Measured data

Why does this page exist?

The words "fast", "secure" and "search-friendly" mean nothing until there is a number, a method and a date behind them. Every RootCR quote includes a measurement report; this page is the public version of the same thing.

Every figure states when and how it was measured. Where the measurement has limits, those are written out too — in the second to last section.

1. How much AI traffic do the sites RootCR builds receive?

Language models do not guess: they download the page. That download leaves a trace in the server log, and so it can be counted.

Period: 11 February – 25 August 2026 (195 days). Source: the access logs of RootCR's own servers. Measured: 25 August 2026.

Which system, how many times?

From classic search over the same period: Google 11,333, Yandex 8,751, Bing 6,253 and Apple 2,472 requests.

What does a "live human-triggered fetch" mean?

Most bots come to collect data for later use. One group is different — the user-triggered fetches of ChatGPT, Claude and Perplexity — because it only starts when somebody has just asked something. This is not scheduled harvesting: a person typed a question, and the assistant opened this page to answer it.

Over 195 days that happened 1,350 times on the sites RootCR built. Roughly once every three and a half hours, day and night.

Per site, as a share

2. Speed and hygiene — measured against 25 Hungarian developers' own sites

If somebody sells professional websites, their own site is their business card. So on 25 August 2026 RootCR measured the homepages of 25 Hungarian web developers and agencies — with exactly the same method, in the same hour, from the same machine as its own five sites.

Of the 25, 19 responded; six either returned a 403 error to a datacentre machine or did not answer at all within thirty seconds.

Measured 25 August 2026, median of five requests per page, with a desktop Chrome user agent. Values in ms are server response time (TTFB).

What this does not say: RootCR is not the fastest in the field. Three competitor sites responded faster (18, 31 and 49 ms in the 25 August 2026 measurement) — from behind a content delivery network. Those same sites, however, load 89 and 90 JavaScript files and up to 23 trackers, from 522–713 KB (same measurement).

Where there is no contest: RootCR runs zero tracking scripts on all five of its sites, while exactly one of the 19 responding competitors has none (25 August 2026). This is not a setting, and it does not depend on the visitor's consent: the code is simply not there.

Why does speed matter?

There is a second reason, specific to AI search: the user-triggered fetches of ChatGPT and Perplexity are response-time sensitive. A source that answers slowly gets dropped and the next candidate takes its place. Response time is therefore not only a user experience metric but a filter on whether an AI assistant can cite you at all.

3. The method — so you can check it

Counting AI traffic. From the servers' access logs, based on the user agent string of each request. Every request counts once, against the first matching system. Three identifiers fall into the "live human-triggered fetch" category: ChatGPT-User, Claude-User and Perplexity-User — according to their vendors' documentation these only ever start in response to a user's question.

The speed measurement. Five consecutive requests per page, and the median of those — not the best and not the average, so that a single outlier cannot distort it. With a desktop Chrome user agent, following redirects, with a thirty-second timeout. HTML size, JavaScript file count and tracker count come from the downloaded HTML by pattern matching.

How the field was selected. The homepages of the developers and agencies appearing on the first page of Hungarian Google results for the Hungarian searches weboldal készítés (website development), webshop készítés (web store development) and webfejlesztő ügynökség (web development agency). Nothing was cherry-picked: whatever was on the list was measured.

Why are there no names? Because the comparison works without them, and a list of names would add nothing to its truth. If you want the full, named table, ask for it and it will be sent.

4. What this measurement does not tell you

This section exists so that nothing has to be explained away afterwards.

• The measurement was taken from a datacentre machine in Germany. From a Hungarian home connection the numbers come out differently. The ratios hold; the absolute values do not.
• Six competitor sites did not respond. That does not necessarily mean they are slow: it is more likely they filter datacentre IP addresses. For that reason they are excluded from the averages in the table.
• User agent strings can be faked. Some bots may lie about themselves. These numbers are therefore upper bounds; their order of magnitude, however, is stable across the whole log.
• This page says nothing about visitor numbers. It measures downloads by AI systems, not how many customers arrived. The latter is the client's data, not RootCR's — where it is available and publishable, it appears in the case study.
• Speed is not everything. A fast site can still be a bad site. Speed is here because it can be measured objectively, not because it is the most important thing.

5. What does this mean for you?

If you are commissioning a site today, there are two questions worth asking any developer — including RootCR:

1. What number can they show about their own work, and what was the method? If only adjectives come back, that is not an answer.
2. How many third-party tracking scripts will run on my site, and who gets access to my visitors' data?

RootCR's own answers are on this page, in checkable form. Ask for a quote, and your site will get the same kind of report at handover.

Frequently asked questions

Updated: August 25, 2026