How we measure

How we measure: four channels, one rhythm.

Every Monday from 4 a.m. AI answers, Google and press run, ads on the first Monday of the month, each for your brand and for every tracked competitor. This page shows what Rivly counts in each channel, where the data comes from, how the Rivly Index is built from it and where the measurement reaches its limits.

Four channels

What Rivly counts in every channel, and what it does not.

GEO

AI visibility

What: The share of AI answers in which your brand is mentioned. Depending on the plan, 15 or 20 buying questions are measured in four AI systems, every week.
How: Every question is asked on Monday, without history and without a sign-in, from Germany. The result is the mean of the four systems, each counts equally.
Where from: We query ChatGPT and Gemini the way their answer appears to people in the application, ChatGPT with web search. We query Perplexity and Claude through their interfaces. Every answer stays readable word for word, together with the sources the system quotes.
Mention: Your brand counts as mentioned when the answer names it. If your name has a common form, it only counts together with your domain or your Google listing.
Position: The order of the mention inside the answer. Places 1 to 3 count as mentioned up front.
Variation: When Rivly asks the same question once more on the same day, in our test runs about one question in five changes on whether your brand is mentioned (19 % of 64 questions). The weekly report therefore names a change only when the mean of the last up to four weeks rises or falls by 5 percentage points, or a single week by 10 percentage points against the previous one.
Limits: AI answers move from week to week. A single week is a snapshot; the direction only shows in the history across several weeks.

Search visibility

What: The position of your domain in the Google top 10 of your market and whether Google shows an AI Overview.
How: Desktop, the location of your market (Germany, Austria or Switzerland), without personalisation. Search volume as a monthly average.
Where from: Google search for your focus queries, queried again every Monday. The search volumes are updated once a month.
Index: Feeds into the Rivly Index with 35 %, as your share of the search demand: every search query counts with its search volume and the usual click rate of your position, from 28 % at position 1 to 2 % at position 10.
Limits: What is measured is the position, not your traffic. How many people click through stands in your own analytics.

Advertising

What: Publicly visible, active ads of your brand and of your competitors.
How: Once a month, on the first Monday, for your brand and for every tracked competitor. What is counted is what runs at the moment of the measurement. Until the next measurement the channel keeps this value.
Where from: The Meta Ad Library API and the public Google Ads Transparency Center. Rivly reads only what is visible to everyone there.
Winning ads: Ads that have been running for 30 days or more. The longer an ad runs, the more likely its message works.
Limits: Budgets and reach stay out of the picture. Neither is publicly visible.

Press

What: Editorial articles of the last 30 days that mention the brand.
How: Every Monday for your brand and for every tracked competitor. The window moves along, it is always the last 30 days.
Where from: Google News. Trade media are reported separately from press portals, because AI systems use them as a source more often.
Check: Every hit is checked against the brand, namesakes are discarded. If a competitor mostly writes about other topics, only its articles on your topics count in the comparison.
Limits: Print and broadcast are not included. What is counted is what can be found online.

Rivly Index 0 to 100

Formula: The weighted mean of the part scores: AI visibility 50 %, search visibility 35 %, advertising 10 %, press 5 %. Each part score is your value divided by the benchmark of the channel, capped at 100.
Benchmark: The highest of three values: the strongest tracked brand in your field, the strongest brand the AI names in your answers, and a minimum level. For search visibility the strongest other firm of the same kind in the same Google results counts instead of the AI answers: the same offer for the same customers in the same market. Portals, directories, media, brands from the range of a shop, clearly larger providers and the website of a brand on searches for its name do not count there. If there is no such firm, the tracked field counts, at least the minimum level. That firm serves only as the benchmark and does not appear in the ranking. If your field changes, the history gets a marker.
AI answers: Only for AI visibility. What counts is a brand named in at least 2 buying questions by at least 2 AI systems. Media, directories, regulators, AI assistants and, for a shop, the brands it sells itself do not count.
Minimum level: 25 % of the answers, 2 % of the search demand, 10 active ads, 3 articles in 30 days. So there is no 100 without a real comparison: 2 ads against a field without ads give 46, not 100.
Bands: 0 to 39 behind, 40 to 69 mid-field, from 70 up front.
Example: 41 points of AI visibility, 30 search visibility, 50 advertising and 25 press give 41 × 0.50 + 30 × 0.35 + 50 × 0.10 + 25 × 0.05 = 37.
Gaps: A channel without a measurement drops out of the calculation and the remaining weights are redistributed. A missing number never pulls your index down. A tracked competitor that the last measurement did not cover stays empty and does not count as a zero.

