaks the comparison.

Before the first run

Create one new account per assistant with a neutral email (never a personal one). Memory and personalization off.
Same network and location for the whole wave. Write down the model name shown that day for each assistant.
Open the tracker workbook, sheet CAPTURE LOG. Only the yellow cells are filled.
Create a folder named wave1_YYYY-MM-DD for screenshots.

The run order

Run	Prompt	Assistant	Rule	Screenshot name
1	P01	Assistant A	new chat, paste the exact prompt, no follow-up	`PP01_A.png`
2	P01	Assistant B	new chat, paste the exact prompt, no follow-up	`PP01_B.png`
3	P01	Assistant C	new chat, paste the exact prompt, no follow-up	`PP01_C.png`
4	P02	Assistant A	new chat, paste the exact prompt, no follow-up	`PP02_A.png`
5	P02	Assistant B	new chat, paste the exact prompt, no follow-up	`PP02_B.png`
6	P02	Assistant C	new chat, paste the exact prompt, no follow-up	`PP02_C.png`
7	P03	Assistant A	new chat, paste the exact prompt, no follow-up	`PP03_A.png`
8	P03	Assistant B	new chat, paste the exact prompt, no follow-up	`PP03_B.png`
9	P03	Assistant C	new chat, paste the exact prompt, no follow-up	`PP03_C.png`
10	P04	Assistant A	new chat, paste the exact prompt, no follow-up	`PP04_A.png`
11	P04	Assistant B	new chat, paste the exact prompt, no follow-up	`PP04_B.png`
12	P04	Assistant C	new chat, paste the exact prompt, no follow-up	`PP04_C.png`
…	…	…	the full order of 116 runs is in the tracker workbook, sheet Run order	…
112	P05	Assistant A	stability re-run, same conditions	`SP05_A.png`
113	P08	Assistant A	stability re-run, same conditions	`SP08_A.png`
114	P13	Assistant A	stability re-run, same conditions	`SP13_A.png`
115	M01	Assistant A	stability re-run, same conditions	`SM01_A.png`
116	M07	Assistant A	stability re-run, same conditions	`SM07_A.png`

Rule for every run: new chat, paste the exact prompt from the workbook, read the answer, fill the row, screenshot if EKU appears, close the chat. Zero follow-up questions: asking "and EKU?" contaminates the run.

What to record per run

Field	How to fill it
mentions_eku	yes or no
mentions_maximus	yes or no
position	1, 2 or 3+ when EKU appears
competitors_mentioned	names, in the order the answer gives them
sentiment	positive, neutral or negative about EKU, if mentioned
cites_sources	yes or no; cited_domains: list them
screenshot	file name, only when EKU appears
notes	anything odd: a refusal, a product mix-up, a retired term

The report, one page, the same day

Share of visibility in this set: runs that mention EKU divided by 111 (never called market share).
Maximus named: runs that mention Maximus (expected near 0: that is an honest baseline).
Average position when EKU appears; top competitors named; domains cited (that list is the target for citations and press).
Stability: write down how many of the five re-runs changed and how. Treat a small swing between waves with care when several did.
Mandatory caveat, verbatim: "Fixed buyer questions, run by hand under controlled conditions, three assistants, one date. Specific to this set and moment; not market share and not real demand. Its value is the like-for-like comparison between waves."

Tracker workbook Baseline table

Closing

Strategic conclusion

The July AI panel is a single reading and is not repeated: its instrument stays with the firm that measured. From October, EKU's own measure is the locked question series, a separate instrument that starts its own trend. Only later waves of that same series compare with wave 1.

Commercial implication

The first like-for-like comparison of AI visibility across assistants arrives with wave 2, after the first pages are published.

Recommended use

Marketing lead runs it in October (the "Measure" line); the analytics owner files the report next to the baseline table.

NextThe numbers tableWhere the wave-1 result is filed.

Validation status. Protocol locked 3 Aug 2026; run order generated 30 Sep 2026. A first automated reading on one assistant was taken 1 to 2 Oct 2026 with Kanbina Lumina's tool; the manual wave on three assistants is still to run.

Working foundation for executive review. Not approved external copy. Prepared for EKU Power Drives, 30 September 2026, by KanbinaLUMINA