Skip to content
RankX AI
Start freeTry free

Cookie settings

Choose which cookies we can use. You can change this at any time from “Cookie settings” in the footer.

Essential

Security, bot protection and remembering your choice. The site needs them to run, and they do not identify you. Always on.

Google Analytics cookies that tell us which pages are read and which links are followed, so we can improve them. Off until you switch it on.

Tells advertisers when their advert led to your visit. Includes the Meta (Facebook) pixel and the LinkedIn Insight Tag, which set cookies on this domain (LinkedIn also on linkedin.com) and can be told if you later start a trial.

More in the cookie policy and the privacy policy.

How RankX AI measures AI visibility.

RankX AI publishes how every number is made: when checks run, where each answer is read from, how a mention is decided, and what a rate can and cannot tell you.

Choose your platformSee Pricing

7-day free trial, no card. On Shopify, install the app and start free.

The schedule

RankX AI checks on a schedule: trends, not screenshots.

Three passes run overnight, UTC: shopping answers, then the AI assistants, then Google rankings and their AI Overviews. Each prompt joins on your cadence.

One scheduled checkIllustrative

  1. Schedule

    The passes fire overnight. Each prompt joins when its own cadence is due.

    01:00 UTC02:00 UTC03:00 UTC

  2. Assistants queried

    Each due prompt is asked once on each assistant it targets.

  3. Answers parsed

    Brands, rivals and every cited URL are read from the answer.

  4. Two-pass mention verdict

    A language model reads it, then a deterministic check has to agree.

  5. Stored

    The answer is kept with its verdict.

    NamedNot named

FIG. 01One overnight pass, from schedule to stored verdict

01:00 UTC

AI Shopping

Shopping prompts run against the shopping surfaces, and every product card in the answers is matched against your own catalogue.

02:00 UTC

AI visibility

Every prompt that is due is asked once on each of the assistants it targets, out of ChatGPT, Claude, Gemini and Perplexity, and each answer is stored with its verdict.

03:00 UTC

Google rankings

Tracked keywords are checked in Google search, and AI Overview presence and citations are recorded on the same check.

You choose the cadence, set per website and adjustable per prompt, from daily to monthly, and each prompt carries its own list of assistants. Every paid plan can run daily; a trial account re-checks weekly. Any prompt can also be re-checked on demand. Each scheduled check asks each prompt once per assistant it targets, and the rates aggregate those single readings across your panel and the window.

One prompt, 10 runsSample data

Where to buy replacement shaker cabinet doors in the UK?

  1. Run 1: named
  2. Run 2: not named
  3. Run 3: named
  4. Run 4: named
  5. Run 5: not named
  6. Run 6: not named
  7. Run 7: named
  8. Run 8: not named
  9. Run 9: not named
  10. Run 10: not named

40%

Mention rate: named in 4 of 10 runs

FIG. 02Ten runs of one prompt: the mention rate is the share of filled dots

The source

Two assistants are read from the interface a person uses.

ChatGPT’s and Gemini’s citations are largely absent from their programmatic legs, so RankX AI reads the consumer interface.

Where each answer is read fromIllustrative

RankX AI asks

Consumer interface

The answer a person sees in the app

  • ChatGPT
  • Gemini

Fallback for ChatGPT and Gemini: the official interface

Official interface

Read programmatically

  • Claude
  • Perplexity
FIG. 03Consumer interface or official interface, assistant by assistant

  • ChatGPT

    Consumer interface

    The answer a person sees in the app, official interface as the fallback.

  • Claude

    Official interface

    No comparable consumer surface to read.

  • Gemini

    Consumer interface

    The answer a person sees in the app, official interface as the fallback.

  • Perplexity

    Official interface

    Returns its source list alongside the answer.

A citation that only renders in the app a buyer uses is invisible to a purely programmatic check.

The verdict

Mentions count only once a second pass confirms it’s you.

A language model reads each stored answer, deterministic checks verify the brand match before anything is counted, and an unreadable answer is recorded as unknown.

Your brand: Halewick SupplySample data

  1. Stored answer

    For painted oak shaker doors, try Halewick Supply. Halewick Joinery also makes doors to order.

