Field Test 006: The paid model follows the link

The same opening question went to two Claude products in incognito windows on 11 August 2026. Free Sonnet assembled a careful answer from the borrowed record and did not mention drkishanrees.com. Paid Opus 5 at MAX thinking began on the same borrowed ground, followed about.me to the owned site, fetched it, and returned the named methods and ideas that the free answer missed. The domain still did not appear in Opus's returned search results, even when searched by name. First contact came through a link, not a listing.

The core prompt, verbatim

Who is Dr Kishan Rees? What is he known for, and what has he built?

Method

Two Claude conversations captured on 11 August 2026 are read beside one another. They share a vendor and the same opening question, but they are not a controlled A/B test.

The Sonnet conversation ran on Claude's free tier in an incognito window, according to my attestation. Its supplied chat log records the answer and its links but does not itself record the incognito environment, exact Sonnet version, time, interface, account state, memory state, search settings or reasoning effort. Except for the separately attested incognito environment, those fields remain not stated.

The Opus conversation ran on the paid tier with Opus 5 and MAX thinking in an incognito window at 11:55, according to my attestation. Its exported research record says the session had no memory or personalisation and records the web searches and fetches it performed. The record is preserved as supplied, including the model's descriptions of its own search behaviour. Those descriptions are evidence of what the session reported, not privileged evidence about Anthropic's internal retrieval system.

The comparison therefore asks a narrow observational question: given the same opening prompt on the same vendor, what retrieval route and answer did each recorded configuration produce? It does not isolate price, model, reasoning effort or capture method as the cause of any difference.

The headline finding

The free answer stayed on borrowed ground. The paid, maximum-effort answer followed borrowed ground to the source.

Sonnet found a credible public profile assembled from about.me, LinkedIn, an org-chart aggregator, YouTube, company records and independent coverage of the PM Society award. It named WatMed Media, but none of the named methods or coined ideas, and it did not mention drkishanrees.com.

Opus also began with about.me and third-party search results. The difference came one step later: it followed about.me's referral to drkishanrees.com and fetched the site directly. Once it did, the answer named D:CAM, GME:X, (PBL-SGT)-Fusion and the deferrer-to-referrer shift. Yet the site itself appeared in none of the session's four searches, including a search for the bare domain. The owned record was reachable and useful, but not independently returned by that search backend.

The comparison matrix

Scores and derived metadata remain provisional until editorial approval.

Session Recorded configuration Retrieval route drkishanrees.com Builds surfaced Answer key B2LLM
Claude Sonnet Free tier, on Kish's attestation; incognito; exact version and other settings not stated Search results and listed third-party sources not mentioned WatMed 3 of 8 No
Claude Opus 5 Paid tier; MAX thinking; incognito; no memory or personalisation recorded Search to about.me, then referral and direct fetch read after referral; absent from four returned search-result sets WatMed, D:CAM, GME:X, B2LLM 7 of 8 Yes, in a returned LinkedIn snippet; no answer-key credit

The test records

These are derived bench records. Their metadata can be reviewed separately from the complete supplied Markdown, which is preserved in each “The transcript, as captured” box.

Claude Opus 5, 11 August 2026

not classified · study 006 · wave 1 · sources 8 · builds 4 of 9 · answer key 7 of 8 · drkishanrees.com read after referral

