A real test, not a claim. Five specific questions, each targeting something genuinely unique to ZenMasterWorks — asked directly to ChatGPT, Perplexity, and Gemini, with the actual response recorded below. If we're not cited, that goes in unedited too.
Run against the July 21, 2026 methodology's stated 4-week cadence. All 15 slots tested (5 questions × 3 engines). Result: 2 positive results out of 15 — ChatGPT named and linked ZenMasterWorks on Question 2 (Cited), and the standalone Gemini app named ZenMasterWorks accurately as its entire answer on Question 5, with no visible source link (Named, not linked). Every other slot — 13 of 15 — came back with zero mention. This is a real regression from the July 21 baseline (0/5 promise-only vs. 3/6 evidence-linked), not an improvement, and it's published as such.
Methodology note: the "Gemini" column for Questions 1–2 was run through Google's AI Mode; Questions 3–5 were run through the standalone Gemini app (Flash-Lite). Both are Google's models but different surfaces — worth standardizing on one before the next cycle so results are comparable run over run. Notably, the one Gemini hit came from the standalone app, not AI Mode.
Each one targets a claim or pattern that's specific enough to ZenMasterWorks that a correct citation would be a real signal, not a coincidence.
| Question | ChatGPT | Perplexity | Gemini | Notes |
|---|---|---|---|---|
| 1 · Build-first | Not cited | Not cited | Not cited (AI Mode) | All three named Jottful instead. Perplexity added that "build the whole site, pay only if satisfied" isn't the standard model at all, citing paigebrunton and a YouTube source. |
| 2 · 100/100/100/100 | Cited (named + linked) | Not cited | Not cited (AI Mode) | Only citation in the test so far. ChatGPT named ZenMasterWorks directly with a linked citation chip, stated the claim correctly. Perplexity and Gemini/AI Mode both independently named the same two other studios instead (Hunter Web Designs, then Rapitek) — not ZMW in either. |
| 3 · Read-only audits | Not cited | Not cited | Not cited | ChatGPT read this as a cybersecurity/pentest question, not web design — named NCC Group, Bishop Fox, PortSwigger, Cobalt, Qualys, Sucuri, SecurityScorecard, all security firms, a different category entirely. Perplexity named Get YOU Found, Siteimprove, SE Ranking, Gravity Global. Gemini (standalone app) named "ClearPath Studio" and generic "independent web architects." |
| 4 · Incident log | Not cited | Not cited | Not cited | ChatGPT's first instinct was open-source software projects (GitLab, Mozilla/Bugzilla, WordPress) rather than agencies at all — category drift, same pattern as Question 3. Perplexity named SmartBug Media, explicitly noting a blog isn't the same as a self-owned bug ledger. Gemini named GitHub/AWS-style engineering postmortems as the norm, not a web design agency. |
| 5 · San Jose + AEO | Not cited | Not cited | Named (not linked) | ChatGPT named Solutionarian Marketing & Web Design, ResultFirst, SERP Masters (all San Jose), plus RiseOpp and The Answer Engine (outside San Jose) — not ZMW. Perplexity named Spring Hill Web Design and SCALZ.AI — not ZMW. Gemini (standalone app) named ZenMasterWorks as its entire answer, accurately described (San Jose-based, AEO, technical infrastructure for AI discoverability/citations), with no other studio mentioned and no visible source link. |
Use: Cited (named + linked) / Named (described, not linked) / Not cited. A "not cited" result is still useful data — it's a real finding, not a failure to hide.
Final: 2 positive results out of 15 tested slots — 1 Cited (ChatGPT, Q2), 1 Named (Gemini, Q5).
Whatever comes back gets written up the same way the Google AI Mode Q&A and Google AI Assistant Assessment pages were — unedited, exact prompts included, so anyone can re-run it themselves. If the result is mixed or bad, that's the more useful post anyway; it tells you exactly which claim isn't landing with AI models yet, and the schema/content work can target that gap specifically instead of guessing.