  2. Pass 1

    A language model flags possible mentions

    • Halewick SupplyPossible match
    • Halewick JoineryPossible match

  3. Pass 2

    A deterministic check has to agree

    • Halewick SupplyNamed
    • Halewick JoineryNot your brand

    Only the confirmed mention is counted.

FIG. 04A mention counts only when the second pass agrees

Mention

A language model reads the stored answer for your brand and your rivals, and a deterministic verification pass then has to agree the match really is your brand before anything is counted.

A loose match that counts a rival as you would corrupt every rate downstream.

Sentiment

How the answer describes you is read only when a verified mention exists. No mention means no sentiment call and no invented neutral score.

Sentiment about an absence is fiction, so RankX AI does not record one.

Citations

The pages behind an answer come from the assistant's own source list where one is returned, plus every URL the answer text itself carries, with the domain extracted for each.

Who gets cited is the fact you can act on, page by page.

Unknown

An answer that returns no analysable verdict is recorded as unknown and left out of the denominator. Every rate divides by the answers actually read.

Unknown is never rounded down to zero, because the two are different claims.

10 stored answersSample data

Read: 8 (4 named)

  • Named
  • Named
  • Named
  • Named
  • Not named
  • Not named
  • Not named
  • Not named

Unknown: 2, set aside

  • Unknown
  • Unknown
  • Counting unknown as a miss would give 4 ÷ 10 = 40%
  • RankX AI leaves unknown out: 4 ÷ 8 = 50%
FIG. 05Unknown is left out of the rate, never counted as a miss

The other pipelines

Google AI Overviews and shopping are measured apart.

AI Overviews are read from your tracked keywords’ rank checks, and shopping answers are matched product by product against your catalogue over a 28-day window.

Google AI Overviews

Every rank check records whether an AI Overview appeared on the keyword, whether it cited your site, and every URL it cited. Nothing is inferred from a single look: the record is per keyword, per check, so citation questions are answered from a run of checks rather than one screenshot.

AI Shopping

Shopping answers on ChatGPT Shopping and Google AI Mode are read over a 28-day window: four whole weeks, so a weekly check cadence contributes evenly. A product card matches your catalogue by an evidence ladder, your domain first, then your seller name, an exact URL, a GTIN, and finally title similarity, and the method that matched is stored with every row.

A match that cannot say how it matched is a guess with a percentage sign.

Questions

What people ask about the methodology.

The mechanics behind every RankX AI figure, in direct answers.

Why does RankX AI report shares instead of a single answer?

Because a single AI answer is an anecdote. One independent study (SparkToro, 2,961 prompt repetitions) found under a 1-in-100 chance that two runs of the same question return the same brand list, so RankX AI reports the share of tracked answers that name you, across your whole prompt panel and the reporting window, and shows the trend rather than a screenshot.

How many times is each prompt checked?

Once per assistant per scheduled check, and you can force a fresh check on any single prompt whenever you want. The published rates aggregate those single readings across every prompt and every check in the window, which smooths run-to-run variance without multiplying what each check costs you.

Why read the consumer ChatGPT instead of the API?

Because the two return different answers, and the consumer interface is the one a buyer actually sees. Citations render in ChatGPT's and Gemini's consumer interfaces and are largely absent from their programmatic legs, so RankX AI reads the consumer surface for those two, with the official interface as a fallback. Claude and Perplexity are checked through their official interfaces.

Will these numbers match what I see when I ask an assistant myself?

Not necessarily, and no honest tool should promise otherwise. Assistants vary their answers run to run and by account, region and time of day. That variance is why RankX AI reports shares over a window rather than predicting your next answer, and why an unreadable answer is recorded as unknown instead of being counted either way.

More detail in the documentation, and what each measurement means on the AI Visibility page.

Start here

See where you show up in AI answers today.

Start with your website. RankX AI recognises your platform, suggests the questions to track and audits your pages, with first results minutes after setup.

Start with your websiteRun a Free AI Audit

7-day free trial. No credit card required. On Shopify, install the app and Shopify bills your plan.