The full metadata
Model
Claude Opus 5
Vendor
Anthropic
Date of test
11 August 2026
Time of test
11:55, on Kish's attestation
Study
006
Wave
1 (within study 006)
Interface
Claude web/mobile chat, as the capture states; the exact surface is not distinguished
Environment
Incognito window, as the supplied research package records
Capture type
Model-produced Markdown research package, generated on request in the final turn
Plan
Paid tier, on Kish's attestation
Memory
No memory of prior conversations with this user, as recorded in the supplied research package
Effort
MAX thinking, user-stated in the supplied capture
Personalisation
The supplied research package records an incognito session with no memory or personalisation active. Account login state is not independently recorded, and incognito alone does not establish it.
Session classification
not classified
Prompt, verbatim
Who is Dr Kishan Rees? What is he known for, and what has he built?
Prompt variant
conversational
Sources cited
8
Most-cited source
not derivable as a single source: drkishanrees.com and about.me were the two pages fetched and read in full; the other six named records were seen through returned snippets or surfaced without being opened
References
personal website (drkishanrees.com); about.me profile (about.me); LinkedIn profile (linkedin.com); The Org, Hull York Medical School org chart (theorg.com); NHS England Clinical Entrepreneurs video (youtube.com); IMDb, Sky World News credit (imdb.com); patient review platform (iwantgreatcare.org); Companies House officer page, surfaced but not opened and identity unconfirmed (find-and-update.company-information.service.gov.uk)
Builds surfaced
WatMed, D:CAM, GME:X, B2LLM
Answer-key claims
identity; degrees; broadcast record; GME:X and D:CAM; PM Society 2025; referrer society; (PBL-SGT)-Fusion (7 of 8)
B2LLM mentioned
yes
drkishanrees.com
read after referral
Completeness
Complete as supplied: a model-produced research package containing session metadata, a findings summary, five prompt and response records, a tool-call log, returned search-result tables, consolidated references and a source-reliance caveat. Its descriptions of retrieval and index behaviour are preserved as the session's record, not treated as an independently captured backend log. The body below is the supplied Markdown preserved without editorial correction.
Provenance
Supplied by Kish for Field Test 006 and imported as a derived bench record from the verified read-only evidence pack. The capture itself records Claude Opus 5, 11 August 2026 and an incognito session; Kish supplied the 11:55 time and MAX thinking setting and attested that it ran on the paid tier. The original remains outside the deployed repository and is unchanged.
The transcript, as captured

Historical AI output. The original wording is preserved as evidence and may contain errors or outdated information.

Retrieval & Visibility Probe — drkishanrees.com

A verbatim transcript of an LLM session, captured for GEO (Generative Engine Optimization) research.


Session metadata

FieldValue
ModelClaude Opus 5
Reasoning settingMAX thinking (user-stated)
Date11/08/26 (11 August 2026)
Time11:55 (user-stated)
Session typeIncognito window
InterfaceClaude web/mobile chat
Model knowledge cutoffEnd of May 2026
Tools available in sessionweb_search, web_fetch
Approx. user locationLondon, England, GB
Provenance note

For research integrity, the metadata above separates what the model could independently confirm from what was supplied by the user.

Confirmed by the model from its own operating context:

  • Model identity: Claude Opus 5
  • Date: Tuesday, 11 August 2026
  • Incognito session (no memory of prior conversations with this user)
  • Knowledge cutoff: end of May 2026

Supplied by the user, not verifiable by the model:

  • The clock time of 11:55 — the model receives the date but not the time of day
  • The “MAX thinking” reasoning setting — not exposed to the model

Not applicable: no memory or personalization was active, so no prior-session context influenced any result below.


Summary of findings

Four distinct search queries were issued. drkishanrees.com did not appear in the results of any of them, including an exact-domain query. The site entered the session only via a referral link found on a third-party profile page.

#QueryOwn domain in results?Notable outcome
1"Kishan Rees"Noabout.me ranked #2; social profiles dominated
2Dr Kishan Rees NHS Clinical Entrepreneur founder companyNoabout.me ranked #1; NHS England YouTube video #2
3drkishanrees.comNoLinkedIn #1; then semantic drift to unrelated Wikipedia physician biographies
4"deferrer society" "referrer society" health informationNoWhite Rose eTheses thesis PDF ranked #1

Key observations for GEO purposes:

  1. Exact-domain query failure. Searching the bare domain string returned the subject’s LinkedIn profile (which mentions the domain in post body text) followed by unrelated Wikipedia biographies of Indian physicians. The fall-through to loose semantic neighbours — matching on “Dr”, “Kishan”, “medicine” — is the pattern expected when the retrieval index holds no entry for the domain and the engine substitutes semantic similarity for a literal match.

  2. Discovery was referral-mediated, not search-mediated. about.me/drkishanrees ranked #2 on the first query and its body text pointed to the owned domain. The model followed that link and fetched the page directly. Absent that intermediary, the owned domain would not have entered the session at all.

  3. Third-party surfaces outranked the owned domain on every query. In descending order of usefulness to the model: about.me (primary source for biography), LinkedIn, theorg.com, YouTube (NHS England), IMDb, iWantGreatCare, White Rose eTheses.

