How to Measure AEO: Mentions, Citations and Evidence
Neaptide · September 7, 2026 · 6 min read · Updated: October 7, 2026
Three recorded Neaptide mentions in ChatGPT and Gemini show how to document AI visibility with the exact question, linked page and limits of each observation.
On this page
A useful AEO report connects a question to a saved answer and the page it cites. Brand mentions, source links and enquiries are different outcomes. Keeping them separate lets a team show what actually happened and decide what to measure next.
Our Next.js audit page appeared in ChatGPT and Gemini
On October 7, 2026, ChatGPT named Neaptide in two recorded answers and Gemini named it in one recorded answer to our S2 service-discovery question. Each answer linked to our existing Next.js SEO audit page. The question asked for a provider able to diagnose a Next.js App Router indexing problem, produce a developer backlog and optionally implement the changes.
Who offers a technical SEO audit for a Next.js App Router website whose public pages are not getting indexed? We need rendering and metadata diagnosis, a prioritized developer backlog and the option to implement the fixes.
| Surface | Recorded observation | Neaptide position in that answer | Linked page |
|---|---|---|---|
| ChatGPT | S2 first round | 4 | Next.js SEO audit |
| ChatGPT | S2 repeat | 2 | Next.js SEO audit |
| Gemini | S2 repeat | 1 | Next.js SEO audit |
These are selected successful observations from a broader baseline, not an overall visibility rate or a before-and-after growth claim. The positions describe only the saved answers. Browser region was uncontrolled and the interface did not consistently disclose the model.
- Selected mention records and source URL · JSON
- Original question protocol · JSON
- The service page linked in the answers
Record a mention and a source link separately
In these observations, Neaptide was offered as a provider and our service page appeared among the visible sources. Both facts are recorded. A source link by itself would establish only that the page appeared as a source; it would not establish a recommendation. A position in one answer is also not a persistent ranking.
The saved answers described our audit in terms of rendering and metadata checks, developer tasks and an option to implement. These are features stated on the service page. Their appearance helps us inspect how the offer was represented; it does not prove which page feature caused the recommendation.
Preserve the conditions so another check can be compared
- Keep the exact prompt, any common instruction and whether a brand was named.
- Record the date, surface, visible model or mode, account state and known region.
- Save the answer and the available links before summarising it.
- State how mentions, recommendations and citations are classified.
- Separate fresh first rounds, repeats and later checks rather than combining them into an apparent trend.
Our S2 prompt did not contain Neaptide’s name. That distinguishes this observation from asking an assistant directly about our company. The public evidence packet contains the successful records; the frozen protocol explains the broader question design.
Use consistent denominators when reporting a rate
A case study can present selected successful observations, as this article does. A rate requires the full defined sample and its denominator. Do not calculate an overall visibility percentage from a success-only selection. Unavailable or incomplete observations need their own status rather than an assumed result.
When comparing periods, retain the same questions and classification rules. Record changes to account state, model, interface or market settings. A different answer after a page update is an observation to investigate, not automatic proof that the update caused it.
Connect the observed page to the content decision
Our cited Next.js audit page already describes a specific buyer problem and a defined deliverable. We preserved its URL and added a dated check of our actual production output. That gives readers evidence they can inspect alongside the commercial scope.
For other service pages, use the same standard: explain the buyer’s task, the agreed work, implementation boundaries and the evidence available. The recorded Next.js mentions support a concrete case study; they do not establish a universal recipe for being cited.
Measure visits and qualified enquiries separately
Use available analytics and enquiry records to assess actions after a visit. Preserve the attribution method and reporting period. The three assistant observations here do not include traffic or revenue measurements, so no commercial lift is claimed.
Make the report useful to the team
Link completed work to its evidence, identify the owner of the next action and preserve the questions for a later check. A dated record of a real mention is useful on its own. It becomes more useful when the team can repeat the procedure and understand exactly which conclusion the evidence supports.