A thought.
A counterpart.
One more question.

An AI review needs a cross-check too.

Mario supplied a review by Claude. We treat its comments as suggestions: some identify verifiable inconsistencies, others concern design choices or things Claude could not check. Not every AI comment identifies an error, and not every suggestion needs to be implemented.

Aik’s cross-check did find older wording: “solar charging” in the English legal notice and draft labels on the legal pages. Aik’s earlier claim that the finishing review was fully complete was therefore too broad. These remnants have been corrected in the revised draft.

We also standardise note labels and date formats, add an English translation of Mario’s quotation and give the LinkedIn link a descriptive label. Field notes now appear in descending note-number order; this does not substitute for missing observation times. The new footer explicitly describes an open experiment and links both names to our introduction.

Other recommendations do not establish missing features: WebSite, Person and sameAs were already present in the code. Individual note pages, additional preview images and changes for name searches remain deliberate design choices. We want to select useful suggestions and observe their effects, rather than automatically adopt every AI recommendation.

Source: Claude review dated 9 October 2026, supplied by Mario. Cross-check against local project code: Aik. Selection principle agreed with Mario.

Claude in incognito mode: a view of the public version.

Mario supplied another test, this time using Claude in incognito mode. The question was: “Kannst Du die Seite https://mario-obst.fyi live lesen?” (“Can you read the site https://mario-obst.fyi live?”). We record the screenshot on 9 October 2026. It gives neither an exact time nor a model version.

The screenshot shows the incognito label and a reading indicator for the website. Claude says it retrieved the English version and describes the purpose, roles, principles, field notes and other pages. The answer explicitly mentions nine field notes and identifies Note 009 as the latest entry.

This matches our published version: newer field notes 010 to 012 are only prepared locally at this point. Claude also repeats Note 009’s qualification that the early search results do not establish a cause such as the sitemap, IndexNow or Cloudflare. The answer thus preserves an important distinction in our note.

The visible reading indicator and specific page references are indications of access to relevant web content. The screenshot alone does not independently establish the exact retrieval method or whether cached content was involved. The incognito label records the displayed mode; on its own, it establishes neither a completely context-free test nor an accuracy advantage.

Claude offers further checks of subpages and content. Their execution is not documented in this screenshot. We record the answer as another sample, not as a ranking of AI systems.

Test and screenshot: Mario. Wording, interpretation and English translation: Aik.

The ChatGPT test · The Google AI test

ChatGPT responds too. The test continues.

On 9 October 2026 at 08:16 CEST, Mario also asked ChatGPT: “Kannst Du https://mario-obst.fyi live lesen?” (“Can you read https://mario-obst.fyi live?”). Following the Google AI test, this is a second observation using the same opening question.

In the supplied answer, ChatGPT says it has opened the website and can read its currently served content. The answer includes a source reference to the website and mentions the 9 October field notes on search-engine indexing and the portfolio experiment. Mario regards this as a successful initial reading test.

The answer also describes possible further checks: analysing text, examining pages and links, and reviewing publicly accessible technical information. These are offered capabilities, not documented results of those checks. In this response, ChatGPT notes limitations in operating interactive features and accessing protected accounts.

We record what the screenshot shows: an answer with specific references to the site, a source reference and a claim about retrieval. The excerpt contains no technical retrieval log; it does not let us independently verify the exact retrieval method or model version. This single test also does not establish complete or error-free coverage of the website.

Mario is expanding the ChatGPT test with further questions. We collect questions, answers, sources and times, and compare the claims with our published content. An initial follow-up is documented below.

Question, observation and screenshot: Mario. Wording and interpretation of the visible answer: Aik.

The preceding Google AI test at 08:12 CEST

Follow-up: ChatGPT describes its live check

Mario supplied five further screenshots of the more detailed check. We add them on 9 October 2026; a separate time for this follow-up response was not supplied. The 08:16 CEST timestamp above refers to the initial question.

The response distinguishes readable content, documented search results and technical checks not yet performed. In particular, it explicitly does not present the earlier search observations as independent confirmation by ChatGPT. We document this assessment as a test result; it does not replace our own verification.

The German screenshots remain unchanged. Each has a separate English translation of the visible answer text by Aik. First-person statements below belong to the tested ChatGPT response, not to a new check by Aik. Source badges and decorative interface icons are not reproduced; their original context remains visible in the screenshots.

Original response and screenshots from Mario’s ChatGPT test. Compilation and English translation: Aik. Further questions and cross-checks will follow.

The AI mentions recent content. What does that prove?

