In short
Of 2,165 Google Drive links in research papers, 620 no longer open (28.6%). The share climbs with the paper's age, from 14.5% for papers first published in 2025-2026 to 76.5% for 2013-2018. Of 249 Zenodo links from the same papers, 247 still resolve.
Older links fail more often
Each link is dated by the first publication year of the earliest paper that cites it. 14.5% of links from papers first published in 2025 and 2026 already no longer open. For papers from 2013 to 2018, the share is 76.5%.
The other bands hold between 181 and 672 links each. 2013-2018 holds 102, so its range is the widest. No year before 2018 has 30 links that could be judged, so those years are grouped.
Show these figures as a table
| Published | Links checked | No longer open | 95% range |
|---|---|---|---|
| 2013-2018 | 102 | 78 (76.5%) | 67.4% to 83.6% |
| 2019-2020 | 181 | 72 (39.8%) | 32.9% to 47.1% |
| 2021-2022 | 617 | 213 (34.5%) | 30.9% to 38.4% |
| 2023-2024 | 672 | 171 (25.4%) | 22.3% to 28.9% |
| 2025-2026 | 593 | 86 (14.5%) | 11.9% to 17.6% |
| All links | 2,165 | 620 (28.6%) | 26.8% to 30.6% |
Deleted or behind a sign-in wall
A link stops working in one of two ways. Google Drive either answers with a 404 error or a page titled "Page not found", which the study counts as deleted, or it asks the reader to sign in or request access. Deletion accounts for most of the loss and grows with age, from 5.6% of links in 2025-2026 papers to 38.2% in 2013-2018 papers. The sign-in wall stays between 7.7% and 9.1% in every band from 2019 on, then jumps to 38.2% for 2013-2018.
Drive's older link format, whose IDs start with 0B, meets a sign-in wall far more often. 51 of 110 links in that format (46.4%) hit one, against 166 of 2,024 links in the current format (8.2%), and 79 of the 110 older-format links come from 2013-2018 papers. File links and folder links fail at similar rates overall, 30.0% and 27.4%.
Show these figures as a table
| Published | Links checked | Deleted | Sign-in wall |
|---|---|---|---|
| 2013-2018 | 102 | 39 (38.2%) | 39 (38.2%) |
| 2019-2020 | 181 | 58 (32.0%) | 14 (7.7%) |
| 2021-2022 | 617 | 163 (26.4%) | 50 (8.1%) |
| 2023-2024 | 672 | 110 (16.4%) | 61 (9.1%) |
| 2025-2026 | 593 | 33 (5.6%) | 53 (8.9%) |
| All links | 2,165 | 403 (18.6%) | 217 (10.0%) |
Links broken by a typesetting dash
16 links failed as printed because a typesetting dash had replaced a hyphen inside the Drive ID. A typesetter's hyphen, an en dash or an em dash looks almost like a keyboard hyphen on the page, but each is a different character, so the address no longer matches the item. Google Drive answers these links with an error page.
When we swapped each dash for a plain hyphen and tried again, 9 of the 16 opened, 2 hit a sign-in wall and 5 were deleted. The other links that could not be judged as printed had a truncated ID (13), a query string glued on or encoded (4), a server error (4), other text stuck to the end (3), a space inside the ID (2), or a shared-drive ID the check flags as malformed (1). All 43 are left out of the shares above.
As printed in the paper
drive.google.com/drive/folders/1CinmrXQ‐KCVqbtR6YLFNVd5_VgV8XAoY?usp=sharing
U+2010 hyphenGoogle Drive answers Error 400 (Bad request)
With a plain keyboard hyphen
drive.google.com/drive/folders/1CinmrXQ-KCVqbtR6YLFNVd5_VgV8XAoY?usp=sharing
U+002D hyphen-minusOpens a folder titled CONSENSUS
How Zenodo links compare
The same papers also link to research repositories, so we checked those links the same way as a benchmark. 247 of 249 Zenodo links and Zenodo DOIs resolve to their record (99.2%, 95% range 97.1% to 99.8%), including all 166 Zenodo DOIs. The 2 that fail are both on sandbox.zenodo.org, Zenodo's test site, where records are not meant to last.
In every band, a far smaller share fails than for Google Drive. From 2019 on, even the top of each Zenodo range sits below the bottom of the Drive range for the same years. The 2013-2018 row rests on a single Zenodo DOI, so it says little. figshare, OSF and Dryad links gave no usable figure (see the FAQ).
| Published | No longer open on Google Drive | No longer resolve on Zenodo |
|---|---|---|
| 2013-2018 | 76.5%78 of 102 links | 0.0%0 of 1 link, 95% range 0.0% to 79.3% |
| 2019-2020 | 39.8%72 of 181 links | 0.0%0 of 30 links, 95% range 0.0% to 11.4% |
| 2021-2022 | 34.5%213 of 617 links | 2.6%1 of 39 links, 95% range 0.5% to 13.2% |
| 2023-2024 | 25.4%171 of 672 links | 1.2%1 of 84 links, 95% range 0.2% to 6.4% |
| 2025-2026 | 14.5%86 of 593 links | 0.0%0 of 95 links, 95% range 0.0% to 3.9% |
| All | 28.6%620 of 2,165 links | 0.8%2 of 249 links, 95% range 0.2% to 2.9% |
Advice for researchers
In this study, data deposited in a research repository such as Zenodo was still reachable in almost every case checked, while more than a quarter of Google Drive links had stopped working.