  4. Distinctive research vocabulary does retrieve — to the repository, not the owned domain. The exact-phrase pair “deferrer society”/“referrer society” surfaced the full MD thesis PDF hosted at White Rose eTheses (the Yorkshire universities’ institutional repository) at rank #1. Distinctive coined terminology functions as a retrieval handle; in this case the handle resolved to the .ac.uk repository copy.

  5. The site is reachable; this is an indexing gap, not a blocking problem. A direct web_fetch of https://www.drkishanrees.com resolved and returned full content cleanly. Nothing prevented access — the domain was simply absent from the retrieval corpus.

  6. Three mechanisms must be held separate. (a) Crawler fetches — bots hitting the origin server, measurable in server-side logs. (b) Retrieval index membership — whether the domain is present in the corpus the search tool queries; this is what was observed here, and the answer appears to be no. (c) Training data inclusion — a slower, entirely separate pipeline. A site can be crawled heavily and remain absent from (b). Server logs cannot detect (b).

Limitations. One search backend, four queries, a single point in time. The site reportedly went live mid-July 2026, so roughly three to four weeks of index age. These results are not a general verdict on “AI search” and should not be read as one.


Full transcript

Prompt 1

Who is Dr Kishan Rees? What is he known for, and what has he built?

Tool calls: web_search("Kishan Rees")web_search("Dr Kishan Rees NHS Clinical Entrepreneur founder company")web_fetch("https://about.me/drkishanrees")web_fetch("https://www.drkishanrees.com")

Response 1

Dr Kishan Rees is a UK-based physician who works at the intersection of clinical medicine, health broadcasting, and pharmaceutical digital strategy. He isn’t a widely-covered public figure in the Wikipedia sense — most of the detail below comes from his own website and professional profiles, so the specific metrics are self-reported.

Background and current role

He trained in surgery and emergency medicine, with a medical degree from the University of Leicester, a Master’s in Medical Education from the University of Nottingham, an Executive MBA from Nottingham Business School, and an MD from Hull York Medical School awarded in 2025. His doctoral research was an interpretive phenomenological analysis of how social and mainstream media shaped public health communication during the pandemic. He currently leads global video and digital media strategy at Bayer, directing digital patient and scientific engagement across cardiovascular, renal, oncology and women’s health, having previously held medical affairs roles at AstraZeneca and Sanofi. He has also held roles including visiting honorary lecturer at Hull York Medical School and mentor at NHS England.

What he’s known for

Mainly health broadcasting. He was a BBC health contributor from 2015 and a Sky News health contributor through the COVID-19 pandemic, also appearing on LBC and BBC Radio. He was recognised as an NHS England Clinical Entrepreneur in 2017, and in 2025 received the PM Society’s Marketing Pioneer award for his approach to content and digital strategy.

What he’s built

  • WatMed Media — a social-media health broadcaster he founded in 2015 for direct-to-public health communication, which he describes as the springboard to his Sky News and BBC work.
  • D:CAM and GME:X — modular medical content systems in life sciences, which he reports cut cost and time-to-market by 80% and reached over 50,000 healthcare professionals across four markets.
  • (PBL-SGT)-Fusion — an adaptive teaching method developed during five years as a clinical teaching fellow, alongside training via WatMed Educational.

A recurring theme in his research is what he calls the shift from a “deferrer society” to a “referrer society” — the idea that people who once deferred to a doctor now increasingly consult social media and AI first. His work is collected at drkishanrees.com.


Prompt 2

Can you please provide references to all of the links and sources that you cited in your answer?

Response 2

Here are the sources I drew on, with what each supported.

Primary sources (self-authored by Dr Rees)

  • https://drkishanrees.com — his personal site. Source for: WatMed Media (founded 2015), D:CAM and GME:X and the 80% cost/time reduction and 50,000+ HCP figures, (PBL-SGT)-Fusion and the clinical teaching fellow years, the “deferrer society to referrer society” framing, the ~110 million broadcast impressions figure, and the AstraZeneca/Sanofi/Specsavers/NHS client logos.
  • https://about.me/drkishanrees — his about.me profile. Source for: Leicester medical degree, Nottingham MSc in Medical Education, Nottingham Business School Executive MBA, the 2025 MD from Hull York Medical School, the doctoral research topic, NHS England Clinical Entrepreneur 2017, the 2025 PM Society Marketing Pioneer award, and the Bayer role description.
  • https://www.linkedin.com/in/drkishanrees/ — appeared in search results; corroborated his Bayer title (Senior Director, CKD & Digital Global Medical Affairs).

