Colophon and accessibility

Editorial policy · Last reviewed 11 August 2026

Editorial independence and disclosures

This is my personal website, maintained in a personal capacity. The views expressed here are my own and do not represent Bayer AG or any other current or former employer.

The editorial rule is simple: confidential or proprietary information is not published. References to organisations and professional work are included for professional, educational and historical context. Company names and marks remain the property of their respective owners; no endorsement of this website is implied.

I correct material errors openly. If you believe something is inaccurate, confidential, misattributed or used without appropriate rights, contact me.

Accessibility

This site aims for WCAG 2.2 AA, and the reason is personal.

My dyslexia, diagnosed only in 2017 after years of teaching others, taught me the same lesson from the inside: education fails when it cannot adapt to the learner.

So the site is built the way I wish more reading material were: a high-legibility typeface, a comfortable line length of roughly 66 characters, strong colour contrast, visible focus states, full keyboard navigation, and respect for reduced-motion preferences. If anything here is hard for you to use, tell me and I will fix it.

How it is built

The site is static, fast and content-first: built with Astro, with the long-form copy stored as Markdown files, hosted on Cloudflare Pages.

The typefaces are open source and self-hosted: Inter by Rasmus Andersson for body text, Source Serif 4 by Frank Grießhammer for Adobe for headings, and Caveat by Impallari Type as an occasional handwriting accent. All three are used under the SIL Open Font License, and each face’s licence ships with the site: Inter, Source Serif 4 and Caveat.

Traffic is measured with Plausible Analytics. The point is to learn which public pages are useful, not who is reading them. Plausible records aggregate patterns such as page views, referring sites, approximate location, browser and device type, and which files are downloaded. It sets no cookies, creates no persistent identifiers, stores no raw IP addresses, and does not follow people across sites or devices. Visitor data is processed and stored in the EU; Plausible’s data policy explains the detail.

Built to be read by machines too

My doctorate found that we now live in a referrer society: people check, compare, and increasingly ask an AI. So this site treats language models as a third audience alongside readers and search engines, an idea from my own work supervising academic study on Generative Engine Optimisation.

In practice: every page carries schema.org structured data describing one consistent entity; the content is written in clean, answer-shaped HTML; robots.txt deliberately welcomes reputable AI crawlers rather than blocking them; and an experimental llms.txt file offers models a canonical summary. That last standard is young and adoption is uncertain, which is exactly why trying it is cheap.

And the measurement is public. Field Test 001, the pre-launch baseline publishes the transcripts of what AI models said about me before this site went live as a citable dataset, with any editorial redaction marked: the before measurement the site is built to move.

A companion strand measures the other half of the question. The Bot Crawls 001 records not what the models say about me but what their software actually did at the edge in the three days around launch: which crawlers came, what they took, and one hour of requests wearing AI crawler names to probe for credentials, every figure recomputed from banked Cloudflare responses.

This page is the case study of its own thesis: if AI systems increasingly retrieve, cite and accurately reuse evidence from this site, the experiment is working.

Corrections and changes

This website is regularly subjected to structured adversarial review, using human judgement and AI-assisted analysis to test the accuracy and consistency of its claims, and its overall quality.

This is a record of material editorial and structural changes, not every new publication, typographical correction or layout adjustment.

