Methodology
We ran 1,000 real searches across nine industries. Most of them are the kind of search you'd make when you're picking a product or a provider.
For each search we recorded what Google's AI Overview cited. Then we asked the same question to Perplexity, Claude, Gemini and ChatGPT, and recorded what each of them cited.
We also captured the searches the AIs ran themselves before answering. That turned out to matter a lot.
Then we compared the cited pages against the pages that ranked in Google's top 30 for the same search but didn't get cited. That's 9,795 different pages, and 12,985 page and search pairs.
Why Google's top 30
We only looked at pages in Google's top 30, because that's where AI Overviews do most of their citing. If you leave out YouTube, Reddit and Google's own pages, 77% of what AI Overviews cited ranked in the top 30.
And the odds drop pretty fast even inside the top 30. 61% of pages in positions 1 to 3 got cited, but only 9% of pages in positions 11 to 30.
How to read the numbers
- "% more citations" is the lift. So 23 citations per 100 pages instead of 16 is 48% more. We always show the "out of 100" numbers too, because a lift gets big fast when the starting number is small.
- Every comparison is within the same search. Some searches are just easier to get cited in than others, so we only ever compare pages competing for the same one.
- "Held in 40 of 40 re-tests" means it isn't a fluke. We split the searches in half at random, ran the check on each half, and did that 40 times. If a pattern only shows up in a handful of those, you shouldn't act on it.
- ChatGPT's numbers are going to look tiny. It cited about 2 pages per 100 here, so small differences in its column are just noise. We explain why in Why ChatGPT is different.
Finding 1: A year in the title
Within the same search, across pages ranking in Google's top 30:
| AI engine | Cited, year in title | Cited, no year | Lift | Re-tests held |
|---|---|---|---|---|
| Claude | 25.9 in 100 | 8.9 in 100 | +192% | 40 of 40 |
| Gemini | 15.5 in 100 | 7.2 in 100 | +115% | 40 of 40 |
| Perplexity | 23.1 in 100 | 15.6 in 100 | +48% | 40 of 40 |
| Google AI Overviews | 21.0 in 100 | 17.5 in 100 | +20% | 0 of 40 |
| ChatGPT | 1.5 in 100 | 2.1 in 100 | -29% | 0 of 40 |
On Claude, Gemini and Perplexity, this held in every single re-test. That's about as solid as a study like this gets.
Google gave a small lift of 20%, but it didn't hold up in any of the 40 re-tests. So don't bank on it there.
ChatGPT's -29% is based on about 2 citations per 100, which is too few to read anything into.
The year does more than mark a page as fresh. We checked how recently each page was updated too, and the year still kept almost all of its effect. It's 3.25 times the odds on Claude and 1.60 on Perplexity, but only 1.25 on Google.
Where a year belongs
Cited per 100, year in title against no year, within the same search:
| Search type | Perplexity | Claude | Gemini | ChatGPT | |
|---|---|---|---|---|---|
| "Best of" | 24 vs 12 | 25 vs 6 | 13 vs 3 | 22 vs 12 | 2 vs 2 |
| Product or service category | 25 vs 15 | 31 vs 7 | 15 vs 5 | 20 vs 20 | 0 vs 2 |
| Cost and pricing | 23 vs 12 | 24 vs 8 | 19 vs 8 | 23 vs 17 | 2 vs 2 |
| Reviews | 25 vs 19 | 28 vs 13 | 15 vs 9 | 23 vs 18 | 3 vs 4 |
| Comparisons ("X vs Y") | 21 vs 19 | 23 vs 15 | 23 vs 17 | 19 vs 23 | 0 vs 0 |
"Best of", category and pricing pages gained the most. Comparison pages gained the least. That makes sense, because an "X vs Y" answer doesn't go out of date as quickly as a "best of" list does.
Finding 2: The AI adds the year to its own search
This was the result that surprised us most, and it explains why the year works.