RHYTHM

Rhythm

Measurement: Every Monday from 04:00, time zone Europe/Berlin. AI answers, Google positions and press run in the same pass, so the numbers of one week fit together. For projects with locations, the position in Google Maps, the rating and the number of reviews per location belong to it. ChatGPT and Gemini need up to 90 minutes for their answers, and Rivly waits for them.
Report: The weekly report goes out as soon as the measurement of your project is finished, at the latest by 08:00.
Monthly: On the first Monday of the month, ads, Google Business, backlinks, search volumes, the Lighthouse measurement (mobile) and the technology of the website are added, together with a pass through the pages of your website within the page quota of your plan. If the first measurement was less than 14 days ago, this part follows in the next month.
First measurement: It starts right after sign-up and takes about 30 minutes. If it fails, you can repeat it once. Changes against the previous week appear from the second measurement on.
Live check: Between Mondays, ask ChatGPT and Gemini a question your customers ask in the AI live check and see after a short wait whom the AI recommends. The number of questions per project and week depends on your plan, the quota renews on Monday, and the same question within 24 hours costs nothing. The live check is not a measurement: the index and the history do not change because of it.
Comparison: Every change compares against the last measurement of the previous measured week, whether that was the Monday measurement or the first measurement. Measurements of the same week are not compared with each other.
Trial: The trial runs for 14 days and covers two Mondays. So you see the full weekly rhythm before anything is charged.

Data quality

What “no value” means

An answer that is technically missing does not count as a loss, the denominator gets smaller. That is why Rivly shows four different states instead of an empty cell.

not collected yet

We have not measured there yet, for example before a channel was started or for a competitor you added after the last measurement. The field stays empty until the next measurement covers it.

Measured and nothing found. A zero is a result.

—

There were no hits in the period, for instance no articles in 30 days.

not meas.

The request failed. That does not count as a loss: Rivly repeats it the same day at 12:00, at the latest with the next measurement.

If a request is missing entirely, Rivly shows the value measured last together with its date. There are no estimated numbers, and “not collected yet” is never counted as a zero.

Access to your website

Who accesses your website, what is read and how you block it.

Rivly retrieves only publicly reachable pages, never areas behind a sign-in: pages of your domain for the set-up, the website check and your locations, and at set-up also the home page and legal notice of possible competitors. Two programs are in use. They identify themselves in the user agent, which is how you recognise them in your server log.

RIVLYBOT

RivlyBot: reading your domain and the home pages of competitors

Who: Rivly's reading program for every request our servers make to a website. User agent: Mozilla/5.0 (compatible; RivlyBot/1.0; +https://rivly.de/wie-wir-messen). The requests come from servers in Germany.
What: Your domain: at set-up the home page, sitemap, legal notice, contact, locations, about page and two town pages, so that Rivly recognises name, type of business, offers and places; in the website check the pages of your quota; for your locations the home page, legal notice, sitemap and the pages your Google Business Profiles link to. Other domains: at set-up the home page of up to 20 firms that Google, Google Maps or AI answers show as your competitors, to check that the website is reachable and belongs to the firm; for firms named only by a language model, just the robots.txt. If you add a competitor yourself, RivlyBot reads its home page and legal notice for the company name. For the Meta Ad Library, RivlyBot looks for the linked Facebook presence on the home page of your brand and of each competitor. Rivly keeps only features: address, response status, title, main heading, meta description, language, word count, indexability and the details recognised about the firm. The text of the page is not stored.
How often: Your domain: when your project is set up, one short pass for the recognition (up to 20 requests within a few seconds) and up to 40 pages for the website check. After that a short round every week over the home page, menu, footer and new addresses in the sitemap (up to 60 pages), and once a month a full pass within the page quota of your plan and the pages for your locations. A single page is added when you request a check in Rivly. Other domains: per website the robots.txt and the home page, once at set-up, the legal notice as well when a competitor is added, and for the Ad Library once a month as long as no Facebook presence has been found. Except in the recognition pass, RivlyBot reads at most two pages of a website at the same time, at least half a second apart.
Blocking: Before the first page, RivlyBot reads the robots.txt of the domain and follows the group “User-agent: RivlyBot”, or, where there is none, the group “User-agent: *”. This applies to your domain and to competitors' websites. Add “User-agent: RivlyBot” followed by “Disallow: /” for the whole website or “Disallow: /path/” for one area. RivlyBot follows it, and a waiting time given there (Crawl-delay) of up to 20 seconds. Rivly does not retrieve a blocked website: the set-up then relies on Google, Google Maps and AI answers, and Rivly does not rate blocked pages.
Your data: If RivlyBot reads your website because a Rivly customer tracks you as a competitor or because Google, Google Maps or AI answers show you as a competitor, we process public details of your company: company name, website address, offers, places and short supporting excerpts, not the page text. We do not record names of owners or representatives from the legal notice as a separate item. The legal basis is Art. 6(1)(f) GDPR (an accurate competitor comparison for our customers). The controller is Vladyslav Mazur, Mazur Software- und E-Commerce-Dienstleistungen, Kleingesee-Brunnenstr. 17, 91327 Gößweinstein, Germany, kontakt@rivly.de. Recipients are the customer to whom we show you as a competitor, our host Hetzner in Germany and, for the recognition, OpenRouter, Inc. (USA). We keep the details until the customer deletes their account and delete them earlier if you object. You can exclude RivlyBot via robots.txt, object to the processing and request access, rectification or erasure, informally to kontakt@rivly.de; we reply within one month. You can also complain to the Bavarian Data Protection Authority (BayLDA). We do not inform operators individually because that would reveal the customer who tracks you as a competitor (Art. 14(5)(b) GDPR). More in the privacy policy, section 9 letter b.