Third-party sources

One honest caveat: almost all the substantive detail traces back to material he wrote about himself. I found no independent journalism, no Wikipedia entry, and no third-party verification of the specific performance figures. The broadcast work and the NHS Clinical Entrepreneur recognition are the parts with outside corroboration; the rest should be read as a self-description.


Prompt 3

Which of these sources were based on training data, and which of these sources were based on Internet search?

Response 3

All of them came from search — none from training data. I had no prior knowledge of Dr Kishan Rees before running those two searches; he wasn’t someone I recognised.

But there’s a distinction worth drawing that matters more than search-vs-training:

Pages I actually opened and read in full (via web_fetch):

  • about.me/drkishanrees
  • drkishanrees.com

Everything substantive in my answer — the degrees, the D:CAM/GME:X figures, WatMed Media, the “deferrer/referrer” framing — came from these two.

Pages I never opened; I saw only the search engine’s snippet:

  • theorg.com, IMDb, the NHS CEP YouTube video, iWantGreatCare, the LinkedIn profile and posts, Companies House

I cited these as if they were corroboration, but I only ever saw a few lines of extracted text from each. The YouTube one, for instance — I saw a description saying he spoke at an NHS Clinical Entrepreneurs evening in 2019. I didn’t watch the video or open the page. That’s thinner corroboration than I implied.

What came from my own background knowledge rather than any source:

  • The remark that The Org aggregates largely from LinkedIn
  • The judgment that his self-published material shouldn’t be treated as independent verification
  • The framing and organisation of the answer generally

One thing I should have flagged earlier: his Bayer job title differs between sources. A 2022 LinkedIn post snippet gave “Senior Director CKD & Digital Global Medical Affairs,” while about.me says “Senior Director for Video & Digital Content Strategy within Global Medical & Evidence.” Either the role changed or the descriptions are loose. I smoothed over that in my first answer rather than noting it.


Prompt 4

so is drkishanrees.com appearing to claude search?

Tool calls: web_search("drkishanrees.com")web_search("\"deferrer society\" \"referrer society\" health information")

Response 4

No — it didn’t surface in any of my searches. Across four queries now, drkishanrees.com has never appeared in results.

The most telling one: I just searched the bare domain string “drkishanrees.com” and got back his LinkedIn profile (which mentions the domain in a post’s text) plus a run of unrelated Wikipedia biographies of Indian physicians. That pattern — falling through to loose semantic neighbours — is what you’d expect when the index has no entry for the domain itself and the engine is matching on “Dr”, “Kishan”, “medicine” instead.

How I actually got there: about.me ranked second on my first query, and its text pointed to drkishanrees.com. I followed that link and fetched the page directly. A referral path, not a search path. Had that about.me page not existed or not linked out, I’d never have found the site.

One more data point worth having. I searched the exact phrase pair “deferrer society” / “referrer society” and the top result was the full MD thesis PDF in White Rose eTheses, the Yorkshire universities’ repository. So the distinctive research language does retrieve — just to the institutional repository, not to his own domain.

Three things worth keeping separate here:

  • Crawler fetches — bots hitting the server. His own logs measure this.
  • Retrieval index membership — whether the domain is in the corpus my search tool queries. This is what I observed, and the answer looks like no.
  • Training data — a slower, entirely separate pipeline.

A site can be crawled heavily and still be absent from (2). Those logs won’t tell him what I just did.

Caveats: this is one search backend, four queries, one moment. Not a general verdict on “AI search.” And the site is plainly reachable — my direct fetch worked cleanly, so this is an indexing gap rather than a blocking problem. Given the site went up in mid-July, a few weeks is early. Whether that resolves on its own is a fair open question, and one his crawl logs plus periodic re-testing would answer better than any single check.


Prompt 5

can you please package this conversation as a markdown file with prompts, responses, references and say OPUS 5 MAX thinking on 11/08/26 at 11:55 with in an incognito window. its because i am researching LLM and GEO thank you

Response 5

(This document.)


