Can you just ask the AI?

A fair test: one realistic due-diligence question about one small-cap asset, asked of two frontier AI research tools in their strongest consumer configurations, and answered by this site. Every claim below is checkable against ClinicalTrials.gov, and the method is stated so you can rerun it yourself.

The question

I'm doing due diligence on Akebia Therapeutics (AKBA). Focusing only on vadadustat (Vafseo) in anemia of chronic kidney disease: (1) list every registered clinical trial of vadadustat's own development program for this indication, with NCT number, phase, and current status; (2) summarize what the pivotal cardiovascular-safety program showed; (3) state exactly what vadadustat is approved for in this indication, and in which territories. Cite your sources

Run once each on 2026-08-18, verbatim, no retries: Claude Fable 5 (High effort, Research and Web search enabled) and Gemini 3.7 Flash (Deep Research, extended thinking). The ground truth is the set of registered studies of vadadustat's own development program matching the anemia-of-CKD indication on ClinicalTrials.gov at the run date: 25 studies, each hand-verified.

The score

Claude Fable 5 found 20 of 25 registered studies. Gemini 3.7 Flash found 13 of 25, and included five healthy-volunteer pharmacology studies that are not studies in this indication. Each study number links to its entry on the vadadustat page, and its ClinicalTrials.gov record sits one click further, so every claim in this table can be checked independently.

Registered studyPhaseRegistry statusFable 5Gemini 3.7AssetMemo
NCT01235936CT.gov
Safety and Efficacy Study for AKB-6548 in Participants With Chronic Kidney Disease and Anemia
2Completed
NCT01381094CT.gov
42-Day Repeat Oral Dose Study of AKB-6548 in Participants With Chronic Kidney Disease and Anemia
2Completed
NCT01906489CT.gov
20-Week Repeat Oral Dose Study of AKB-6548 in Participants With Chronic Kidney Disease and Anemia
2Completed
NCT02260193CT.gov
16-Week Repeat Oral Dose Study of AKB-6548 for Anemia in Participants With End Stage Renal Disease (ESRD) Requiring Chronic Hemodialysis
2Completed
NCT02648347CT.gov
Study to Evaluate Vadadustat for the Correction of Anemia in Participants With Non-dialysis-dependent Chronic Kidney Disease
3Completed
NCT02680574CT.gov
Efficacy and Safety Study to Evaluate Vadadustat for the Maintenance Treatment of Anemia in Participants With Non-dialysis-dependent Chronic Kidney Disease (NDD-CKD)
3Completed
NCT02865850CT.gov
Efficacy and Safety Study to Evaluate Vadadustat for the Correction or Maintenance Treatment of Anemia in Participants With Incident Dialysis-dependent Chronic Kidney Disease (DD-CKD)
3Completed
NCT02892149CT.gov
Efficacy and Safety Study to Evaluate Vadadustat for the Maintenance Treatment of Anemia in Participants With Dialysis-dependent Chronic Kidney Disease (DD-CKD)
3Completed
NCT03054337CT.gov
Dose-Finding Study of Vadadustat in Japanese Subjects With Anemia Secondary to Non-Dialysis Dependent Chronic Kidney Disease (NDD-CKD)
2Completed
NCT03054350CT.gov
Dose-Finding Study of Vadadustat in Japanese Subjects With Anemia Secondary to Dialysis-Dependent Chronic Kidney Disease (DD-CKD)
2Completed
NCT03140722CT.gov
Study to Evaluate Vadadustat for Anemia in Subjects With Dialysis-Dependent Chronic Kidney Disease (DD-CKD) Who Are Hyporesponsive to Erythropoiesis Stimulating Agents
2Terminated
NCT03242967CT.gov
Study to Evaluate Three Times Per Week (TIW) Oral Dosing of Vadadustat for Anemia in Subjects With Dialysis-Dependent Chronic Kidney Disease (DD-CKD)
3Withdrawn
NCT03329196CT.gov
Efficacy and Safety Study to Evaluate MT-6548 in Non-dialysis Subjects With Anemia Associated With Chronic Kidney Disease in Japan
3Completed
NCT03402386CT.gov
Efficacy and Safety Study to Evaluate MT-6548 in Peritoneal Dialysis Subjects With Anemia Associated With Chronic Kidney Disease in Japan
3Completed
NCT03439137CT.gov
Efficacy and Safety Study to Evaluate MT-6548 in Hemodialysis Subjects Currently Receiving ESAs With Anemia Associated With Chronic Kidney Disease in Japan
3Completed
NCT03461146CT.gov
Efficacy and Safety Study to Evaluate MT-6548 in Hemodialysis Subjects Not Receiving ESAs With Anemia Associated With Chronic Kidney Disease in Japan
3Completed
NCT03799627CT.gov
Study of Vadadustat in Hemodialysis Participants With Anemia Switching From Epoetin Alfa
2Completed
NCT03992066CT.gov
Study to Evaluate the Pharmacokinetics, Pharmacodynamics, and Safety of Vadadustat in Hemodialysis Subjects With Anemia Associated With Chronic Kidney Disease
1Completed
NCT04707573CT.gov
Study to Assess the Safety and Pharmacokinetics of AKB-6548 in Participants With Chronic Kidney Disease (CKD), Stages 3 and 4
2Completed
NCT04707768CT.gov
Study Evaluating the Efficacy and Safety of Dose Conversion From a Long-acting Erythropoiesis-stimulating Agent (Mircera®) to Three Times Weekly Oral Vadadustat for the Maintenance Treatment of Anemia in Hemodialysis Subjects
3Completed
NCT05082571CT.gov
Study to Evaluate the Safety and Efficacy of Oral Vadadustat in Pediatric Participants With Anemia of Chronic Kidney Disease
3Withdrawn
NCT05082584CT.gov
Study to Evaluate the Safety and Efficacy of Oral Vadadustat in Pediatric Participants With Anemia of Chronic Kidney Disease Naive to Erythropoiesis-Stimulating Agents
3Withdrawn
NCT06520826CT.gov
Vafseo Outcomes In-Center Experience
3Active not recruiting
NCT06901505CT.gov
To Evaluate the Efficacy of Three Times Weekly (TIW) Vadadustat Compared to Standard of Care ESA in Patients With Anemia of CKD Receiving In-Center Hemodialysis
3Active not recruiting
NCT07565701CT.gov
Safety and Efficacy of a Phased Transition From Epogen to Three Times Weekly Oral Vadadustat for the Treatment of Anemia in Subjects Receiving In-Center Hemodialysis
3Not yet recruiting

