EverGrow SEO Solution Website-Check

Own measurements.
With method and limitations.

These are measurements I made myself: in server logs, Search Console, web analytics and with my own fetches. Each study states its data basis, its method and what it cannot show. I use retailer data only anonymized.

About the author · Auf Deutsch lesen

Measurement log
Studypublishedperioddata basis
AI bots in robots.txt published period February to October 2026 457,000 German and 12 million home pages worldwide (HTTP Archive), 77 robots.txt files from large shops, one server log
AI assistants in a shop published period 30 Sept–25 Oct 2026 Server logs, web analytics and Search Console of a large online shop
ChatGPT buy box published period 22–24 Oct 2026 1,008 answers, US and Germany, signed out
Merchant listings, EEA published period 9 Aug–23 Oct 2026 531 product searches, Search Console, Merchant Center
Merchant listings published period 15 Apr–18 Oct 2026 Search Console, Merchant Center, several shops
ChatGPT utm outage published period 30 Sept–15 Oct 2026 Server logs and GA4 of a large online retailer
AI readability published period 11 Oct 2026 My own fetches of 100 large online shops
published
Period
HTTP Archive monthly datasets from February to October 2026, tracked on the same home pages
Data basis
HTTP Archive (BigQuery): 457,225 German and 12.0 million home pages worldwide with robots.txt rules; 77 complete robots.txt files from large shops; server logs of one large shop
Counting method
named = explicitly listed as a User-Agent; block indicator = entry with Disallow without Allow (a full block in 54 of 62 entries across 77 shop robots.txt files)

How many German websites and shops block AI bots in robots.txt

4.6 % of German home pages name GPTBot in robots.txt; 2.5 % have Disallow without Allow (block indicator), compared with 8.3 % and 3.5 % worldwide. The picture reverses for large websites: up to rank 100,000, 28.8 % of German pages name GPTBot (23.0 % with Disallow without Allow), compared with 13.9 % worldwide excluding Germany. On the same pages in Germany, naming rises from 3.5 to 4.4 % between February and October.

In short

  • 4.6 % of German home pages name GPTBot in robots.txt; 2.5 % have Disallow without Allow. Worldwide, the figures are 8.3 % and 3.5 %. Across all 13 AI identifiers checked, Germany is below the worldwide figure overall.
  • The picture reverses by size: 28.8 % of German home pages up to CrUX rank 100,000 name GPTBot (23.0 % with Disallow without Allow), compared with 13.9 % worldwide excluding Germany. Germany is still slightly above the worldwide figure up to rank 1 million, but below it in the group up to 10 million: the German long tail has fewer entries than the worldwide one.
  • On the same home pages in Germany, GPTBot naming rises from 3.5 to 4.4 % between February and October 2026; PerplexityBot roughly doubles, and OAI-SearchBot rises to 2.7 times its earlier level. Of 77 large German shops, 4 fully block GPTBot; for OAI-SearchBot, four of seven entries contain only Allow.

Measurements in this study

of German home pages name GPTBot

4.6 %

457,225 .de home pages with robots.txt rules, October 2026 crawl; worldwide 8.3 %

of German websites up to rank 100,000

28.8 %

1,088 home pages with a CrUX rank up to 100,000; worldwide excluding Germany 13.9 %

GPTBot named on the same German pages

3.5 → 4.4 %

432,997 home pages in both datasets, of which 352,715 and 360,897 respectively have rule groups; Disallow without Allow 2.4 → 2.6 %

large German shops fully block GPTBot

4 of 77

robots.txt files from 11 October 2026; 11 name GPTBot, 3 explicitly allow it

Read the study Auf Deutsch lesen

published
Period
30 September to 25 October 2026; purchases 1 to 7 and 16 to 25 October
Data basis
Server logs, web analytics (consented visitors) and Google clicks of a large online shop, anonymized
Levels
Bot fetches, attributed visits and purchases analysed separately

AI assistants in a large online shop: fetches, visits and purchases

AI bots account for around 1.1 % of all requests. 92.8 % of ChatGPT visits to product pages were to products previously fetched by OAI-SearchBot; only 6.5 % had a live fetch in the preceding hour. ChatGPT sessions had 1.5× as many purchases per session as organic Google sessions.

In short

  • AI bots account for around 1.1 % of all requests. More than half of these come from meta-externalagent, followed by OAI-SearchBot (15.0 %) and PerplexityBot (11.8 %). Between 77 and 83 % of live fetches by ChatGPT-User and Claude-User go to product pages.
  • 92.8 % of ChatGPT visits to product pages were to products previously fetched by OAI-SearchBot. This does not establish selection: OAI-SearchBot fetched 69.8 % of all active products in 26 days. The more reliable finding is that only 6.5 % had a live fetch in the preceding hour.
  • ChatGPT sessions had around 1.5 times as many purchases per session as organic Google sessions, and around 2.3 times as many when landing on a product page. This describes groups, not a cause. 75.5 % of ChatGPT sessions start on a product page, compared with 57.8 % for organic Google.

