AI Visibility

AI Visibility

The hosted “AI visibility” tools run a fixed set of prompts against the AI assistants on a schedule, record the answers and count which URLs each answer cites. Every number they show is a citation inside their sampled answers, and they cannot see the one thing that is not a sample: which of your pages the assistants’ crawlers actually fetch.

Flowcourier.Umbraco.Analytics.AI does the lean version inside your site, on your own Umbraco AI profiles and your own key, and puts the two real signals beside it — AEO’s verified AI-crawler fetches and Search Console’s clicks.

What a share means here

Every share on these surfaces is within your prompt set, over the answers that searched the web. Write a different prompt set and you get different numbers. A run where the model answered from memory without searching is counted apart (“Answered without search”) — it says nothing about whether your site would have been found — and never as “not cited”.

How it works

1. Prompts. You write the questions you want assistants to answer with your site — on the dashboard, or on a document’s sub-tab (a prompt aimed at that page), or by letting the model suggest some from the page’s text.
2. Sampling. Every prompt is put to each configured chat profile, with the provider’s web search switched on, every 72 hours by default. The answer and the URLs it cites are stored.
3. Reading. Each cited host is yours, a competitor’s or third party’s, decided from your configuration when a report is built — so editing the competitor list reclassifies history instead of splitting it. Cited URLs on your own hosts are resolved to the documents behind them.

Every run is a model call plus a provider-priced web search on your key. It is off by default; there is a daily run budget; and the dashboard states the sample size above its figures.

Install and switch on

It needs the Analytics AI package, an Umbraco AI chat profile whose model can search the web (an OpenAI Responses-API model such as gpt-5.x or gpt-4.1 today), and an Analytics Pro licence.

{
  "Flowcourier": {
    "Analytics": {
      "AI": {
        "Visibility": {
          "Enabled": true,
          "ProfileAliases": ["sampling"],
          "BrandNames": ["Acme"],
          "BrandDomains": [],
          "CompetitorDomains": ["rival.com"],
          "DailyRunBudget": 100,
          "IntervalHours": 72
        }
      }
    },
    "Licenses": {
      "AnalyticsPro": "FC1.…"
    }
  }
}

Setting	Default	What it does
Enabled	false	The whole feature. Off, nothing is stored and nothing appears. On, three tables are created on the next start (FcAiVisibilityPrompt, FcAiVisibilityRun, FcAiVisibilityCitation).
ProfileAliases	[]	The chat profiles to sample, one answer per prompt per profile. Empty means the default chat profile. Use a dedicated profile without a system-prompt template — a template written for the Copilot steers the sampled answer.
IntervalHours	72	How long after its last run a prompt is due again. Answers are not deterministic; a share only settles over repeated runs.
DailyRunBudget	100	Runs (prompt × profile) per UTC day, shared by the scheduled job and Run now.
MaxPrompts	200	The most prompts a site may track — a cap on the bill.
BrandNames	[]	Names whose presence in an answer’s text counts as a mention.
BrandDomains	[]	Hosts that are “you”. Your site URL’s host and the Umbraco domains are always included.
CompetitorDomains	[]	Hosts that are competitors. Everything else cited is third party.
RetentionDays	365	How long answers are kept.
RunTimeoutSeconds	90	A run that has not answered by then is recorded as a timeout.

Settings → Flowcourier → Analytics AI shows the configuration and the licence state.

The dashboard

Analytics → AI Visibility (it sits between Search queries and AEO’s AI Crawlers):

• Prompts tracked, answers sampled, cited (answers with web search that cite your site), mentioned, competitors cited and answered without search, each against the previous period.
• Cited share over time, one line per model.
• Who gets cited — your site, competitors, third party — and the cited domains.
• Your cited pages, each opening the document, with AEO’s count of verified AI-agent fetches beside it when AEO is installed. Fetches are real traffic; citations are a sample.
• Prompts — add, pause, run now, remove; a row opens the sampled answers: the latest answer as the assistant gave it, its cited sources, and the run history per model.
• Gaps — prompts whose latest answer cites a competitor and not you.
• Real fetches vs sampled citations — pages AI agents fetch but the answers never cite (the pages to restructure for answer engines), and pages cited but never fetched. Computed against AEO’s 200 most-fetched paths in the range, and the card says so.

Run due prompts now samples immediately instead of waiting for the hourly tick, within the same daily budget.

The document sub-tab

On every web page’s Analytics tab, AI Visibility (after AEO’s AI Traffic):

• Whether the answers to the prompts aimed at this page cite it, per model.
• One line of real signals: verified AI-agent fetches and visits from assistants (AEO), Google clicks and impressions (Search Console).
• Citations of this page, whichever prompt produced them, with the exact URL cited.
• Prompts aimed at this page — add your own, or Suggest prompts: one model call that proposes up to five questions the page could answer, from its text and the Google queries it already appears for. Nothing is stored until you add one.
• Cited instead of this page — who the prompts aimed at it cited.

Three findings join the strip: AI assistants cite this page (info), AI agents fetch this page but the sampled answers never cite it (warning, needs AEO), and Competitors cited for prompts aimed at this page (warning).

Licence

Sampling, prompt writes, Run now and Suggest prompts need an Analytics Pro key at Flowcourier:Licenses:AnalyticsPro. Unlicensed, the dashboard and the sub-tab render a banner over whatever data exists, the job stays idle, and a write is refused with the reason. A key added later needs no restart.

Limits

• Engines are whatever Umbraco AI profiles you have: OpenAI today, Anthropic when its provider maps the web-search tool. Perplexity, Gemini and Copilot have no Umbraco AI provider.
• A profile whose model cannot search fails every run with “this profile’s model or provider does not support web search” — pick a model that can.
• Multi-site installs whose Umbraco domains carry a path (example.com/da) may not resolve every cited URL to its document; the citation is still counted for the host.