How AmICited collects data

AmICited does not estimate or model AI visibility. It asks the engines the actual questions, on a schedule, from the country you choose, and stores every answer. Every number in the product is an aggregate of those stored answers, so you can always drill from a metric down to the exact response behind it.

What happens on every run#

When a prompt is due, AmICited runs it once on each engine it is set to use:

1. Ask. The prompt text is sent to every selected engine, localized to the prompt’s country and that country’s language.
2. Store the response. The full answer text is saved, together with the engine, country and timestamp. One prompt on one engine is one response.
3. Record the citations. Every URL the engine listed as a source is stored as a citation, with its domain, title, snippet and its position in the source list.
4. Extract the brand mentions. A small language model reads the answer text and lists every brand, company or product it names, in order of first appearance, plus the URL the answer linked for that brand, if any.
5. Record the extras. Where the engine reports them, AmICited stores the sub-queries it searched (query fan-out) and any sponsored ads or product listings shown in the answer (ads and products).

The prompt is marked finished only when the last engine has reported back. If an engine fails, the run is retried after 5 minutes, and only the failed engine is re-asked.

The engines#

AmICited tracks eight engines: five on every plan and three premium ones. Each is collected the way that best reproduces what a real user sees.

When AmICited reads the Google results page for AI Overviews, it also keeps the plain organic results as a Google Search baseline. You can see it on a prompt’s page, but it is excluded from every citation metric, because a search ranking is not an AI citation. See AI engines for per-prompt engine selection.

Country and language#

Each prompt is tracked from one country (default United States), chosen from 139 supported countries. The language of the query comes from the country: each country maps to one primary language, so Switzerland runs in German and Canada in English. To track the same question in two markets, add it once per country; the same text can exist in several countries. See countries and languages.

How your brand is recognized#

AI answers name brands in many spellings. AmICited matches your brand against the brand name and the domain you set for the domain, after normalizing both: case, spaces, punctuation and accents are ignored, and a bare hostname like acme.com reduces to acme. So “Acme”, “ACME” and “acme.com” all count as you.

Each brand is recorded once per answer, at its earliest position. An answer that names you five times counts as one mention at the position where you first appeared. This keeps long, repetitive answers from inflating your numbers. The full distinction is on mentions vs. citations.

For citations, the cited host is compared with your domain, with a leading www. removed.

Where the data lives#

Every response, citation and brand mention is written to a per-domain data warehouse (BigQuery), alongside your search, ads, site audit and ecommerce data. History is kept in full, not sampled. How far back you can look in the app depends on your plan’s citation history (7 days on the trial, up to unlimited on Enterprise); see plans and limits.

All figures on the dashboard respect the same filters (date range, engine, country, tag), and each one is compared with the previous period of the same length.

Schedules and cost#

Each prompt runs on its own schedule: hourly, every 3 hours, daily, weekly, monthly, yearly or one time. One response costs 0.015 credits, plus a small token-based charge for the brand-extraction step. A prompt tracked daily on 5 engines makes 5 responses a day:

5 × 0.015 = 0.075 credits per day, about 2.25 credits a month.

If the workspace runs out of credits, scheduled runs pause and the domain is marked credit-blocked until credits are added. See run frequency and credits.

Related#