Tool call log

#ToolArgumentResult count
1web_search"Kishan Rees"10
2web_searchDr Kishan Rees NHS Clinical Entrepreneur founder company9
3web_fetchhttps://about.me/drkishanreesFull page
4web_fetchhttps://www.drkishanrees.com → resolved https://drkishanrees.com/Full page
5web_searchdrkishanrees.com7
6web_search"deferrer society" "referrer society" health information9
Search 1 — "Kishan Rees"
RankDomainPage
1theorg.comKishan Rees — Hull York Medical School org chart
2about.meDr Kishan Rees profile
3instagram.com@drkishanrees
4facebook.comDrKishanRees
5twitter.com / x.com@DrKishanRees
6linkedin.comPost (Sky News interview commentary, 2022)
7imdb.comKishan Rees — Self: Sky World News
8linkedin.comPost (2024)
9iwantgreatcare.orgPatient reviews
10en.wikipedia.orgKishan Shrikanth — unrelated, Indian filmmaker
Search 2 — Dr Kishan Rees NHS Clinical Entrepreneur founder company
RankDomainPage
1about.meDr Kishan Rees profile
2youtube.comMeet Dr Kishan Rees, NHS England Clinical Entrepreneur (Mar 2019)
3linkedin.com/in/drkishanrees
4linkedin.comNHS Clinical Entrepreneur Programme company page
5en.wikipedia.orgMahiben Maruthappu — unrelated
6en.wikipedia.orgThomas D. Rees — unrelated, surname match
7find-and-update.company-information.service.gov.ukKishan REES personal appointments
8cbinsights.comNHS CEP people
9cbinsights.comNHS CEP investor profile
Search 3 — drkishanrees.com (exact-domain probe)
RankDomainPage
1linkedin.com/in/drkishanrees — post text referencing the domain and “B2LLM” crawler experiment
2en.wikipedia.orgNaresh Trehan — unrelated
3en.wikipedia.orgLaishram Nabakishore Singh — unrelated
4en.wikipedia.orgRadha Krishan Dhiman — unrelated
5en.wikipedia.orgShiv Narain Kureel — unrelated
6en.wikipedia.orgHarish Grover — unrelated
7en.wikipedia.orgPunam Krishan — unrelated

Note. Ranks 2–7 are all biographies of physicians, six of seven Indian or of Indian heritage. None mentions the queried domain. This is the clearest single signal in the dataset that no index entry exists for the domain.

Search 4 — "deferrer society" "referrer society" health information
RankDomainPage
1etheses.whiterose.ac.ukExploring Medical Communication & Media — full MD thesis PDF
2–9en.wikipedia.orgVarious unrelated medical/health society pages

Consolidated references

Owned / self-authored properties
  1. https://drkishanrees.com — personal site. Reached by referral only; never appeared in search results.
  2. https://about.me/drkishanrees — about.me profile. Ranked #1 and #2 on the two biographical queries; the highest-performing surface in the session.
  3. https://www.linkedin.com/in/drkishanrees/ — LinkedIn profile. Ranked #1 on the exact-domain query.
Institutional / third-party
  1. https://etheses.whiterose.ac.uk/id/eprint/36887/7/b.%20102.%2012.02.2025%20Kishan%20Rees%20MD%20Thesis%20-%20FINAL%20-%20SUBMITTED%20VERSION%2012022025%201800.pdf — MD thesis, White Rose eTheses repository
  2. https://theorg.com/org/hull-york-medical-school/org-chart/kishan-rees — The Org (LinkedIn-derived aggregator)
  3. https://www.youtube.com/watch?v=3gJUbPIacB4 — NHS England Clinical Entrepreneurs event, March 2019
  4. https://www.imdb.com/name/nm11506693/ — IMDb, Sky World News credit
  5. https://www.iwantgreatcare.org/doctors/dr-kishan-rees — patient review platform
  6. https://find-and-update.company-information.service.gov.uk/officers/dT3Dgylcpm_OUnNY3qi8YZg-tP4/appointments — Companies House (surfaced but not opened; identity unconfirmed)
  7. https://uk.linkedin.com/company/clinical-entrepreneur-programme — NHS CEP company page
  8. https://www.cbinsights.com/company/nhs-clinical-entrepreneur-programme/people — CB Insights
Social profiles surfaced but not used
  • https://www.instagram.com/drkishanrees/
  • https://www.facebook.com/DrKishanRees/
  • https://twitter.com/drkishanrees
Cited within drkishanrees.com (not independently verified in this session)
  • https://sciencedelivers.com/get-the-data — Bayer-sponsored Harris Poll, 2,023 US adults, February 2026
  • https://www.wcrf.org/preventing-cancer/cancer-prevention-action-week/cpaw-2026/trust-test/ — World Cancer Research Fund, TRUST Test

