AI citations, with the evidence.

What did the answers show?

These are the six completed observations, including answers that did not cite Elite. A citation means a link to our site inside the generated answer. A separate Sources panel does not count as an answer citation.

Unanswered and unrun checks

Gemini returned an error for the charity website prompt. A second prompt could not be submitted. ChatGPT and Copilot were not run because the available sessions were signed out or at a sign-in page. The Ad Grants and clinic-brief prompts were not run in Google. These gaps are excluded from answered-query counts; they are not evidence of absence.

Which exact prompts did we use?

The wording is preserved below, including an ambiguity revealed by the consultant question. We did not mention Elite in any prompt or ask an engine to recommend us.

1. Charity website providers

   Who can build a donation website for a small UK charity? Compare agencies with published prices and link to your sources.

2. Private consultant website costs

   How much does a private consultant website cost in the UK? Compare providers with published prices and link to your sources.

3. Charity Ad Grants management

   Which UK agencies help small charities manage Google Ad Grants, and what fees do they publish? Link to your sources.

4. Private clinic website planning

   What should a UK private clinic include in a website brief before hiring an agency? Cite useful planning resources.

How were these observations recorded?

We inspected the visible answers and source links manually through the services’ web interfaces on 6 October 2026, using a browser session in Indonesia and UK-focused wording. This was a convenience pilot, with one observation per engine–prompt pair. It was not a UK-location test, a controlled experiment or an independent audit.

Perplexity

Signed-in account; Incognito conversation; Search enabled; default UI, underlying model not reported. Fresh conversation for each prompt.

Google AI Overviews

Standard Google Search with an AI Overview displayed; signed-in account; results labelled not personalised; country shown as Indonesia. Search location was not controlled to the UK.

Gemini

Signed-in account; Temporary chat; Flash. The charity query returned an error. A second prompt did not submit and is not treated as an answered observation.

ChatGPT

Not run: the available session was signed out.

Microsoft Copilot

Not run: the available session reached a sign-in landing page.

Scoring: “brand mentioned” records the business name in answer text; “answer citation” records an Elite hyperlink in the answer; “Sources panel only” records retrieval without an answer citation. An error has unknown values, not a negative score. The downloadable log keeps these fields separate and preserves the exact prompts and session conditions.

Limits: accounts, location, wording, model updates and timing can change answers. The prompts cover different intents, the engines have unequal coverage, and there are no repeat observations. We therefore do not report a visibility percentage, rank engines against each other or infer traffic, enquiries or sales. Raw captures are retained privately because they include signed-in interface details; the public log contains the relevant observations and URLs.

Disclosure: Elite Digital Agency performed and publishes this AI-assisted review of its own visibility. It is a working baseline with an editorial owner, not third-party validation. Report a correction with the observation ID and source.

What did we learn from the pilot?

Clear prices help a comparison, but scope must sit beside them

Our starter and private-consultant pages appeared in price-led answers. That supports keeping useful prices and inclusions visible. It does not establish why the engines selected them. The charity overview blurred starter scope with bespoke payment integration, so we clarified that distinction on our source page. Search summaries may take time to reflect changes.

A retrieved page is not always cited

The private-practice reading path appeared in Perplexity’s Sources panel but was not linked in its answer. That distinction matters when evaluating a visibility report. The planning answer linked official guidance, suggesting that precise source references are useful for this question; it does not imply those publishers are backlink prospects.

Prompt wording can change the market being compared

Google interpreted “private consultant” as a general business website question. A future version should specify “private medical consultant” and preserve the old wording alongside the new one. Rewriting the prompt and treating the results as an unchanged trend would obscure the difference.

For another example of publishing original evidence, see our six-site charity donation-page review, with source-level observations and a practical checklist.

How should a future run work?

1. Fix the target country, account state, engine interface and exact prompt version before testing.
2. Run the same prompts in each available engine, starting a fresh conversation each time. Log unavailable engines and errors separately.
3. Record the date, brand mention, answer links and source-panel links. Keep a private capture with account details removed before sharing.
4. Repeat the same set on later dates and compare like-for-like observations. Keep prompt changes in a new version.
5. Evaluate enquiries and qualified opportunities separately from citations. A generated answer is not a business outcome.

Download this pilot’s seven-row log (six answers and one error), or use our AI visibility checker to prepare a prompt set for your own business.