Eight percent of the addresses AI engines cite carry a year. In the Machine Relations Index answer ledger, 15,451 of 197,384 cited-URL events (7.8 percent) had a number from 2010 to 2029 in the URL path. Almost all of those years are the current one: 2026 is in 10,754 events (5.5 percent), 2025 in 2,169 (1.1 percent), 2024 in 666 (0.3 percent), and every year from 2010 to 2023 together in 1,824 (0.9 percent). Four runs in ten (9,310 of 23,183) cited at least one year-stamped address, and three in ten (6,984) cited at least one stamped 2026.
The year is not spread evenly. A 2026 stamp is in 7.8 percent of the addresses cited on "best X" questions and 2.5 percent of those cited on problem-first questions. It is in 9.7 percent of those cited on AI infrastructure questions and 1.9 percent of those on consumer products. And a year written into a slug behaves differently from a year written into a date folder: slugs stamped 2026 outnumber slugs stamped 2025 by 6.9 to 1, while date folders show 1.7 to 1.
This page counts cited addresses. It does not say an engine prefers this year's page: the ledger has no count of how many 2025 and 2026 pages exist for the same question, so a 5-to-1 ratio of citations is not a 5-to-1 preference.
What was counted #
- Population. Successful runs in the Index's "mri_v2" panel, from 2026-07-06 to 2026-11-07, on its 1,008 buying questions: 168 in each of six question shapes (best X, top lists, how to choose, is it worth it, problem first, X vs Y). Runs flagged as naming a brand in the prompt are excluded. The ledger export is dated 2026-11-07 08:10 UTC.
- Engines. ChatGPT, Claude, Gemini, Google AI Mode and Perplexity. Google AI Overviews are left out because this analysis was not run on them.
- Unit. One cited URL in one answer run, counted once however many times the run repeated it. The same URL cited in two runs counts twice. The set holds 197,384 such events from 23,183 runs, 78,194 distinct URLs and 29,282 root domains.
- A year stamp. A four-digit number from 2010 to 2029 that stands on its own in the URL path, meaning it is not part of a longer number. Query strings and fragments are ignored. No page was fetched or read, so a stamp is a fact about the address and not about the page.
- The questions carry no year. None of the 1,008 question texts contains a four-digit year, so the wording of the question does not push engines toward a stamped address.
Almost every cited year is the current one #
Cited-URL events by the year found in the path (share of all 197,384, 95 percent interval from resampling the 1,008 questions):
| Year in the path | Events | Share | 95% interval |
|---|---|---|---|
| 2026 | 10,754 | 5.45% | 5.06 to 5.81 |
| 2025 | 2,169 | 1.10% | 0.99 to 1.20 |
| 2024 | 666 | 0.34% | not computed |
| 2010 to 2023 | 1,824 | 0.92% | not computed |
| 2027 to 2029 | 38 | 0.02% | not computed |
| Any year | 15,451 | 7.83% | 7.43 to 8.25 |
| No year | 181,933 | 92.17% |
The 2026-to-2025 ratio is 4.96 to 1 (interval 4.45 to 5.56). Counted by distinct URL, 4,048 of the 78,194 are stamped 2026 (5.2 percent) and 981 are stamped 2025 (1.3 percent), so repeats do not produce the pattern. The stamped citations are not one publisher's habit. The 2026 events come from 2,356 root domains and the largest, track360.io, holds 252 of 10,754 (2.3 percent). The 2025 events come from 713 domains and the largest, futureagi.com, holds 34 of 2,169 (1.6 percent).
Last year's stamp is not gone. A 2025-stamped address was cited in 1,857 runs (8.0 percent of all runs) and 713 domains still had one cited. Next year's stamp is not here yet: 2027 appears in 33 events.
Year stamps cluster on best-X questions #
Share of cited URLs stamped 2026 or 2025, by the shape of the question that produced the citation (events matching, out of events):
| Question shape | Cited URLs | Stamped 2026 | Stamped 2025 | 2026 to 2025 ratio (95% interval) |
|---|---|---|---|---|
| Best X | 38,847 | 7.8% (3,047) | 1.0% (403) | 7.6 (5.8 to 10.0) |
| Is it worth it | 30,652 | 6.3% (1,930) | 1.5% (474) | 4.1 (3.2 to 5.2) |
| Top lists | 39,901 | 6.2% (2,490) | 1.2% (469) | 5.3 (4.3 to 6.8) |
| X vs Y | 30,246 | 4.5% (1,373) | 0.9% (257) | 5.3 (4.1 to 7.1) |
| How to choose | 31,480 | 4.0% (1,251) | 0.7% (234) | 5.3 (4.2 to 7.0) |
| Problem first | 26,258 | 2.5% (663) | 1.3% (332) | 2.0 (1.4 to 2.8) |
The 2026 share on best-X questions (95 percent interval 6.9 to 8.8) is more than three times the share on problem-first questions (2.0 to 3.2). On problem-first questions, where the buyer describes a situation and names no product, the 2026 share is the lowest of any shape (2.5 percent) and the 2025 share (1.3 percent) is more than half of it, so a stamp there is far less tied to the current year. Rankings and "is it worth it" pages are the ones that wear a year.