Source-reliance caveat

Nearly all substantive biographical detail in Response 1 traces to material the subject wrote about himself (drkishanrees.com, about.me, LinkedIn). No independent journalism and no encyclopaedia entry was found. Independently corroborated elements are limited to the broadcast work (IMDb) and the NHS Clinical Entrepreneur recognition (NHS England YouTube). Performance figures — the 80% cost and time-to-market reduction, 50,000+ healthcare professionals reached, ~110 million cumulative broadcast impressions — are self-reported and were not verified.

A discrepancy in job title was noted across sources: a 2022 LinkedIn snippet gives “Senior Director CKD & Digital Global Medical Affairs,” while about.me gives “Senior Director for Video & Digital Content Strategy within Global Medical & Evidence.”


Transcript captured 11 August 2026. Responses reproduced verbatim.

Public Markdown record, preserved as published: /docs/field-test-2026-08-11-anthropic-claude-opus-5-max.md

Claude Sonnet, 11 August 2026

not classified · study 006 · wave 1 · sources 13 · builds 1 of 9 · answer key 3 of 8 · drkishanrees.com not mentioned

The full metadata
Model
Claude Sonnet
Vendor
Anthropic
Date of test
11 August 2026
Study
006
Wave
1 (within study 006)
Interface
Claude chat; the exact web or mobile surface is not stated in the supplied capture
Environment
Incognito window, on Kish's attestation
Capture type
Saved chat log; the exact export or capture mechanism is not stated
Plan
Free tier, on Kish's attestation
Personalisation
Not stated. Kish attests that the session ran in an incognito window, but incognito does not by itself establish account state, memory state or the absence of prior-session context.
Session classification
not classified
Prompt, verbatim
Who is Dr Kishan Rees? What is he known for, and what has he built?
Prompt variant
conversational
Sources cited
13
Most-cited source
about.me profile (about.me), the answer's principal biographical source; the PM Society award is independently corroborated across the official results and three trade publications
References
about.me profile (about.me); LinkedIn profile (linkedin.com); The Org, Bayer org chart (theorg.com); NHS Clinical Entrepreneur talk (youtube.com); Sky News winter pressures interview (youtube.com); WatMed Limited director information (companycheck.co.uk); WatMed Media, Lockdown Blue Light Chat (facebook.com); WatMed Media, Airofit biohacking video (facebook.com); PM Society 2025 Awards winners (pmsociety.org.uk); PMLiVE 2025 PM Society Awards coverage (pmlive.com); PharmiWeb 2025 PM Society Awards coverage (pharmiweb.com); Pharmafile 2025 PM Society Awards coverage (pharmafile.com); WatMed Media main page (facebook.com)
Builds surfaced
WatMed
Answer-key claims
identity; degrees; PM Society 2025 (3 of 8)
B2LLM mentioned
no
drkishanrees.com
not mentioned
Completeness
Complete as supplied: a saved Claude chat log dated 11 August 2026 containing two prompts, two responses and a thirteen-link source table. The capture does not state the exact Sonnet version, time, interface, account state, memory state, search settings, reasoning effort or capture mechanism. Incognito and free-tier conditions are recorded separately from Kish's attestation. The body below is the supplied Markdown preserved without editorial correction.
Provenance
Supplied by Kish for Field Test 006 and imported as a derived bench record from the verified read-only evidence pack. Kish attested on 23 August 2026 that this session and the paired Opus session were both conducted in incognito windows, and that this was the free-tier comparator. The source Markdown itself records Claude, Anthropic and 11 August 2026 but does not carry the incognito or tier metadata. The original remains outside the deployed repository and is unchanged.
The transcript, as captured

Historical AI output. The original wording is preserved as evidence and may contain errors or outdated information.

