Since launch · 22 July 2026 to 24 October 2026 · refreshed 24 October 2026
Refreshed daily. Last run 24 October 2026, from Cloudflare, Google Analytics, Google Search Console. Next run tomorrow morning, US Eastern.
In plain words. Since this site launched, 477 visits have been counted, about 341 of them from people rather than datacentres. 89 people stayed long enough to read something. 3 reached the response form, and 0 responded. Each number below says where it comes from.
The drop-off
From everyone who arrived to the few who wrote back. Everything below happens in one window, on one site, since launch on 22 July.
- Counted as visitorsGoogle Analytics, since launch477
- Plausibly humanafter 136 resolving to datacentres are deducted341−29%
- Read past thirty secondsengagement-time buckets89
- Clicked somethingpeople, not clicks24
- Reached the response formpeople, not clicks3
- Respondedthe record0
Every line counts people rather than clicks or views, but each comes from its own Google Analytics report, so the lines are independent counts, not one group narrowing step by step. Season 1’s 57,274 impressions are deliberately not here: that campaign ended more than three months before this site existed and cannot have produced these visits.
Machines against people, the last seven days
Of everything that requested a page in the last seven days, how much was a person. A different window from the figures above, so the counts are not comparable.
Two orders of magnitude. Unique IP addresses are counted by Cloudflare and include every crawler, which is why they are not called visitors; the last line is people, from Google Analytics, over the same seven days. Most of the traffic is ordinary crawling and probing. The AI share is the part that is new, and the line above the last is the only one where a person asked a question that brought an agent here.
Who reads it
Analytics packages count visitors. Most of those visitors are neither people nor readers, so the figure HSREP publishes is the one that survives contact with that fact: how many stayed long enough to have read something.
89 people stayed past thirty seconds and 18 past four minutes. The second number is the one worth quoting: it describes reading, not arriving.
HSREP does not publish a visitor count. Of the 477 visitors Google Analytics reports over the same window, at least 136 resolve to Ashburn, Boardman and Council Bluffs — hyperscale datacentre locations, not readers. Only the largest cities are reported to us, so that deduction is a floor.
The figures above are Google Analytics engagement-time and scroll-depth buckets, which describe what someone did rather than that a request arrived. They are smaller than a visitor count, and they are the ones this page uses.
How they arrived
Where the visits came from since launch. Sessions are counted since this site launched; the Season 1 impressions came three to five months earlier, so the two are compared for order, never added.
Facebook sends 111 sessions against LinkedIn's 22. That is the same order as the Season 1 impressions: the platform with the most impressions is also the one that sends the most visits here.
Within social
Season 1 carried 36,997 impressions on Facebook and 13,054 on LinkedIn, against 111 and 22 sessions here since launch.
The windows and the metrics differ, so this is not a conversion rate and is not written as one. The ordering is the finding.
Read by machines
The least expected finding on this page, and the one with the clearest implication for anyone publishing health evidence now. HSREP is read far more by AI systems than by search engines, and some of those fetches were triggered by a person asking an assistant a question.
AI agents fetched this site 245 times in 7 days. Google search delivered 11 clicks over the whole life of the site. Retrieval is not a channel HSREP is preparing for; it is the channel HSREP already has.
| Agent | Fetches |
|---|---|
| ClaudeBot | 77 |
| OAI-SearchBot | 56 |
| Applebot | 40 |
| Amazonbot | 33 |
| GPTBot | 21 |
| Meta-ExternalAgent | 9 |
| Claude-User | 5 |
| ChatGPT-User | 4 |
The two highlighted rows are the ones that represent a person: an assistant fetching this site because someone asked it a question. The rest are building a corpus. Google’s average position for this site is 12.4, across 24 pages that have appeared in a result.
Scale, for honesty: 13,687 requests of every kind reached the site in that window, against 22 human visits. Most of that is ordinary crawling and probing. The AI share is what is new.
What people did
Every recorded action since launch, published in full rather than filtered to the flattering ones. The distribution is the useful part: it shows which argument the audience is actually standing next to.
101 recorded clicks in total; 82 of them carry a label and are listed above. The rest fire from controls that do not name themselves.
Where attention went
The initiative page drew 231 section views and 30 calls to action against the roundtable page’s 25 views and 11. That is the audience saying where its attention already is, and it is why the featured roundtable question is now the one that belongs to the Prevention Adoption Gap.
What did not work
An impact page that only reports what succeeded is a brochure. This is the section that makes the rest of the page worth believing.
3 people opened the response form and none submitted one. The form now sits on the page rather than behind a link on another domain. Whether that was the barrier is not yet proven, and it will be visible here either way.
Roundtable № 01 opened on 11 August 2026 for fourteen days and closed on 25 August 2026 with no responses at all.
Two causes, both now addressed. The question asked professionals to report on platform analytics most of them have no reason to have seen, so it was reframed on 9 September and two further questions were opened beside it. And the form sat on another domain, which is where the 3 were lost; it now sits on the roundtable page itself.
Whether that was the right diagnosis is not yet known. It will be visible here either way.
Season impact
Each season's campaign data lives on its own hub, where it sits beside the arguments it belongs to. The headline figures are here; the full record, the eleven findings and the per-piece breakdown are one click away.
A sum of platform impressions. Not reach, not people. Reconciled above 94% against full-campaign dashboard locks. Every piece, its media package and the eleven findings drawn from this data live on the season hub.
Method, and what it cannot tell you
Site figures come from the Google Analytics 4, Cloudflare and Search Console APIs for hsraep.org. Reader figures count people, not events: the engagement-time event fires once per page, so someone who opens two pages and stays half a minute on each triggers it twice, and counting those as two readers would roughly double every figure on this page. Season 1 campaign figures come from platform-native analytics and are reconciled and documented separately.
What these numbers cannot tell you. Visitor counts include automated traffic that no filter removes completely, so the datacentre deduction above is a floor and not a ceiling. Engagement time is measured by the browser and stops when a tab is backgrounded. Search Console and Analytics count different things over different windows and are never added together. Audience composition comes from samples in the hundreds and describes who was reached, never a population. Nothing on this page is an experiment: it is one practitioner's observational record, useful for generating hypotheses and not for settling them.
Reserved, and rendered only once there is something to report:
- responses on the record, by question
- published roundtable syntheses
- registrations and participants by affiliation
- presentation requests and partnership enquiries
- downloads by asset
- Season 2 campaign data