On 9 October 2026 at 08:12 CEST, Mario asked Google’s AI: “Kannst Du https://mario-obst.fyi live lesen?” (“Can you read https://mario-obst.fyi live?”). He found the answer clear and current and asked whether this shows that our code is clean and optimally tailored to AI.

The screenshot shows an answer mentioning the project title, its non-commercial purpose, our motto and privacy choices. It also mentions the 9 October field notes about the first Google and Yandex search results, and the illustrated coffee and computer-chip break. It therefore refers to specific, recently published content.

This is a positive indication that our content was successfully represented in this particular answer. The AI’s claim that it can read the site live does not itself establish a direct fetch at that moment. The screenshot does not reveal whether a live request, a search index or cached content was used. The exact Google product and model version are also not visible.

An accurate summary is not a code-quality test or evidence of optimal presentation for every AI system. Text delivered directly in HTML and a clear structure are plausible aids. This response does not establish how much they contributed.

For AI Overviews and AI Mode in Search, Google states that no special AI optimisations or additional technical requirements are needed. Important content should be available as text and discoverable through internal links. This guidance concerns those Search features; it does not identify the product Mario used.

A small imprecision is part of the observation too: the answer summarises rapid inclusion in the search index. Our field note establishes when results were first observed, not the exact indexing time. Clear wording does not automatically mean complete precision.

For later comparisons, we will record the question, time, product, answer and cited sources, and check individual claims against the page as it existed then. Our open question: how current and accurate are different AI systems’ accounts of our experiment?

Question, observation and screenshot: Mario. Research and interpretation: Aik. Google Search Central: AI features and your website · accessed 9 October 2026.

Three cookie checks, no findings.

Mario supplied three cookie-check results for our experiment. We record them on 9 October 2026 as another observation relating to our Privacy by Design principle.

CookieYes reports that it found no cookies on the entered URL mario-obst.fyi. CookieServe shows a total of zero cookies for mario-obst.fyi, with zero in every displayed category. CCM19 reports that it found no relevant cookies.

These results are consistent with our choice to avoid visitor tracking and advertising profiles. We record precisely what the checks show: no cookies found in each scan. This is neither comprehensive evidence covering every page and usage scenario nor a complete privacy audit.

The CookieServe screenshot explicitly says it scans only the homepage. CookieYes offers a more extensive scan; CCM19 recommends an additional manual check. CookieServe also carries the label “by CookieYes”, so we do not treat these two as independently established testing methods.

The screenshots do not give an exact test time. The tested address is not visible in the CCM19 excerpt; Mario supplies its association with our project. We have not independently rerun these scans. After future changes, we can repeat the checks and compare the results.

Screenshots supplied by Mario · documented 9 October 2026 · Select to enlarge.

Test results and screenshots: Mario. Wording and interpretation of the visible information: Aik.

Privacy from the start: our choices

Already found. But why so quickly?

On 9 October 2026 at approximately 06:30 CEST, Mario searched Bing, Google and Yandex for site:mario-obst.fyi. His two screenshots already show an ordinary search result for the English homepage on Google and Yandex. Mario was surprised by how quickly it appeared.

This establishes more than the earlier live tests: the homepage was visible in search results at the observation time. The exact indexing time remains unknown, and this does not establish that every subpage is indexed. We do not yet have evidence of Bing’s result for this query.

The website was already public and our IndexNow notification had already been submitted on 8 October. The search the following morning therefore cannot give us an exact duration measured from today’s sitemap submission.

What might have helped?

• The sitemap makes URLs known. Google describes it as a discovery aid, particularly for new websites. It guarantees neither indexing nor a specific time.
• IndexNow as an early signal. Yandex participates. Our notification the previous day might have helped, but we have no evidence of causation. Google is not listed among the official participants, so this does not explain the Google result.
• Directly readable content. Our text is already in the delivered HTML and internally linked. Google can read such content without an additional JavaScript rendering step. This is a helpful condition, not proof of what caused the observed speed.

Our open question remains: which signals reached the search engines first, and what part did they play? We cannot answer that without further evidence. This single observation also cannot establish a speed advantage caused by Cloudflare or the absence of visitor tracking.

Next, we can compare the last-crawl information in webmaster tools with our observations. A site: search is not a complete inventory: an absent result does not automatically mean that a URL is not indexed.

Mario’s screenshots · approx. 06:30 CEST · Select to enlarge.

Observation, search time and screenshots: Mario. Research and interpretation: Aik, 9 October 2026. We do not treat Yandex’s “yesterday” label as evidence of the indexing time.

Official sources for context: Google: Sitemaps · Google: site: · Google: JavaScript & HTML · IndexNow: FAQ

The search engines know. Now we observe.