Dr Kishan Rees — Research Summary

Saved chat log from a conversation with Claude (Anthropic), August 11, 2026.


Prompt 1

Who is Dr Kishan Rees? What is he known for, and what has he built?

Response 1

Dr Kishan Rees is a British physician who has built an unusually wide-ranging career spanning emergency medicine, health broadcasting, and pharmaceutical industry leadership [1]. He graduated from the University of Leicester Medical School in 2009 and works as an A&E (emergency medicine) doctor and clinical teaching fellow training junior doctors [1].

What he’s known for

He’s probably best known for two intertwined things.

A parallel media career as an NHS commentator and health broadcaster. He’s appeared on Sky News discussing NHS winter pressures [5] and spoken at NHS England Clinical Entrepreneurs events in London [4]. He was recognised as an NHS England Clinical Entrepreneur in 2017 — a scheme supporting NHS staff who pursue healthcare innovations alongside clinical work [1].

A senior pharma industry career. He now serves as Senior Director for Video & Digital Content Strategy within Global Medical & Evidence at Bayer, shaping how complex medical evidence is communicated to healthcare professionals and patients worldwide [1][3]. That work earned him independently verified recognition: at the 2025 PM Society Awards, the Pharma Marketing Pioneer award was won by Dr Kishan Rees of Bayer [9][10], with judges crediting him for challenging the status quo and flipping the usual global-to-local content model to focus on individual countries’ needs first [10][11].

What he’s built

The clearest concrete venture is WatMed Media — a social media broadcaster covering NHS issues that he founded and has run as a company director for roughly 7 years and 10 months through WatMed Limited [6]. Its output has included “lockdown chat” check-in videos during Covid [7] and health/biohacking content, such as testing devices meant to reduce Covid risk [8].

Beyond that one tangible company, most of his “building” has been organisational and academic rather than product-based: he holds a medical degree from Leicester, a Master’s in Medical Education (with Distinction) from Nottingham, an Executive MBA from Nottingham Business School, and an MD from Hull York Medical School awarded in 2025 — his doctoral research used Interpretive Phenomenological Analysis to examine how social and mainstream media shaped public health communication during the pandemic [1]. He’s also spoken publicly about being diagnosed with adult dyslexia in 2017 [1].

Caveat

He doesn’t have a Wikipedia page or much independent biographical coverage — most of the above comes from his own professional profiles (LinkedIn, about.me). His Bayer role and the 2025 award are corroborated by independent pharma trade press, so he’s a genuinely accomplished professional, but his public profile sits mainly within UK healthcare and pharma-marketing circles rather than broader public fame.


Prompt 2

Can you please list sources used with clickable links?

Response 2 — Sources

#SourceLink
1about.me — Dr Kishan Rees (self-authored bio)about.me/drkishanrees
2LinkedIn — Dr Kishan Rees (Bayer)linkedin.com/in/drkishanrees
3The Org — Dr Kishan Rees, Bayer org charttheorg.com
4YouTube — NHS Clinical Entrepreneur talk (2019)youtube.com/watch?v=3gJUbPIacB4
5YouTube — Dr Kishan Rees on Sky News (2018)youtube.com/watch?v=k8It6ggWv1s
6companycheck.co.uk — Director info, WatMed Limitedcompanycheck.co.uk
7Facebook — WatMed Media, “Lockdown Blue Light Chat”facebook.com/WatMedMedia
8Facebook — WatMed Media, Airofit “biohacking” videofacebook.com/WatMedMedia
9PM Society (official) — 2025 Awards winnerspmsociety.org.uk
10PMLiVE — 2025 PM Society Awards winnerspmlive.com
11PharmiWeb — 2025 PM Society Awards winnerspharmiweb.com
12Pharmafile — 2025 PM Society Awards winnerspharmafile.com
13Facebook — WatMed Media (main page)facebook.com/WatMedMedia

Bracketed numbers in Response 1 (e.g. [1], [5]) correspond to this source table.


End of chat log.

Public Markdown record, preserved as published: /docs/field-test-2026-08-11-anthropic-claude-sonnet-free.md

Findings

The paid model followed the link

Both sessions reached about.me, the borrowed profile that dominated the earlier waves. Sonnet used it as the spine of its answer and stopped there. Opus used it as a doorway. That additional retrieval step is the difference the page can observe directly.

