AI visibility (share of model)
More and more people ask an AI assistant directly instead of a search engine — and receive a finished answer with concrete recommendations. AI visibility measures whether and how your brand appears in such AI answers: you store brand and comparison terms, your own example questions and regions, XICTRAQ periodically queries an AI model and evaluates each answer — citation rate, mentioned URLs, sentiment and share of voice against your own comparison terms. Available from the Business plan.
# What AI visibility shows
Per profile you see three figures at a glance: how often your brand is mentioned in the AI answers at all (citation rate), what share your brand takes versus your own comparison terms (share of voice) and in what sentiment the brand is talked about. Below that a time series so you can tell whether visibility develops over the weeks — and a list of the most recently evaluated probes per question and region.
# Configuring a profile
A profile bundles everything that defines a brand. You can create any number of profiles — for example per product line, per sub-brand or per managed customer website. Optionally you link a profile to one of your monitored websites; that is the prerequisite for deterioration to be delivered as an alert.
Brand & term lists
You store the brand name (up to 120 characters) and two free term lists: brand terms (spelling variants, product names, domain) and comparison terms (terms you want to measure yourself against), each entry up to 120 characters. Both are pure free text — XICTRAQ does not prescribe competitors and keeps no external term database. Detection is case-insensitive and word-boundary precise: a term like “Meta” deliberately does not match inside a word such as “Metallbau”. If the brand term list stays empty, the brand name itself is used as the search term.
Your own example questions (prompts)
You phrase realistic questions a prospect would ask an AI assistant (“Which providers for … are there?”, “What do I use to …?”), each up to 500 characters. How many questions are evaluated per profile depends on the plan: up to 10 on the Business plan, 25 on the Agency plan. Exactly these questions are evaluated — so you test precisely the purchase intent that matters for your business, instead of generic metrics.
Regions (DE / AT / CH)
Per profile you choose one or more regions: Germany, Austria, Switzerland. The region is purely a context hint in the prompt (“The question comes from a person in …”) — no external geo service is queried and no IP is processed. Each question is evaluated once per chosen region, so you see regional differences separately.
Interval & on/off
The query interval is daily, weekly or monthly. A daily cron run checks which profiles are due and runs the probes; weekly is the default and sufficient for most brands. A toggle pauses a profile at any time without deleting it.
# The metrics in detail
Citation rate
The share of evaluated answers in which at least one brand term appears. A rate of 64 % means: in roughly two out of three answers to your example questions your brand is mentioned. The profile figure is the average over the most recently evaluated probes (a sliding window of the latest runs). This is the core figure of AI visibility.
Mentioned URLs
If the brand is mentioned, XICTRAQ additionally extracts the addresses linked in the answer (up to ten per probe) — both full links and domains the AI names only as plain text with common endings such as .de, .com, .at or .ch. This shows whether and which of your pages appear as a source and which other pages are linked in the same answers. Answers without a brand mention store no addresses.
Sentiment
If the brand is mentioned in an answer, the brand-related sentences are classified heuristically via a bilingual word lexicon: positive, neutral or negative. Answers without a brand mention get no sentiment. The average over the rated probes yields the profile sentiment. This is rough orientation — a word-list heuristic, not a third-party sentiment service — and does not replace reading individual answers.
Share of voice
The share of your brand mentions among all relevant mentions — i.e. brand terms divided by (brand terms plus your comparison terms), in percent. 58 % means: of the counted relevant mentions, a good half goes to your brand. If neither brand nor comparison term is mentioned in a probe, the value stays “not available” instead of 0 %. Since the comparison terms are your own, you compare exactly what matters to you.
"Which website monitoring providers are there in Germany?"
Mention
cited
Sentiment
positive
Share of Voice
67 %
URLs found
2
# Alert on falling visibility
A profile linked to one of your websites can warn you when the citation rate drops noticeably. After each profile run, XICTRAQ compares the newer half against the older half from the sliding window of the latest runs. If the newer rate falls below 60 % of the older one, a "warning"-level analytics alert is created and delivered through your configured channels.
When an alert fires
The check runs against the profile’s latest runs: at least six recent probes must exist, then this window is split into a newer and an older half. Only when the newer half’s citation rate falls below 60 % of the older half does this count as a relevant deterioration. So the alert fires on a real drop, not on the normal noise of individual answers.
Prerequisite: linked website
Without a linked website there is no alert target — such profiles evaluate silently without notifying. Once you link a profile to a monitored website, delivery runs through exactly the alert channels you have already set up for that website.
Delivery via your channels
The alert goes through the same delivery layer as every other notification — email, Slack, Teams, Discord, Telegram, webhook or browser push, depending on your configuration, including dedupe and retries. No additional third-party service is involved.
The citation rate in AI answers has dropped noticeably (from about 62% to 34%).
# Data frugality & limits
- Only your own inputs are processed: brand name, term lists, example questions, regions.
- The region is purely prompt context (DE/AT/CH) — no external geo service, no IP processing.
- The evaluation uses the models of one provider (no multi-engine mix); each run records the model that answered.
- Comparison terms are your own free-text lists — XICTRAQ does not prescribe competitors and keeps no external database.
- Sentiment is a word-list heuristic for rough orientation, not a binding verdict.
- All evaluations are fail-safe: if a probe fails, it is skipped and the rest continues.
# Plans, limits & budget
Each probe is a real AI query and counts towards the monthly probe budget of AI visibility, a counter of its own, separate from the AI assistant’s quota. Three plan-dependent limits govern consumption: the number of profiles, the number of example questions per profile and the monthly probe budget. Once the monthly budget is used up, queries pause until the next month rolls over; values already collected are kept. Many questions, short intervals and several regions raise consumption — one question across three regions is three probes.
| Limit | Business | Agency |
|---|---|---|
| Profiles per account | 3 | 10 |
| Example questions per profile | 10 | 25 |
| Probes per month | 600 | 2.500 |
The monthly budget can be raised via the AI extension add-on (as with the AI assistant) — each booked tier unlocks additional probes per month. The profile overview always shows your current quota, the consumption of the running month and the remainder.
Frequently asked questions
# How it works in detail
How a probe works: at the configured interval the service sends your stored sample questions to an AI model, optionally with a regional context (DE, AT or CH). The reply is checked across four dimensions. First, mention or citation: whether your brand appears in the answer and in what wording. Second, Share of Voice: how often your brand terms are named relative to the competitor terms you entered. Third, sentiment toward your brand, which is only assessed when the brand is actually mentioned, otherwise it shows as "not available". Fourth, the URLs and domains the AI names, which indicate the sources the answer draws on.
If your brand is not mentioned in a probe, that is not an error but a valid result: it points to low AI visibility and therefore to a need to act on your content and on findability for AI systems (often referred to as GEO). Across several probes a trend emerges that is more meaningful than any single value.
An important note on interpretation: one AI model is deliberately queried as a sample in order to deliver comparable and budgetable measurements. AI answers are not deterministic, so the same question can produce slightly different results at different times. Treat the values as indications and a development line, not as an exact or complete picture of your presence across all AI systems.
On the budget: each individual probe consumes a portion of your plan-based monthly AI budget. Many prompts, short intervals and several regions increase consumption accordingly. Plan your prompts and interval to fit your available quota. This feature is available from the Business plan.