AEO NIUEXA

AI Visibility

12 min read

What AI Visibility Measures

AI Visibility asks AI engines the questions your buyers ask and records, for every answer, whether it names your brand, which pages it cites and whether it gets your facts right. Open it from AI Visibility in the navigation.

The AI Visibility overview: tracked brands, mentions in the last 30 days, active citations and unread alerts, then priority actions such as a content gap and mentions lost on Perplexity and ChatGPT.
The overview: totals for all your brands and the priority actions, such as lost mentions and content gaps.

Five engines can be asked. Each is labelled with how its answer is obtained: we call the provider's API, which is not the same as a person using the consumer app.

  • ChatGPT (OpenAI API with web search)
  • Perplexity (API)
  • Claude (Anthropic API with web search)
  • Gemini (API with Google Search)
  • Google (search results via Serper): counts only when the results page shows an AI Overview. A featured snippet, knowledge panel or ordinary result is labelled but never counted as a mention.

Set Up a Brand

  1. Open the site's dashboard, go to the AI Visibility tab and choose Enable Tracking. Setting up is free.
  2. Enter the brand name and what the business sells. "Other" or "General" cannot be turned into questions: leave the industry blank and describe the business instead.
  3. On the brand page, under Brand settings, add the other names AI answers may use (an abbreviation, an old name) and your other domains (a shop, a help centre). An answer counts as yours when it names any of them as a whole word, and a cited page counts as yours when it is on any of your domains.
  4. Choose the engines to ask, and add the competitors you want to compare with (Discover suggestions proposes some, free with a daily limit).
  5. Pick the questions to track (next section), then run a check or turn on weekly monitoring.
The top of a brand page: mentioned in 43.7% of answers to unbranded questions, 2nd of 3 brands, with Run new check, Pause tracking and How these numbers are measured; below it, the last counted weekly run compared with the one before it: answers that mention you, answers that cite your pages, average position when named and open wrong facts.
The brand page opens with the weekly share, your rank among your competitors and the last weekly run compared with the one before it.

Questions

The Questions page holds the questions buyers ask AI assistants about your market. You can keep as many as you like; up to 10 are tracked, and only tracked questions are asked in a check.

The Questions page: tracked 4 of 10, a table of questions with target page, owner, source and the latest answer of each engine, marked named you, answered without naming you, or no AI Overview.
Each question shows the latest answer of each engine: named you, answered without naming you, or no AI Overview.
  • Add a question by hand, with an optional target page and owner.
  • Paste questions from sales or support, up to 200 at a time, written as customers asked them. They are added untracked.
  • Import from Search Console: upload the export ZIP or its Queries file (at most 1 MB). Question-like queries are listed by impressions and nothing is ticked for you. The file is only read here; nothing is sent to an AI.
  • Suggest questions on the brand page writes questions from your business description. They are added untracked; free, with a daily limit.
  • Archive removes a question and keeps its earlier answers.

Questions that name your brand ("Harbor Lane Coffee vs Ridgeway Roasters") are counted apart from unbranded ones, because an engine almost always names you when the question does.

Checks

Run new check asks every tracked question on each of the brand's engines and costs 10 credits. The credits are held when the check starts and charged only if at least one engine answered; a check where no engine answers is free and shown as failed. An engine whose provider is not available at that moment is skipped and named. Pause tracking stops new checks and weekly runs until you resume.

Weekly Monitoring

Once a week, on the day and hour you choose (in UTC), the brand's tracked questions are asked again on its engines, so you can compare the answers week by week. Each weekly run costs 10 credits, charged only if an engine answered.

The Weekly monitoring panel: on, tracked questions 4, the five engines asked, day Monday and hour 08:00 UTC, the next run, and the monthly budget for weekly runs across all brands.
Weekly monitoring: choose the day and hour, and agree a monthly budget for all your brands.
  • Before it can be turned on the brand needs a tracked question and an available engine, your email must be confirmed, and you must agree a monthly budget.
  • Monthly budget: one cap, in credits per calendar month (UTC), for the weekly runs of all your brands. The panel shows the projected spend. When the cap is reached, the month's remaining weekly runs are skipped, never charged. A new cap needs your agreement again. Checks you start yourself are never counted against the cap.
  • Missed and late runs: a run that could not start in its slot because of downtime is not replayed; it moves to the next week and you get one notice. A run that waited for a free worker runs late. An interrupted run is retried once.
  • Time limit: a run asks each question on all engines before moving to the next question. If it reaches its time limit, the last questions are left out, not whole engines.

Reading the Results

Weekly share of AI answers

The main number: the share of answers to your unbranded tracked questions that mention you, from weekly runs only. It is the average of the last 4 weekly runs that count (a run counts when at least half of its planned answers arrived), shown once there are 3, with its lowest and highest week and how it was measured. Engines that did not answer and Google searches without an AI Overview are left out. It is a count of stored answers, never a forecast. Changing the tracked questions or the engines starts a new series.