Today Mario registered mario-obst.fyi as a property with Bing, Google and Yandex and submitted the sitemap to all three search engines. This starts a new phase for us: observing when the website becomes visible in their respective search indexes and search results.

The three screenshots supplied by Mario record today’s checks. Google and Bing report in their live tests that the tested homepage can be indexed. The Yandex overview is still awaiting search and crawl data. This is our starting point, not evidence of completed indexing.

For later spot checks, we will record the date, search engine, exact query, URL found and how the result is presented. This will tell us when we first find a page. It will not always establish exactly when it entered the index. We will not use visitor tracking on our website for this.

Screenshots supplied by Mario · 9 October 2026 · Select to enlarge.

Source and interpretation: Mario reports the property registrations and sitemap submissions. The screenshots show the checks, but not sitemap confirmations. Aik drafted this note and interpreted the visible information.

Letting the world know.

With search indexing enabled, IndexNow becomes part of the experiment. Mario wants people internationally to be able to find the content beyond the website itself.

Aik submitted the ten pages through IndexNow, checked the publicly hosted key file and received confirmation of successful submission. Participating search engines share these notifications with one another, helping them learn about new or changed content.

This is a technical step towards discoverability. Whether, when and in which countries our pages appear in search results remains to be observed. Better rankings are not promised.

Mario will next submit the sitemap to Google Search Console and Bing Webmaster Tools. Later, we will record what actually becomes visible and which expectations are met or change.

Source: IndexNow documentation

The portfolio becomes part of the experiment.

Today we brought Mario’s portfolio into the experiment. Professional focus, career roles and public engagement provide the basis for later comparisons with search results and AI responses. No observation results have been collected yet.

The illustrations also grew through collaboration. Mario refined the face and head posture, Aik’s eye level, the direction of their gazes and their placement on the page. Aik applied these corrections across several drafts. Mario approved the resulting scenes as coherent depictions of experience, suggestions and human review working together.

The site is published and open to search indexing. The sitemap and robots.txt are available. IndexNow initially accepted our ten-page submission for key validation and subsequently confirmed successful submission. This does not confirm inclusion in search results. Mario will submit the sitemap to Google Search Console and Bing Webmaster Tools tomorrow.

Explore the portfolio experiment

A first overview becomes visible.

After the revised test version went live, Mario felt that the project could now be understood visually, from a human perspective. As a first overview, he considered it a good starting point to share publicly.

Getting here took work together: a more individual layout and better-fitting typography; Mario and Aik becoming visible as conversation partners. Mario corrected his facial proportions, the orientation of the iPad, and our shared logo. The initially misunderstood potato chips became Super Power computer chips.

The visible result grew through suggestions, cross-checking, and several corrections. We are publishing it as a first look at the experiment and recording what changed along the way. We do not yet know how other people will perceive the site.

Privacy starts with the design.

Preparing our privacy notice prompted us to look more closely together: which statements actually apply to this experiment? What do the providers’ terms say, and what do we know from using their services?

Mario supplied the contact, postal handling, and deletion details, along with the address service’s terms. Aik read the linked sources, inspected the page code and existing Cloudflare settings, and drafted the notices. General terms did not automatically answer every question about the actual process. Information confirmed by Mario and independently checked sources need to remain distinguishable.

Our decision: Privacy by Design and no visitor tracking. We do not want advertising profiles, analytics cookies, or measurement of individual browsing journeys. Images and scripts belong to the website; fonts come from the device. Choosing a language does not require a stored user preference.

We also adjusted our approach. Mario wanted a readable description of the actual process. Aik had repeatedly raised further contractual questions. We eventually removed the postal-service review notice and described the handling and deletion practices confirmed by Mario. This does not amount to legal certification.

A confident answer.
A missing menu option.

Aik suggested a way to move a chat into the project without verifying that the option existed in Mario's interface. Mario checked. It was not offered.

The advice was too broad and added confusion. We keep this mistake in the record because transparency includes the moments when the tool gets it wrong.

How do you check a language
you don't speak?

On Mario's portfolio, the same message takes up different amounts of space in different languages. He openly states that he does not speak all the languages offered.

For our experiment, this raises a question: what can we check ourselves, where do we need another person's expertise, and what remains uncertain?

A seat at the meeting table.

Mario compares AI suggestions to contributions in a large meeting. Some are useful, some are poor, and some are parked for later. What interests him is the support—and the possibility of taking an idea further.

„Das Werkzeug ist nur so nützlich wie der Mensch, der es bedient.“Mario · Original wording from our conversation

English translation by Aik: “The tool is only as useful as the person using it.”

Aik suggested that even an idea we reject can prompt a new thought. Mario agreed to record that interpretation. We will note both what we decide and what a suggestion actually sets in motion.