The Pipeline · Issue One, October 2026
Attention Is Not a Leading Indicator
62 compound names appear in both a measured search-demand instrument and the trials registry. The two orderings disagree so completely that neither can stand in for the other.
Read with Feature I, whose registry counts form one of the two orderings, and with Feature II, which supplies the most plausible explanation for the gap.
Two orderings of one list
Two instruments, both measured, both dated, neither derived from the other. One is the registered study count from Feature I, retrieved 2026-10-13. The other is a search-volume instrument covering the United States, measured 2026-10-12 and stored in this project’s dataset with its own method note.
Joining them is done on an exact name match and nothing cleverer: a row appears below only where the demand instrument’s key is character-for-character a corpus compound slug. That matches 62 of 77 names and leaves 15 unmatched, which are simply absent rather than approximated. A fuzzy match would be an authored judgement wearing the clothes of data.
Order the 62 by search volume and then by registered studies, and the two orderings cross almost everywhere. That picture is the whole argument, and it is drawn rather than asserted.
The two measures are never added, and they are never combined into a single index.
The two measures are never added, and they are never combined into a single index. They are different instruments measuring different things on different timescales, and the only honest operation to perform on them together is to compare their orderings.
Where the orderings cross
In one direction, attention far ahead of registration. GHK-Cu carries 442,730 in measured monthly volume against 3 registered study records. MOTS-c carries 174,530 against 5. TB-500 carries 136,370 against 1. CJC-1295 carries 70,210 against 1. BPC-157, the name most associated with this category in public, returns 4 registered records.
In the other direction, registration far ahead of attention. Gonadorelin returns 1,164 registered records against 10,370 in monthly volume. Somatropin returns 818 against 17,260. Triptorelin returns 298 against 2,300. Lixisenatide returns 75 against 20.
The names in each direction are taken from the extremes of a mechanical ranking of the joined set, volume against registered records. That ranking is a device for choosing which rows to name. It is not published as a metric and nobody should quote it as one.
The agreement at the very top is precisely what makes the disagreement everywhere else easy to miss.
There are names where the two instruments agree, and they are the ones everybody already talks about: semaglutide and tirzepatide sit near the top of both orderings. The agreement at the very top is precisely what makes the disagreement everywhere else easy to miss.
| Compound | Monthly searches | Registered records | Of which recruiting |
|---|---|---|---|
| GHK-Cu | 442,730 | 3 | 1 |
| MOTS-c | 174,530 | 5 | 3 |
| TB-500 | 136,370 | 1 | 1 |
| CJC-1295 | 70,210 | 1 | 0 |
| Gonadorelin | 10,370 | 1,164 | 118 |
| Somatropin | 17,260 | 818 | 45 |
| Triptorelin | 2,300 | 298 | 38 |
| Lixisenatide | 20 | 75 | 2 |
| Semaglutide | 393,840 | 781 | 168 |
| Tirzepatide | 1,272,620 | 298 | 97 |
Why this is not a paradox
Four things about the instruments, before anyone builds a theory on the gap.
The demand figure is monthly and current. The registry figure is cumulative over decades. A compound that was registered heavily in the nineteen-nineties and is quiet now scores high on one and low on the other for reasons that have nothing to do with the argument here. Gonadorelin and somatropin are both in that position.
The demand instrument returns a null for terms its advertising platform restricts, and this category is restricted. A null in that file is an absence of advertiser data and never a measured zero, which is why the volumes used here are the instrument’s own per-compound aggregates with their null counts carried alongside rather than treated as zeroes.
Three separate demand instruments exist in this project’s dataset and they are not interchangeable. Only one is used on this page, it is named at every figure, and no figure from the other two appears anywhere in this issue.
A compound name is not a product. Search volume under a name aggregates people looking for very different things, and so does a registry count.
The desk’s reading
The reading, and it is a reading.
Attention is not pointing at what is coming. It is pointing at what is currently obtainable by another road.
Demand in this category does not anticipate development. It substitutes for it. The names with the largest gap between attention and registration are, with striking consistency, names that reach a person through a compounding or research-supply route rather than through an approved application, which is exactly the channel Feature II found the federal publication record is preoccupied with. Attention is not pointing at what is coming. It is pointing at what is currently obtainable by another road.
If that is right, then treating search demand as a pipeline signal gets the direction of causation backwards, and an operator who builds a position on it is taking channel risk while believing they are taking development risk. That is the most expensive kind of mistake, because it is priced as the wrong thing.
This desk should declare its own position on the point. Measured demand for the subject of this publication, the business and regulatory shape of the sector, is near zero: the obvious head terms are measured in the low hundreds of searches a month and the specific ones in the tens. This magazine is not a demand asset and was not built as one. It exists because the registers it reads are public, checkable and almost never read side by side, and that is the only justification it has.
Sources, and the queries that produced every figure above
- Google Ads search-volume instrument, United States, held as data/keywords/volumes-us.json. Measured for this project on 2026-10-12 and stored with its own method note. Retrieved 2026-10-12. One of three separate demand instruments in the dataset. The dataset’s own note states that the three must never share a denominator, so only this one is used here and it is named at every figure. A null in this instrument is an absence of advertiser data, never a measured zero.
- ClinicalTrials.gov Data API, version 2, studies endpoint. U.S. National Library of Medicine, National Institutes of Health. Retrieved 2026-10-13. One query per corpus compound name. totalCount is the number of study records the registry returns for the term on the retrieval date. The full result set, one row per compound with the URL that produced it, is published in data/quarterly/clinicaltrials.json.https://clinicaltrials.gov/api/v2/studies?query.intr=<compound name>&countTotal=true&pageSize=1
- Peptide research corpus: 77 compound records and 77 evidence profiles. This project’s validated dataset, data/compounds and data/evidence. Counts are taken directly from the dataset files. A derived tier on an evidence profile is labelled INFERRED in the data and carries its own basis string, which is reproduced wherever a tier is printed.