Case study and method
AI visibility test: how often AI named a UK NetSuite consultancy
A UK consultancy that does NetSuite project rescue was named in 8 of 88 AI answers, all from ChatGPT. Its two most-named rivals were each named in 48.
Captured on 14 and 15 October 2026. The firm and its rivals are anonymised.
Method
Five buyer-style prompts, each asked word for word in a fresh chat, three runs per engine where access allowed. The Prompt Pack method uses five engines: 5 prompts × 5 engines × 3 runs = 75 answers.
The five prompts
- Which independent firms can help a UK finance team rescue a troubled NetSuite implementation?
- What does a NetSuite project-rescue assessment cover, and who provides this service in the UK?
- Which independent NetSuite consultants can work alongside our existing implementation partner to recover a struggling project?
- What should a UK finance director do when important NetSuite requirements keep moving into phase two, and which firms can help?
- Who can independently assess whether a NetSuite recovery plan will meet our finance reporting and control needs?
Why 88, not 75
This test added two extra Duck.ai models, so 105 answers were planned. Two engines stopped early: Perplexity gave 5 of 15 before asking for sign-up, and one Duck.ai model gave 8 of 15 before a daily limit. That left 88 captured: 65 from the pack’s five engines and 23 from the two extra models. Missing runs were recorded as barriers, not guessed.
| Engine, as tested | Planned | Captured | Named | Cited |
|---|---|---|---|---|
| ChatGPT (signed out, default model, web search) | 15 | 15 | 8 | 6 |
| Google AI Overviews (signed out) | 15 | 15 | 0 | 0 |
| Gemini (Flash-Lite, signed out) | 15 | 15 | 0 | 0 |
| Microsoft Copilot (signed in, work account) | 15 | 15 | 0 | 0 |
| Perplexity (guest) | 15 | 5 | 0 | 0 |
| Duck.ai, GPT-6 Luna (guest) * | 15 | 15 | 0 | 0 |
| Duck.ai, GPT-5.4 mini (guest) * | 15 | 8 | 0 | 0 |
| Total | 105 | 88 | 8 | 6 |
* Extra model, not part of the pack’s five engines. ChatGPT also recommended the firm in 8 of its 15 answers; no other engine did.
Results by prompt
| Prompt | ChatGPT runs naming the firm |
|---|---|
| Q1 | 3 of 3 |
| Q2 | 0 of 3 |
| Q3 | 3 of 3 |
| Q4 | 2 of 3 |
| Q5 | 0 of 3 |
For Q2 and Q5, no engine named the firm. Neither question had a page on its site that answered it.
Rivals
The two most-named rivals were each named in 48 of the 88 answers, and cited in 30 and 39. A third was named in 35 and cited in 33.
Definitions
- Named
- The firm’s name, or a clear variant, appears in the answer.
- Cited
- A link to the firm’s own site appears in the answer’s sources. A “website” button that only opens a Google search does not count.
- Recommended
- The answer shortlists or recommends the firm. A neutral mention is named, not recommended.
- Rivals
- Every other firm named in the answer.
What this does and doesn’t show
- It is one firm, five prompts and two days. These are counts of answers, not market share or a chance of winning work.
- Not being named is not proof of being invisible. It is evidence for those dates, engines and questions.
- Copilot was signed in to a work account, which can shape answers.
- Some captures don’t show every source link, so cited counts can be understated.
- The firm was advised to write pages for Q2 and Q5, then re-test in six to eight weeks. That re-test has not happened yet.
FAQ
Why were 88 answers captured, not 75?
The test added two extra Duck.ai models, so 105 were planned. Perplexity stopped after 5 of 15 and one Duck.ai model after 8 of 15, leaving 88.
Does 8 of 88 mean the firm has a 9% share of AI recommendations?
No. These are counts of answers to five questions on two days, not market share or a chance of winning work.
Were the firm and its rivals real?
Yes. They are anonymised here.
Run the same test on your firm
The ERP Buyer Prompt Pack has the prompts, scoring rules and a scoreboard that tallies itself. It is also included in Found, the ERP SEO playbook, as chapter 8.