Measurements in this study

of requests come from AI bots

1.1 %

Share of all requests, 30 Sept–25 Oct 2026; rounded denominator

of ChatGPT product visits had a live fetch beforehand

6.5 %

ChatGPT-User on the same page in the preceding hour; previously fetched by OAI-SearchBot: 92.8 %

as many purchases per session as organic Google

2.3×

ChatGPT, landing on a product page; all landing pages 1.5 times; web analytics, consented visitors

of products visited via ChatGPT had no Google click

16 %

in 90 days of Search Console data; periods do not coincide; Jaccard index 0.34

Read the study Auf Deutsch lesen

published
Period
22 to 24 October 2026
Data basis
1,008 answers without a reported capture error, 504 each for the US and Germany; 168 questions, each asked twice per day
Price comparison
31 mixed-source US answers for 17 product questions/models; cross-check of all four apparent price cases

ChatGPT buy box and merchant feeds

US: 65.4 % web cards, 30.7 % OpenAI feed. The cheapest confirmed model card comes first in 23 of 31 mixed-source answers. Four apparent feed price premiums do not establish an actual price disadvantage after review. Germany: no defined OpenAI feed marker; 20.5 % Google Shopping cards.

In short

  • In the US, 65.4 % of cards are classified as web cards; 30.7 % carry the defined OpenAI feed marker. Web cards occupy the first card position in 71.4 % of cases. Cards without the defined source markers are common in this measurement; it does not establish the actual visibility of products without a feed. A web card does not prove that the merchant supplies no feed.
  • The cheapest confirmed model card comes first in 23 of 31 mixed-source US answers. After cross-checking, four apparent cases with a more expensive feed card first do not establish a price disadvantage for the same purchasable offer. The sample is small; this does not disprove preferential treatment either.
  • In Germany, none of the 1,201 cards carry the defined OpenAI feed marker. 20.5 % carry Google Shopping markers. This applies to ChatGPT while signed out during the measured period, not to all interfaces or the actual feed participation of all merchants.

Measurements in this study

web cards in the US

65.4 %

868 of 1,327 cards without a defined source marker; OpenAI feed: 30.7 %

web cards in first position in the US

71.4 %

300 of 420 first card positions; OpenAI feed: 26.7 %

cheapest confirmed model card first

23/31

Mixed-source US answers; four apparent price cases, no confirmed price disadvantage

Google Shopping cards in Germany

20.5 %

246 of 1,201 cards; none with the defined OpenAI feed marker

Read the study Auf Deutsch lesen

published
Period
Shop data 9 August to 18 October 2026, my own measurement 20 to 23 October
Data basis
Data from shops I work with, with permission (anonymized), here Shop A, plus my own measurements: 531 product searches on four days in DE, AT, CH and UK
Baseline
1 to 14 September 2026 = 100

Merchant listings in the EEA

In Germany, 90 % of product items in the new grid lead to comparison sites. Shop A: Search Console merchant listings −63 %, Merchant Center −9 %, organic clicks on product pages +6.5 % (previous year −15.6 %).

In short

  • Since mid-September 2026, Google has shown a grid of comparison sites instead of merchants for product searches in Germany: 90 percent of the product items in the block lead there. In Austria it is 46 percent; in Switzerland and the UK, the carousel stays as before.
  • In Search Console (search type Web), clicks on “Merchant listings” in the EEA drop from 15 September by 63 percent in Shop A. The Merchant Center, in contrast, counts only −9 percent for all free listings, compared with −11 percent in the same period of the previous year.
  • In Shop A, the gain in organic clicks on product pages made up for a good half of the loss from merchant listings on these pages by calculation: without merchant listings, they rose by 6.5 percent, while they fell by 16 percent in the same period of the previous year. All clicks from web search were 9 percentage points above the previous year’s trend.