A year in the slug is a different thing from a year in a date folder #
Many publishers put the publication date in the path (/2025/07/30/...). That year says when a page was written. A year in the slug (best-iot-microcontrollers-for-2025) says what year the page is about. Splitting the stamped events by whether the year is its own folder in the path:
| Where the year sits | Stamped 2026 | Stamped 2025 | Stamped 2024 | 2026 to 2025 |
|---|---|---|---|---|
| In the slug or file name (not its own folder) | 9,376 | 1,356 | 320 | 6.9 to 1 |
In a date folder (/2025/07/...) |
1,378 | 813 | 346 | 1.7 to 1 |
Addresses that name a year in the slug are cited far less once the year has passed: a slug stamped 2025 is cited one-seventh as often as a slug stamped 2026. Addresses that only record a publication date lose far less: a 2025 date folder is cited at about 60 percent the rate of a 2026 one. The ledger does not say why, and does not say whether the engines read either year.
Claude wears the year most, and every engine follows the pattern #
Share of cited URLs stamped 2026 or 2025, by engine (events matching, out of events):
| Engine | Cited URLs | Stamped 2026 | Stamped 2025 | 2026 to 2025 |
|---|---|---|---|---|
| Claude | 33,192 | 9.5% (3,165) | 1.3% (425) | 7.4 |
| Gemini | 17,416 | 4.8% (829) | 1.3% (235) | 3.5 |
| Perplexity | 105,295 | 4.8% (5,005) | 1.1% (1,201) | 4.2 |
| Google AI Mode | 6,124 | 4.4% (268) | 1.1% (68) | 3.9 |
| ChatGPT | 35,357 | 4.2% (1,487) | 0.7% (240) | 6.2 |
Claude cites 2026-stamped addresses at about twice the rate of the other four engines. Gemini and Google AI Mode are the engines that cite last year's stamp closest to the 2026 rate, at 3.5 and 3.9 to 1. Perplexity supplies 105,295 of the 197,384 events, so the pooled figures are weighted toward it.
The vertical matters more than the engine #
Share of cited URLs stamped 2026, highest and lowest of the 24 verticals (events matching, out of events):
| Vertical | Cited URLs | Stamped 2026 |
|---|---|---|
| AI infrastructure | 9,978 | 9.7% (964) |
| iGaming and betting | 8,412 | 8.7% (732) |
| AI visibility and GEO | 12,965 | 8.4% (1,090) |
| Healthcare services | 8,379 | 7.9% (666) |
| Local services | 6,438 | 3.0% (193) |
| Consumer health | 12,832 | 2.5% (325) |
| Review-sparse industrial | 6,333 | 2.0% (128) |
| Consumer products | 12,423 | 1.9% (234) |
The spread between the top and bottom vertical is five to one. A page's year stamp is common where the category changes every few months and rare where it does not.
The share of 2025 stamps is falling on Perplexity #
Share of cited URLs stamped 2026 and 2025 by month, Perplexity against the other four engines combined (events matching, out of events; November is 1 to 7 only):
| Month | Perplexity events | Perplexity 2026 | Perplexity 2025 | Other engines' events | Others 2026 | Others 2025 |
|---|---|---|---|---|---|---|
| July (from the 6th) | 11,967 | 3.6% | 1.4% | 22,120 | 5.5% | 1.2% |
| August | 15,135 | 3.6% | 1.5% | 8,873 | 6.0% | 1.0% |
| September | 23,636 | 4.8% | 1.5% | 12,955 | 6.8% | 1.1% |
| October | 47,428 | 5.2% | 0.9% | 41,066 | 6.5% | 1.0% |
| November (1 to 7) | 7,129 | 5.6% | 0.5% | 7,075 | 6.3% | 0.8% |
On Perplexity, the 2025-stamped share drops from about 1.5 percent in July to September to 0.9 percent in October and 0.5 percent in the first week of November, while its 2026 share rises from 3.6 to 5.6. On the other four engines the 2026 share moves between 5.5 and 6.8 percent and the 2025 share drifts from 1.2 to 0.8. Perplexity's October volume doubled, so a change in which questions or runs were sampled could explain part of the move; this page did not test that.