- Deposit the data in a research repository and cite its DOI. All 166 Zenodo DOIs in the study resolve.
- If a Drive link has to go in the paper, open it from the final PDF in a private browser window before publication. That catches a sign-in wall and a link damaged by typesetting.
- Copy the link from the typeset proof, because typesetting is where dashes replace hyphens.
How the study was done
We searched Europe PMC for open-access papers in PubMed Central whose full text mentions drive.google.com, and collected 2,892 papers with every Drive link in each. We kept only papers under a licence that allows commercial reuse, which Europe PMC records as CC0, CC BY, CC BY-SA or CC BY-ND, and looked those licences up at 12:20 UTC on 13 October 2026. The filter removed 779 papers: 444 under CC BY-NC-ND, 174 under CC BY-NC, 47 under CC BY-NC-SA and 114 with no licence recorded. That left 2,113 papers and 2,485 Drive links. 6 of those papers could not be read because Europe PMC returned a server error, so 2,107 were read, and 2,088 of them contain at least one Drive link.
Google Drive's robots.txt disallows 202 of the 2,485 links, all 171 in the /open form, 24 uc download links and 7 others, so we did not request them. 14 more fit neither the file nor the folder form and were not requested either. The remaining 2,269 file and folder links point to 2,207 distinct Drive items. Each item was requested once between 12:41 and 14:41 UTC on 13 October 2026, one request every 3 seconds, and dated by the earliest paper that cites it.
A link counts as working when Drive shows the item by name, or when it serves the file itself, which the check tells from the MIME type in the response. It counts as deleted when Drive says the item is not found, and as a sign-in wall when Drive sends the reader to a Google sign-in or an access request. Anything else counts as unclear and is left out of the shares. The ranges are 95% confidence intervals. The 4 links that returned a server error, and the 16 with a typesetting dash, were requested again at about 14:48 UTC.
The benchmark checked 336 Zenodo, figshare, OSF and Dryad links and Zenodo DOIs from the same papers by the same rules, between 14:50 and 17:06 UTC on 13 October 2026, with each site's robots.txt and crawl delay respected on every redirect. zenodo.org timed out often, so the run was paused and resumed, and a timeout was retried rather than counted. Every link ended with a result.
Limits
- Each link was checked once, on one day. A link that was down that day counts as no longer open.
- A sign-in wall may hide a file that still exists. It counts as no longer open because a reader cannot reach it.
- The 202 links Google Drive's robots.txt disallows, and 14 links of no recognised form, were not requested, so the study says nothing about them.
- Only papers whose licence allows commercial reuse are included. Papers under a non-commercial licence may fare differently.
- 29 links counted as deleted have an ID length no working link has, so some may have been printed wrong. Without them, the overall share is 27.7%.
- The 2013-2018 band is the smallest, at 102 links, so its figure is the least precise.
- The Zenodo comparison covers Zenodo only, and compares different links from the same set of papers. It measures how often each kind of link fails, and says nothing about whether a given file would have survived elsewhere.
Download the data
The files below hold every link checked and its result, so the figures on this page can be checked and reused. They are published under CC BY 4.0. Lines starting with # at the top of each CSV hold the title, licence and citation.
google-drive-links.csv
One row for each of the 2,207 Drive items checked: the link as printed, file or folder, publication year and band, the papers citing it, the result (working, deleted, sign-in wall or unclear), the HTTP status, the evidence for the result, and the recheck result where there was one.
zenodo-benchmark.csv
One row for each of the 336 benchmark links: Zenodo, figshare, OSF and Dryad links and Zenodo DOIs, with the result, the final address and the HTTP status.
method.md
The method and classification rules, every link that could not be judged with its cause, and the recheck results.
How to cite this study
The dataset, on Zenodo
Boreland, M. (2026). Google Drive link rot in research papers (dataset) (Version 1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23347953
BibTeX
@misc{boreland2026drive,
author = {Boreland, Mark},
title = {Google Drive link rot in research papers (dataset)},
publisher = {Zenodo},
year = {2026},
version = {1.0},
doi = {10.5281/zenodo.23347953},
url = {https://doi.org/10.5281/zenodo.23347953}
}This page
Boreland, M. (2026). 28.6% of Google Drive links in research papers no longer open. NudgeHost. https://www.nudgehost.com/research/google-drive-link-rot