Before answering, the AIs run their own web searches, and we logged every one of them. Claude added a year to its search on 77% of searches, Gemini on 87%, and ChatGPT on 60%.
| Claude | Gemini | ChatGPT | |
|---|---|---|---|
| Searches where the AI added a year | 77% | 87% | 60% |
| Year in title, when it searched with a year | 26.7 vs 6.7 per 100 (+300%) | 16.0 vs 6.4 per 100 (+149%) | 1.8 vs 2.1 (-18%) |
| Year in title, when it didn't | 22.3 vs 18.7 (+19%) | 11.9 vs 22.0 (reversed) | 1.0 vs 2.1 (-53%) |
When Claude or Gemini put a year in its own search, pages with a year in the title got almost all of the citations. When they didn't, the advantage went away.
So what your title actually matches is the search the AI typed, and most of the time that search had a year in it.
ChatGPT points the same way, because pages with a year do a little less badly when it searched with one. But both of those numbers are based on about 2 citations per 100, so you can't conclude anything from them.
One thing to keep in mind is that the "didn't search with a year" row is based on only 107 searches for Claude and 36 for Gemini. So the direction is pretty clear, but don't lean too hard on the exact numbers in that row.
Finding 3: Use the current year
This is probably the most useful thing in the whole study. Here's how titles with the current year did against titles with an older year:
| AI engine | Cited, 2026 | Cited, older year | Lift |
|---|---|---|---|
| Claude | 30.3 in 100 | 8.6 in 100 | +252% |
| Google AI Overviews | 23.0 in 100 | 10.8 in 100 | +113% |
| Perplexity | 25.8 in 100 | 15.6 in 100 | +66% |
| Gemini | 15.7 in 100 | 9.7 in 100 | +62% |
| ChatGPT | 2.4 in 100 | 1.5 in 100 | +56% |
The Google number surprised us. Having a year in the title barely does anything on Google, but titles with the current year still got 113% more citations than titles with an older one.
So an old year in your title actually hurts you, and that includes Google.
Every January, go and update the year in the titles that have one, and refresh the page while you're in there.
Finding 4: Google rewards pages that are actually up to date
Among pages that publish a date, within the same search:
| AI engine | Updated 2025 or later | Older | Lift |
|---|---|---|---|
| Google AI Overviews | 22.3 in 100 | 10.3 in 100 | +117% |
| Claude | 19.2 in 100 | 7.5 in 100 | +158% |
| Gemini | 13.6 in 100 | 6.9 in 100 | +98% |
| Perplexity | 19.7 in 100 | 13.2 in 100 | +49% |
| ChatGPT | 1.6 in 100 | 1.1 in 100 | +47% |
On Google this held in all 40 re-tests. And updated pages did better on every other engine too.
The date that matters on Google is the one it shows next to your result. Pages showing a date from the last year got 186% more AI Overview citations than pages showing a date over two years old.
A visible "Updated" date on the page got Google to show the fresh date 87% of the time. When the dates were only in the page's code, that dropped to 40%.
So you want to do all three. Actually update the page, show a visible "Updated" date on it, and put the current year in the title.
Just changing the year on a page you haven't touched won't do much for you on Google.
Finding 5: A title that contains the search
Here's what happened when the title had every word of the search in it, within the same search:
| AI engine | Cited, title has every search word | Cited, it doesn't | Lift |
|---|---|---|---|
| Perplexity | 22.8 in 100 | 15.9 in 100 | +43% |
| Claude | 19.2 in 100 | 11.0 in 100 | +74% |
| Google AI Overviews | 23.1 in 100 | 17.6 in 100 | +31% |
| Gemini | 12.1 in 100 | 9.8 in 100 | +23% |
| ChatGPT | 2.0 in 100 | 1.4 in 100 | +40% |
On Perplexity this held in all 40 re-tests. It also held after we took into account whether the page had a question heading, and how well its content matched the search.
So on Perplexity, the title is doing something on its own, on top of what the rest of the page does.
When the exact search phrase was in the title, the lift looked even bigger. That's +116% on Claude, +62% on Perplexity, +59% on Gemini, +31% on Google and +22% on ChatGPT. We only found that one while digging through the data, so treat it as something worth testing rather than a result.
