We inserted or updated “2026” in the title tag and H1 of 11 pages on our blog, changing nothing else.
Citations to those pages rose 56% and 61% across two test groups. A page we never touched rose 63% over the same period, so we cannot credit the edit for the growth.
Almost all of the lift came from a single URL, which produced 93% of the second group’s net increase. Across all 11 pages, six rose, three fell, and two stayed at zero.
My read: if adding the year does anything, the effect is smaller than the week to week noise in citation data. On this evidence, it is not a lever worth prioritizing, though it costs almost nothing to include in new content.
The bigger lesson: a 61% jump looked like a win until we checked a page we had not edited. Any GEO test without an untouched comparison point will produce results you cannot trust.
Why we tested this
The idea came from a pattern in our own data. Most of our highest cited blog posts already carry “2026” in the title tag. Some of them do not carry it in the H1, which is an inconsistency in its own right, but the title tag pattern was hard to miss.
The reasoning behind it is intuitive enough. AI search engines are widely assumed to favor fresh content. A year in the title is the cheapest possible freshness signal, it takes seconds to add, and a large part of the SEO industry does it as a matter of habit. If it works for AI citations the way it is assumed to work for clicks, it would be one of the highest return edits available.
Assumed is the operative word. We could not find anyone who had tested it against a reference point. So we did.
What we did
We selected 11 pages from the Otterly blog and edited two things on each: the title tag and the H1. Nothing else. No content updates, no new sections, no refreshed statistics, no changes to the last updated date. Where a page’s body still referenced 2025 data, we left it alone, because fixing it would have introduced a second variable.
The treatment varied slightly across pages. On most, we added a year where there was none. On one, we changed an existing “2025” to “2026”. On two, the title tag already had the year and only the H1 was out of sync, so those got a lighter edit.
The edits went out in two waves, on July 9 and July 13, giving us two test groups with staggered measurement windows.
Edited
Baseline (15 days)
Post-edit reading (15 days)
Group A
July 9
June 25 to July 9
July 10 to July 24
Group B
July 13
June 29 to July 13
July 14 to July 28
We also tracked three pages we never edited:
One page that already had “2026” in its title before the study began.
Two pages with no year in the title.
We call these reference pages rather than a control group, since they were not matched to the test pages on topic, format or citation volume. They are still the most useful part of the experiment. Citation counts move on their own, and without something untouched to compare against, there is no way to tell whether an edit did anything or whether the whole category simply had a good couple of weeks.
Group B page by page, because the spread inside the group tells you far more than the average:
Page
Change
The 25 Best AI SEO Tools
+183%
AI Brand Monitoring Tools (2025 changed to 2026)
+165%
Top Tools for Google AI Overview Tracking
+93%
Best AI Search Monitoring Tools
+86%
How often does ChatGPT trigger a Web Search?
-7%
Mastering Brand Monitoring on ChatGPT
-49%
Optimizing Content for AI Search
-75%
AI Search Engines in APAC
no change
Search Volume on AI Search
no change
Four pages up, three down, two flat at zero. We could not find a variable that separated the winners from the losers. Not traffic, not publication date, not content format, not baseline citation volume.
The aggregate was mostly one page
This is the finding we would have missed if we had stopped at the group level.
Group B gained 833 citations in net. One URL, “Best AI Search Monitoring Tools”, accounted for 775 of them. That is 93% of the entire group’s net increase from a single page. Across all 11 test pages in both groups, the net gain was 896 citations, and the same URL accounts for 86% of that.
Remove it, and the remaining eight pages in Group B went from 463 to 521 citations. A 12.5% increase, against a reference page that rose 63% over the same period.
There is a further wrinkle. That page already had “2026” in its title tag before the experiment began. Its edit was a reformatting of an existing year plus an H1 update, which is the lightest treatment in the whole study. The page that carried almost all of the growth received the smallest change.
A volume weighted average lets one large page decide the outcome for an entire group. The 61% figure describes where citations landed. It does not describe how our pages responded to an edit.
Why we are calling it inconclusive
The edited pages rose. That is true, and if we had stopped measuring there, it would have made a good headline.
The problem is the reference page that already had the year in its title and that we never touched. It rose 64% during Group A’s window and 63% during Group B’s. In both cases it matched or beat the pages we actually edited, which came in at 56% and 61%.
The second problem is the page that should have given us our cleanest read. “How often does ChatGPT trigger a Web Search?” had 351 citations at baseline and went from no year to a year, one of the highest volume pages in the study to make that transition. It moved -7%.
The third problem is the design. The reference pages were not matched to the test pages, the treatment varied across them, and the group level result came down to a single URL.
None of this proves that adding the year does nothing. It means this experiment could not separate an effect from the noise around it. Those are different claims, and the difference matters if you are deciding whether to bother with the edit on your own site.
Five reasons it may have come out flat
This is the part we find more useful than the result.
1. The natural variance in citation counts is larger than the effect we were looking for.
Reference pages we never touched moved by 60% and more over a two week span. One of them dropped 75%. Our crawler analytics also swung sharply across pages during the same period, though roughly a third of those observations were missing from the report, so we treat crawler activity as context rather than evidence. If the true effect of a year in the title is a few percentage points, an experiment at this scale has no chance of seeing it. The signal is smaller than the variance in the outcome.
2. The year in the title may be a symptom rather than a cause.
Think about which pages get a year in the title in the first place. Listicles. Rankings. Buyer’s guides. Annual studies. Those are exactly the formats AI engines already like to cite, because they are structured, comparative and easy to extract from. The correlation we noticed in our own data, that top cited pages tend to carry the year, may be a format effect wearing a year effect’s clothing. Adding “2026” to a page does not turn it into a listicle.
3. The freshness signals contradict each other.
Several of our test pages now say 2026 in the title while the visible last updated date says 2024 or 2025, the body still cites 2025 figures, and in one case the URL slug still says 2025. We left all of that intact on purpose, to isolate the variable. It may be the reason the variable does nothing. A model assembling an answer has more than one date signal available, and the title string is probably the weakest of them. Changing the label without changing the content may be exactly the kind of thing these systems are built to see through.
4. A title does not get you into a conversation you were not already in.
Two of our pages had zero citations before the edit and zero after. That is not a failure of the edit, it is a category error in what we expected from it. If a page is not in the consideration set for a given prompt, its title is not what is keeping it out. At best, a year in the title is a tiebreaker between sources a model is already weighing, not an entry ticket. We would leave pages like these out of the sample next time.
5. We could see crawler visits, but not index refresh.
Agent Analytics told us bots were hitting the site during the test window. What it does not tell us is when each engine actually re-read a specific page and started serving the updated title in its answers. Crawl activity and index refresh are not the same event. A 15 day window is an assumption about that lag, not a measurement of it, and some pages may have been evaluated on the old title for much of the post-edit period.
So should you add the year?
Yes, but as hygiene rather than as a lever. Nothing here suggests you should expect a citation lift from it: of the eleven pages we edited, six rose, three fell, and two stayed at zero. It costs nothing, some LLM fan-out queries do include the year when they search the web, and our own most cited pages carry it, though those are also the listicles and guides that tend to get cited anyway. The rule we are settling on: add the year to content that is genuinely current or being refreshed, not as a standalone edit to a page you are not otherwise touching.
What we are testing next
The next experiment in this series looks at semantic triples, restructuring factual sentences into a clean subject, predicate, object format to see whether it makes content easier for models to extract and cite. It is running now, and we will publish the result the same way we published this one, whichever direction it goes.
If you want the null results as well as the wins, follow along.