The top of the brand page

The brand page opens with the weekly share and, when you track competitors, your place among them ("2nd of 3 brands"). How these numbers are measured next to Run new check sums up the rules below.

  • The four figures compare the newest weekly run that counts with the counted run before it, on unbranded questions only, and show the counts behind each share ("6 of 14 answers"): answers that mention you, answers that cite your pages, your average position when named (1 means the answer named you before any other brand) and the open wrong facts, with how many are new since the earlier run. The first counted run of a series has nothing to compare with, and says so.
  • Weekly share in this series plots the share run by run, for the current series only (its last 12 weekly runs at most), with the answers received out of those planned under each date. A run that does not count leaves a gap in the line, never a guessed value. The notes under the chart list the answers that never arrived, by run and engine, and the Google searches without an AI Overview; Show the numbers opens the same runs as a table.
  • Brands in AI answers ranks you and your tracked competitors over the same weekly runs as the weekly share. Each brand's share is the mean of the runs' shares, worked out like the weekly share, so your row matches it; the counts are totals over those runs, with the answers that cited a page on each brand's site. A competitor is named in an answer when the check recorded it, or when its name appears in the answer text as a whole word, so one you add later is also found in those runs' answers.
Weekly share in this series: one point per counted weekly run from 28 Sep to 19 Oct, ending at 42.9%, with 19 of 20 answers on 5 Oct and the notes "no answer from Perplexity (1 question)" and "Google showed no AI Overview for 5 searches". Brands in AI answers: Copper Kettle Coffee 60.0%, Harbor Lane Coffee (you) 43.7%, Ridgeway Roasters 38.0%, each with the answers that named it and cited its site.
The trend of the current series and the brands named in the same answers.

AI Visibility score (0-100)

A different measure: a 30-day composite of five categories, each scored from 0 to 100 before the weights below are applied.

Presence
30%
How often the brand appears in completed checks
Sentiment
20%
Keyword sentiment of the sentences that name the brand
Authority
20%
Citation frequency and strength
Position
15%
Where the brand appears in AI responses
Competitive
15%
Share of voice against configured competitors; not measured (and left out of the total) when no competitor is configured

A score older than 30 days is shown as "No current score" with the last score and its date.

Score breakdown: Presence 79, Sentiment 72, Authority 61, Position 90, Competitive 74; and the content pipeline: open content gaps, briefs, articles, unread alerts and wrong facts to correct.
The score breakdown and the content pipeline on the brand page.

Who AI Cites

The Sources page lists the pages the engines cite when they answer your tracked questions, from the last 4 weekly runs that count: your own pages, your competitors' and everyone else's, with best-of lists apart. Each row shows how often a site was cited, by which engine and for which questions.

Who AI cites: from the last 4 weekly runs, your pages were cited 42 times, shown by engine, by question and by page.
Sources: where the engines take their answers from, yours first.

Progress Run by Run

The Progress page shows each tracked question across the last runs, weekly and manual, oldest first: for every engine whether it mentioned you, and under each cell your pages that were cited. A change you made to the questions or engines is marked above the run it affected.

The Progress page: four tracked questions across five runs from 28 Sep to 19 Oct, with one mark per engine and the cited pages under each cell.
Progress: one row per question, one column per run.

Alerts and the Weekly Email

After each weekly run, the answers are compared with the two runs before it. You get an alert when:

  • Mention lost - an engine named you before but not in the last two runs
  • Your page no longer cited - an engine cited one of your pages before but none in the last two runs
  • Competitor cited instead of you - a competitor was cited in the last two runs and you were not mentioned or cited
  • New mention - an engine names you after not doing so in the two runs before
  • Negative answer - the words about you were mostly negative (a keyword check, not a judgement of the facts: read the answer)

A skipped, failed or missed run raises one notice. Brands without weekly monitoring use rules that need repeated evidence, for example two later answers to the same question on the same engine before a citation counts as lost. The Alerts page groups open alerts by kind; filter by brand, kind and severity, and mark them read or dismiss them.

The Alerts page: six open and six unread alerts, open alerts grouped by kind (Mention lost, Competitor cited instead of you, Negative answer, New mention, Your page no longer cited) and the filters.
Alerts, grouped by kind.
Weekly digest: after a weekly run that raised a medium or higher alert or a new mention, we email you a short digest. Turn it off under Profile → Email notifications or with the one-click unsubscribe link in every email. Alerts always stay in the app.

Answers and Citations

Mentions and Citations on the brand page list every stored answer and every cited page. An answer's page shows the question, the full answer, whether and where your brand was named, its keyword sentiment and the pages it cited, marked as yours, a competitor's or a best-of list.

Acting on the Results