Measurements in this study

of product items in the German grid lead to comparison sites

90 %

Mobile, all groups, four measurement days, range 89–91 %; Google Shopping 10 %

merchant listings in Search Console, EEA

−63 %

Shop A, search type Web, clicks 21 Sept–4 Oct vs 1–14 Sept 2026; Google’s new layout in the EEA

free product listings in Merchant Center, EEA

−9 %

Shop A, same period; previous year −11 %

organic clicks on product pages, Shop A, EEA

+6.5 %

Without merchant listings, same period; previous year −15.6 %, so 22 percentage points better

Read the study Auf Deutsch lesen

published
Period
15 April to 18 October 2026
Data basis
Search Console and Merchant Center, several shops, anonymized
Baseline
15 April to 14 May 2026 = 100

Merchant listing impressions

From May 16 to October 8, 2026, Search Console counted 3 to 7 times the desktop impressions for merchant listings in several shops, CTR fell to 0.2 to 0.9 %. On October 9 it was over everywhere.

In short

  • Starting on 16 to 18 May 2026, desktop impressions for merchant listings rose in each of the shops to 3–7 times the April level within five days and stayed there for almost five months. Nothing rose on mobile.
  • Clicks did not follow. Desktop CTR fell from 0.8 to 9.4 percent to 0.2 to 0.9 percent. Almost half of the impressions came from countries where the shops did not get a single click.
  • On 9 October it ended in all shops on the same day. Since then, desktop impressions have been at 4 to 6 percent of the April level, and desktop CTR is at the level of mobile. I read this as a change in counting, not as lost traffic.

Measurements in this study

desktop impressions, merchant listings

3–7×

25 May–4 Oct vs 15 Apr–14 May 2026: Shop A 6.5×, Shop B 3.1×, Shop C 5.8×

desktop CTR in the same period

0.2–0.9 %

Previously 0.8 to 9.4 %; clicks were at 57 to 83 % of the April level

impressions from countries without a single click

46–63 %

1–14 Sept, all devices, Shop A and C; Shop B: 42 % from countries with fewer than five clicks

end in all shops on the same day

9 October

Since then desktop impressions at 4 to 6 % of the April level, desktop CTR 5 to 17 %

Read the study Auf Deutsch lesen

published
Period
30 September to 15 October 2026, by hour in UTC
Data basis
Server logs and GA4 of a large online retailer, anonymized, plus my own spot checks
Baseline
1 to 7 October, same hour = 100

ChatGPT utm outage

Tagged ChatGPT clicks −73 to −79 %, ChatGPT live fetches only about −20 %: mostly the tag was missing, not the visibility.

In short

  • The break came within an hour: on 8 October between 21:00 and 22:00 UTC, tagged ChatGPT clicks per 100 live fetches fell from 40 to 18, and to 10 an hour later. Afterwards, they were 73 to 79 percent below the previous week.
  • Demand fell much less: ChatGPT-User fetches were only around 20 percent below the previous week. Instead of 34, only 9 to 11 tagged clicks arrived per 100 fetches. Most of the gap was missing tagging.
  • The tag has been returning since 15 October, at this retailer from the afternoon (UTC). That week cannot be repaired in your reports, only marked and put into context.

Measurements in this study

tagged ChatGPT clicks

−73 %

Per day, 9–12 Oct vs 1–7 Oct; on 13–15 Oct even −79 %

ChatGPT-User fetches

−20 %

Same period, confirmed OpenAI addresses only; 13–15 Oct −23 %

clicks per 100 fetches

34 → 11

1–7 Oct vs 9–12 Oct; on 13–15 Oct only 9

offset through “direct”

≈ 55 %

Additional direct mobile landings on product pages, equivalent to ≈ 55 % of missing ChatGPT landings; measurable only in part of the shop (GA4, 9–14 Oct)

Read the study Auf Deutsch lesen

published
Measurement
Snapshot from 11 October 2026
Data basis
100 large B2C shops in Germany, my own fetches with an honest identifier, respecting robots.txt
Product pages
Up to three per shop, 56 shops measurable

AI readability of online shops

ChatGPT search is blocked at only 1 of 90 shops. But only 38 % provide a GTIN, only 16 % mark up variants as ProductGroup.

In short

  • ChatGPT search is blocked in robots.txt at only 1 of 90 shops. 4 of the 5 shops that block GPTBot (training) leave ChatGPT search open.
  • 49 of 56 measurable shops have Product markup in the raw HTML, but only 38 % provide a GTIN and only 16 % mark up variants as ProductGroup.
  • At 3 of 54 shops, the price appears only after JavaScript runs. According to an analysis by Vercel, the crawlers of ChatGPT and Claude do not execute JavaScript.

Measurements in this study

blocks ChatGPT search

1 of 90

OAI-SearchBot in robots.txt

reject unknown bots

31 %

31 of 100 shops: error code 403 or 429, or a challenge page

provide a GTIN

38 %

Anywhere in the HTML, among 56 shops with measurable product pages

respond fast enough

61 %

INP within the good range, the weakest of the three loading metrics

Read the study Auf Deutsch lesen