RIVLYCHECK

RivlyCheck: checking files and home page

Who: The checking program of the free tools on rivly.de and of the website check. User agent: RivlyCheck/1.0 (+https://rivly.de/tools/).
What: Files and pages of the domain being checked: robots.txt, llms.txt, sitemap, home page and, for the free tools, up to ten more pages. The website check also requests single pages with the identifiers of the reading programs of well-known AI providers (such as GPTBot, ClaudeBot, PerplexityBot) to see whether your server shuts them out. These requests are spaced out: at most two at a time, at least one second apart.
How often: On request: once when someone enters a domain in a free tool. For your project, on your instruction: at set-up, every week in a short version and once a month (Terms § 8(4)).
Blocking: A rule in robots.txt does not reach RivlyCheck: the program reads the file in order to check it. For your own domain, which you have set up as a project in Rivly, the retrieval happens on your instruction, and RivlyBot still reads your pages only according to your robots.txt; you end the instruction by deleting the project. If you operate a domain and want to exclude RivlyCheck from reading it, block the user agent RivlyCheck in your firewall or on your server. Questions are answered at kontakt@rivly.de.

Sources and licences

Which third-party data Rivly uses, and under which licence.

Rivly collects the measurements itself. One public data set comes from outside: the directory of places. It is under a free licence, which we name here.

SOURCE

Directory of places: GeoNames

Source: Places, spellings, coordinates and postcodes in Germany, Austria and Switzerland come from GeoNames. The data set is updated once a month.
Used for: Rivly uses it to recognise the places of your business on your website and in your Google listings, to match spellings (for example “Frankfurt am Main” and “Frankfurt”) and to relate search queries and competitors to your catchment area: the places your website names, with their surroundings.
Licence: Contains data from GeoNames, licensed under Creative Commons Attribution 4.0 (CC BY 4.0). Rivly has modified the data: without historical and abandoned places, spellings merged, postcodes assigned to places. geonames.org · CC BY 4.0

Transparency

Changes to the methodology

3 October 2026 Rivly Index, version 5: search visibility counts The Rivly Index now counts four channels: AI visibility 50 %, search visibility 35 %, advertising 10 %, press 5 %. Before that, search visibility did not feed into the number, AI visibility and advertising counted in equal parts, press with less. Search visibility is your share of the search demand, measured against the strongest other firm of the same kind in the same Google results, even if you do not track it as a competitor. Rivly recalculates earlier weeks from the values measured at the time, so the history does not jump because of the formula.
3 October 2026 AI answers: ChatGPT and Gemini as in the application ChatGPT and Gemini are now queried the way their answer appears to people in the application, ChatGPT with web search. What is measured is the share of answers in which your brand is mentioned, and each of the four systems counts equally. Before that, every question ran once per system through the provider's interface. The values of the two methods are not directly comparable: in the history, the marker “New measurement basis” stands at this point and the line across it is dashed.
3 October 2026 Fixed rhythm, no measurement on demand Rivly measures in a fixed rhythm: on Mondays AI answers, Google and press, and on the first Monday of the month also ads, Google Business, backlinks, the Lighthouse measurement and the technology of the website. The measurement on demand is discontinued. For a quick test after a change there is the AI live check, which changes neither the index nor the history.
26 September 2026 Rivly Index, version 4: benchmark from field, AI answers and minimum level Every channel is now measured against the highest of three values: the strongest brand in your field, the strongest brand the AI names in your answers, and a minimum level. Before that only the tracked field counted: a small field gave a high number even when the AI names another brand far more often. Weeks already measured stay unchanged; the history carries a marker at this point.
17 September 2026 Rivly Index, version 3: one rule for brand and competitors A mention counts for your brand and for every competitor by the same rule. Before that, competitors were counted more strictly and therefore stood lower than they were. Weeks already measured stay unchanged; the history carries a marker at this point.
14 September 2026 Points for recommendations Every recommendation shows how many index points it can bring. For that, Rivly calculates your index a second time, with the goal of the recommendation instead of your current value. After two measurements it also shows how much of it actually happened.
13 September 2026 Winning ads: window of 30 days An ad counts as a winning ad after 30 days of run time, previously after 14 days. Short tests drop out of the count: the number of winning ads falls and says more about what really works.

New data never enters an earlier week. If the formula of the index changes, Rivly recalculates earlier weeks from the values measured at the time. If the measurement basis changes, Rivly sets a marker in the history and draws the stretch across that point dashed.