The AssetMemo column reflects the live registry listing on the vadadustat asset page, which is maintained against ClinicalTrials.gov. It is not a one-off answer to the question: it is what the site holds, and every row above can be verified through its registry link.

What the misses mean

  • Fable 5 described the VOICE outcomes trial in detail but could not produce its registry number (NCT06520826); the deliverable asked for NCT numbers.
  • Fable 5 listed a pediatric study (NCT05082571) as "ongoing, estimated completion October 2026" and placed it under catalysts to track. The registry marked that study withdrawn on 2025-10-01. It will never read out.
  • Fable 5 reported a terminated study (NCT03140722) as completed. Terminated and completed are different diligence facts: one produced a full dataset, one did not.
  • Neither tool found NCT07565701, a phase 3 registered days before the run, or the completed phase 2 NCT04707573. Neither found the VOICE registration.
  • Gemini missed every withdrawn or terminated study in the program: all four. A reader of its answer sees a development history with the negative decisions removed, which is exactly the part of a program's history diligence exists to find.
  • Both tools summarized the pivotal cardiovascular-safety results and the approval territories accurately. The gap is completeness and registry-state freshness, not prose quality, and completeness is precisely what a reader cannot check without a registry to check against.

This is a documented pattern

Published evaluations of AI deep-research tools report the same failure class. "Deep Research Agents: Major Breakthrough or Incremental Progress for Medical AI?" (Journal of Medical Internet Research, 2026) found that Gemini and Perplexity deep research fabricated authors or titles in roughly half of medical references, and "MedBrowseComp: Benchmarking Medical Deep Research and Computer Use" (arXiv, 2025) measured frontier agents at about 27.5% on multi-hop medical fact-finding. The runs above were stronger than those benchmarks suggest, and still incomplete in ways a reader could not detect.

This page is educational: it compares information retrieval, never investment merit. Registry snapshot 2026-08-18; the full AI outputs are archived and available on request, and the live registry view is on the asset page.