August 2026

23 August 2026: Published Field Test 006: The paid model follows the link, an observational comparison of two Claude sessions conducted in incognito windows on 11 August 2026. Standardised production attribution across Field Tests 001 to 005 by adding a shared Production note to each study, retaining its original publication date and recording 23 August 2026 as last amended. The notes separate study conduct, evidence provision and editorial direction from AI-assisted evidence ingestion, analysis, drafting, review and implementation: Field Test 001 credits Dr Kishan Rees, Anthropic Claude Fable 5 and OpenAI Codex Sol 5.6; Field Tests 002 to 005 credit Dr Kishan Rees and Anthropic Opus 4.8 (Claude Code); Field Test 006 credits Dr Kishan Rees and OpenAI GPT-5.6 Sol (Codex). Models tested in the sessions are not listed as production contributors, and no transcript evidence was rewritten.
16 August 2026: Corrected the first-published date on the four Ideas entries that went live with the site: B2LLM, MEDSIN, (PBL-SGT)-Fusion and The referrer society. Each carried the date it was written and reviewed on the private bench, 16 or 17 July 2026, rather than the date it was first published to readers. All four are now dated 19 July 2026, the day drkishanrees.com went live. The change applies to the page, its Markdown companion and the structured data. No wording was altered.
13 August 2026: Updated all 15 records in the speaking archive with evidenced session dates, attendance mode and locations, while leaving unknown times unstated. Corrected the date of the Declining Web Traffic panel from 26 April to 23 April 2026. Reclassified the archive’s structured data from Event to CreativeWork, because these pages document completed sessions rather than public, bookable event listings.
13 August 2026: Standardised AI contribution attribution across all Dispatches. Each page now uses a shared Production note containing publication history, verified contributor labels and roles, and final editorial responsibility. After checking the native task records, The bots arrived. The citations did not. now identifies OpenAI GPT-5.6 Sol (Codex) and Anthropic Opus 4.8 (Claude Code); Two desks, one accountable editor identifies OpenAI GPT-5.6 Sol (Codex) and Anthropic Opus 5 (Claude Code); and The First Prompt identifies OpenAI GPT-5.6 Sol (Codex) and Anthropic Opus 5 (Claude Code). The First Prompt now credits Sol for the archive-series concept, Dispatches index prototype and implementation handoff, rather than incorrectly describing Kish’s supplied manuscript as a Codex initial draft.
13 August 2026: Corrected the numbering of the current Dispatch series to follow publication chronology: 01, The bots arrived. The citations did not.; 02, Two desks, one accountable editor; and 03, Three tasks, six conversations. The index displays the series newest-first while retaining those stable numbers. The independent Dispatches from the Archive sequence, including issue 001, is unchanged. Existing routes, canonical URLs and publication dates were preserved, and automated checks now guard against an established Dispatch URL disappearing or changing.
11 August 2026: Published a selective public executive profile in HTML, PDF and Markdown, generated from one reviewed source. It brings together career chronology, leadership scope, qualifications, appointments and professional standing while omitting confidential or proprietary information.
11 August 2026: Refined the D:CAM description to focus on the methodology and public record. A performance figure without an accompanying public source was removed from the site and its machine-readable summaries.
11 August 2026: Consolidated the personal media-audience estimate on the Broadcast page, where its scope and limitations are stated alongside the public record; added aggregate Facebook Insights retained in a 2017 MBA report and linked the 6 April 2020 appearance to contemporaneous Sky News, X and Facebook publications.
11 August 2026: Clarified professional registration and the archival status of WatMed Media; marked a redaction in Field Test 001; and added this editorial policy and public corrections record.
11 August 2026: Reorganised the primary and footer navigation alphabetically: Learn was added, and Story moved from the navigation to the Biography page.
11 August 2026: Revised the 2025 Reuters Pharma DSO session summary to focus on small teams and bridging to the late majority, condensed the marketing-mix discussion, and added a link from the page to contemporaneous public evidence.
9 August 2026: Learn launched publicly: free, practical lessons for working critically, confidently and responsibly with AI.
9 August 2026: The Dispatches index was rebuilt as an integrated reader for current field reports and numbered archival returns.
8 August 2026: Dispatches from the Archive launched with The First Prompt, preserving the original work alongside present-day evidence and commentary.

July 2026

31 July 2026: Corrected the publication date on Bot Crawls 002 from 30 to 31 July 2026 and added navigation between the Bot Crawls studies.
24 July 2026: Dispatches launched with its first field report.
23 July 2026: Bot Crawls 001 published and the public Bot Crawls record added.
19 July 2026: Site launched with its accessibility approach, crawler policy and experimental language-model summary published alongside the public Field Tests record.