Finding 6: What made no difference
These are the title rules most people are still following. None of them made a consistent difference.
| Title tactic | Perplexity | Claude | Gemini | ChatGPT | |
|---|---|---|---|---|---|
| Short titles, under 40 characters | +0% | -5% | -23% | -2% | +90% |
| Long titles, over 65 characters | +1% | -11% | +6% | -4% | -47% |
| "| Brand Name" suffix | +15% | -5% | -12% | -10% | +50% |
| Title written as a question | -4% | +1% | -13% | -7% | -46% |
| A list number, such as "7 best" | -12% | -8% | +17% | +24% | -41% |
Lift in citations, within the same search. On Google and Perplexity, none of these held up in more than 7 of 40 re-tests, and most held up in none. The ChatGPT column jumps around a lot because it's based on about 2 citations per 100.
The drop on Claude for short titles looks like it might be something, but it didn't hold up when we re-tested it. Neither did the list number results, which go in opposite directions on Google and Gemini.
If you want to use a question, put it in a heading. A heading on the page that asks the searcher's question got 26% more Google citations, and that held in all 40 re-tests. So your title doesn't need to be a question for you to get that lift.
Why ChatGPT is different
ChatGPT is in every table above, and its numbers are tiny in all of them. That's because it barely cites the pages we compared.
Only 8% of the pages ChatGPT cited ranked in Google's top 30. For the other engines it was 21% to 40%.
The reason is in its own searches. ChatGPT's searches averaged 9.1 words, and Claude's averaged 5.5. ChatGPT also added 6.4 words on top of what the person typed, where Claude added 2.2, so it goes long-tail and specific fast.
This one surprised us. 46% of ChatGPT's searches used a site: operator, which means it went straight to one website instead of searching the whole web. A search like that isn't going to land on Google's top 30 for the original question.
So a title tag study built on Google's top 30 can't tell you much about ChatGPT, and that's fine.
We do know what gets ChatGPT citing you, though. Our own site went from 72.5 AI citations a day to around 700, tracked in Bing's AI reporting, which covers ChatGPT and Copilot.
The page checks behind that are in The 3 Numbers That Decide If AI Cites Your Page, so that's the one to read next.
What to do
- Put the current year in the title of your "best of", category, pricing and review pages. That's 192% more citations on Claude, 115% on Gemini, 48% on Perplexity and about 20% on Google. ChatGPT is the exception, see point 6.
- Update that year every January, and refresh the page when you do. Titles with the current year got 113% more citations than titles with an older year, even on Google.
- Actually update the page, and show a visible "Updated" date. Recently updated pages got 117% more Google citations, and a visible "Updated" date is what gets Google to show the fresh date.
- Put the words of the search you're targeting in the title. That's 43% more citations on Perplexity, and it points the same way on the other engines.
- Stop spending time on title length, brand suffixes and question titles. They didn't do anything here. They might still matter for clicks in normal search results, but this study didn't measure that.
- For ChatGPT, work on the page rather than the title. Its own searches are long and specific, so run the checks in The 3 Numbers That Decide If AI Cites Your Page.
- Focus on buyer searches. Claude searched the web on 80% of buyer searches and only 9% of informational ones. ChatGPT searched on 86% against 27%. If the AI doesn't search, none of this reaches it.
Then test it on your own pages. Take a group of similar pages, add the current year to half of them, and leave the rest alone. Then compare citations a month later, and you'll know if it works on your site.
Limitations
This shows patterns, so it can't prove cause and effect. A year in the title might go along with other things, like pages that just get refreshed every year anyway. The freshness check and the AIs' own searches rule a lot of that out, but testing it on your own pages is what settles it.
AI answers change between runs. When we asked the same engines the same questions a few hours apart, only 49% to 65% of the cited pages came up again. So there's always going to be some wobble in numbers like these.
The freshness numbers only include pages that publish a date. Pages with no date a machine can read aren't part of Finding 4.
We only tested US, English, desktop results. Things might look different elsewhere.
Scope: 1,000 searches across nine industries, 9,795 distinct pages, Google AI Overviews, Perplexity, Claude, Gemini and ChatGPT (US, desktop, English), September 2026. Every pattern re-tested on 40 random half-splits of the searches.