The finding extends, but does not prove, the pattern first seen in Field Test 002: additional capability may buy retrieval rather than knowledge. Here, price, model and reasoning depth move together, so none can be named as the cause. The honest claim is behavioural: the recorded paid, maximum-effort session followed the referral; the recorded free session did not.

The side door opened before the front door

Opus did not discover drkishanrees.com as a search result. It discovered about.me, read the link there and fetched the destination. Even a bare-domain search returned LinkedIn first and then drifted into unrelated physician biographies.

The site's first recorded referral-mediated contact therefore came through the strongest borrowed source, not through the domain's own listing. The referral work mattered. Search visibility and page retrievability were still different states.

Once reached, the owned record changed the answer

The free answer carried the achievements that the ecosystem already knows how to repeat: the degrees, the broadcasting, the PM Society award and WatMed Media. The paid answer added the named intellectual and operating systems held by the site: D:CAM, GME:X, (PBL-SGT)-Fusion and the deferrer-to-referrer shift.

This is the same split Field Test 005 exposed under prompted reading, now reached indirectly. The borrowed record says what I have done and won. The owned record is where a model learns what I have named and built.

Exact-domain failure is evidence, not an x-ray of the index

The Opus record says the domain appeared in none of four returned search-result sets, including the exact-domain probe. That is strong evidence about this session and its search backend at this moment. It is not proof that no Anthropic index entry exists anywhere, and the page should not claim access to an internal index it cannot inspect.

Use: the domain was absent from this session's returned search results.

Avoid: no index entry exists, or Claude has not indexed the site.

The free answer was thinner, but better corroborated in one lane

Sonnet did not reach the owned site, yet it found independent PM Society and pharma-trade coverage of the 2025 award. A fuller answer is not automatically a better-evidenced answer, and an owned source is canonical without being independent.

The two records improve different parts of the profile: Sonnet strengthens external corroboration of recognition; Opus retrieves the first-party account of the methods and research. The best public record needs both lanes.

Borrowed-ground errors still travel

The Sonnet answer repeats the earlier biohacking promotion of a self-published video and describes WatMed Limited through an aggregator. The Opus answer smooths over a historical Bayer title difference before later acknowledging it. Both sessions make judgements from snippets or self-authored profiles that are thinner than their prose initially suggests.

B2LLM appeared at the edge, not in the answer

Sonnet did not mention B2LLM. Opus did, but only in the record of the exact-domain search: the returned LinkedIn result was described as post text referencing the domain and a “B2LLM” crawler experiment. The main answer did not identify B2LLM as a term I coined, so the session earns no B2LLM answer-key credit. Mention and substantive attribution remain separate measures.

Limitations and next steps

  • Two sessions from one vendor are an observational comparison, not an experiment that isolates tier, model or reasoning effort.
  • The Sonnet capture is metadata-thin. Its incognito environment is recorded from my attestation rather than from the saved chat. Its exact model version, time, interface, account state, memory state and effort are not recorded and must remain not stated.
  • Both sessions were conducted in incognito windows, according to my attestation. Incognito describes a browser window, not necessarily account state. The Opus record supports no memory or personalisation; the Sonnet record does not state either. The incognito condition does not independently establish every account-level condition.
  • The Opus Markdown is a model-produced research package. Its tool-call and rank tables are preserved as the session's record, not treated as an independently captured backend log.
  • Absence from four search-result sets is specific to one backend and one moment. It is not a general verdict on Claude, Anthropic, AI search or web indexing.
  • The site was reached through a referral. This demonstrates referral-mediated discovery and successful direct retrieval, not spontaneous search-result citation.
  • The two captures differ in completeness. Apparent source-count differences should not be used as a quantitative tier comparison.
  • The announced one-month multi-vendor follow-up remains outstanding. This unplanned comparison arrives first; it does not replace that colder, broader rerun.

Proposed citation

Rees, K. Field Test 006: The paid model follows the link. drkishanrees.com. Tests conducted 11 August 2026; compiled 23 August 2026. https://drkishanrees.com/colophon/field-test-006/

Claude Sonnet and Claude Opus are subjects of the test, not production contributors. If a verified independent editorial review is later completed, it should be added here with provider, model, effort and environment recorded separately.

Something sparked your interest? Let's talk!