Can a stamped page be rolled forward? #
The ledger cannot yet say. Only four pairs of addresses on the same domain, identical apart from a 2025 and a 2026 stamp, were cited in this window (seven citations of the 2025 version and seventeen of the 2026 version). Nearly all stamped pages cited here are new addresses each year, not one address with a changed number: 961 of the 965 address stems stamped 2025 have no 2026 twin in the ledger, and 3,837 of the 3,841 stems stamped 2026 have no 2025 twin. Whether a 2026 page keeps its citations in January, whether renaming it to 2027 helps, and whether a redirect holds them are not answerable from a window that ends in November 2026.
What this changes for how pages are named #
- Put the year in the slug of a ranking, and only of a ranking. Best-X, top-list and is-it-worth-it questions are where stamped addresses are cited most. Problem-first and how-to-choose pages carry a stamp in 2.5 and 4.0 percent of citations, and an explainer gains little from a year it will have to drop.
- A year in the slug is cited much less once the year has passed. A slug stamped 2025 is cited one-seventh as often as one stamped 2026. A 2025 date folder is cited at about 60 percent of the 2026 rate. This counts citations; it does not show that a page lost authority.
- Do not delete last year's page. A 2025-stamped address was cited in 8.0 percent of runs this autumn. Keep it, link it forward to the new edition, and judge any redirect on a later window.
- Do not pre-stamp 2027. Thirty-three of 197,384 cited events carry it.
What this page does not show #
- Cause. A stamp may label content that engines select for other reasons, such as recency in the page body. No page was read.
- Preference. The ratio compares citations. It does not control for how many 2025 and 2026 pages exist on the web.
- Buyer demand. The 1,008 questions are the Index's own panel, written for coverage of buying decisions. They are not a sample of what buyers type.
- Citation quality. A cited URL is a source an engine linked or listed. Whether the answer relied on it or named it is a separate question: the Index finds engines name only 17.6 percent of the sources they cite (AI engines cite sources they do not name).
- Google AI Overviews. Not included.
- A January view. The window ends 2026-11-07. It cannot show what happens to 2026 stamps in 2027.
- A count of mistaken stamps. A number such as 2024 in a product code or an ID is counted as a year. This inflates the older-year rows slightly and is the reason the 2010 to 2023 total is shown only as a group.
How the address is read for other signals is in the URL slug study.
Method note #
Source: the Index answer ledger's citation-events table, daily export of 2026-11-07 08:10 UTC, read with DuckDB and joined to the question table for question shape. Filters: panel "mri_v2", runs not flagged as brand-prompted, a non-empty cited URL, engines ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, run dates 2026-07-06 to 2026-11-07. Unit: distinct (answer run, cited URL) pairs. The path is the URL lowercased, with scheme, host, query string and fragment removed. The year is the first number from 2010 to 2029 in the path with no digit directly before or after it; 46 URLs carry both 2025 and 2026 and count once, under the first. A second expression that counts any occurrence returns 10,798 for 2026, 2,181 for 2025 and the same 15,451 for any year. A date folder is a path with a four-digit year as a whole folder (/2025/); every other stamped URL, including a year in a file name, counts as a slug stamp. Intervals resample the 1,008 questions, 2,000 draws, fixed seed, with every cited URL of a question kept together. Vertical, engine and monthly rows report pooled counts and no interval. An address stem is the path with a 2025 or 2026 stamp replaced by a placeholder, matched within one root domain.
Frequently asked questions #
Do AI engines cite URLs with the year in them? #
Some. In the Machine Relations Index, 7.8 percent of cited URLs (15,451 of 197,384) had a year in the path, and 5.5 percent had 2026. This describes the cited addresses and does not show that a year in the URL earns a citation.
Are last year's URLs still cited? #
Yes, less often. Addresses stamped 2025 are 1.1 percent of cited URLs, against 5.5 percent for 2026, and appeared in 8.0 percent of runs. Addresses with a 2025 date folder are cited at about 60 percent of the rate of those with a 2026 folder.
Should I put the year in the slug of a best-of page? #
The evidence says it is common there. A 2026 stamp is in 7.8 percent of cited URLs on best-X questions and 6.2 percent on top lists, against 2.5 percent on problem-first questions. It does not show that adding a stamp earns a citation.
What happens to a year-stamped page next year? #
This ledger cannot say, because its window ends in November 2026. A slug stamped 2025 was cited about one-seventh as often as one stamped 2026 in this window, so expect fewer citations, and plan the new edition at a new address while keeping the old one live.