Does recency still matter in AI search? How much?
A year ago we found that AI bots loved fresh content. In our original study, in June of 2025, 65% of log hits were for content published within the past year.
A year ago citation data was new, so the 2025 study leveraged log file hits and crawler activity that shows what models are pulling in. Now, July 2026 we can layer in citation data and ask the same question: Does recency matter?
I will be straight with you: I felt like I knew the answer before I started. Recency mattering is not a shocking idea.
But I wanted to layer in two things this time that will help us make smarter moves in deciding when and how content recency impacts your AI search strategy.
- First, comparing a page’s last update to its original publish date, because those are not the same thing and I had a hunch the gap mattered.
- Second, get to the implications instead of stopping at “yeah, you have to update your content.” Because that is not a new finding.
-
-
-
-
- Which content types do you prioritize first?
-
-
-
-
-
-
-
- How do you set your own refresh threshold from citation data (instead of guessing at a cadence)?
-
-
-
-
-
-
-
- How do you think about content you do not own (partnerships, earned placements)?
-
-
-
Methodology
I pulled every page cited in non‑branded LLM answers for four brands across four pretty different categories: a national pet retailer, a vacation rental marketplace, a retail energy provider, and a commercial bank. Three engines (ChatGPT, Gemini, and Perplexity), one shared four‑month window (March through June 2026). I deduped to unique conversations and kept the pages cited three or more times, the ones the models actually lean on.
For each page I pulled its last‑modified date from structured signals like schema, sitemaps, and headers. I could date about two thirds of them, 7,683 pages carrying 47,097 citations. Everything below is computed on that dated set, each page counted once.
The verticals are not the same size. Retail energy is much bigger here than the others, but that is a function of how much data we collected for that client, not how important the category is. So I lean on percentages, not raw counts, when I compare across engines or industries, and I check the headline both pooled and equal‑weighted.
July 2026: 75% of the pages LLMs cite were updated in the last year
88% in the last two. Plot it out and the whole thing collapses toward the present, with anything older than three years basically non-existent. And the updates are recent, not just “sometime in the last year”.
Of the pages updated in the last year: more than half were last touched in the previous three months. To be clear, this is about how recently a page was updated, not whether the page itself is new. A refreshed old page counts.
The old page you refreshed beats the new one you wrote
Here is the part I added this year. Take the pages where I could capture both dates and measure them two ways. By last update, 72% look fresh. But if that page was published in the last year, , that citation rate drop to 42%.
The freshness LLMs reward is being manufactured by updates, not by new publishing. More than a quarter of the “fresh” pages were first published over two years ago. They earned their freshness by being maintained.
It is strongest in the pages models rely on most. Among the highest‑volume cited pages, 75% were updated in the last year but only 40% were first published that recently. This reinforces that content is not a set it and forget it strategy. The content you published 2, 3, hell, 10 years ago can still work for you if maintained.
Each LLM cites what it prefers, and that determines its freshness
All three engines cite fresh content, but not equally. The gap is influenced by the type of content each chooses to cite, not by some default preference for recency.
Defined: “fresh‑from‑old”: Pages that are updated in the last year (fresh),first published two or more years ago. In plain terms, it is how much of an engine’s fresh content is refreshed rather than genuinely new.
Gemini: freshest content, 78% of content cited updated in the last year
And 90% of Gemini citations on content updated within two years. The reason is its content mix: marketplaces and aggregators (26% of its citations) plus comparison and reviews (24%) make up half of what it cites, and those are the two freshest page types in the study. Blogs and guides are only 37%. Its citations skew heavily to energy (62%), which is also where the marketplace content lives. Of its fresh pages, 27% were refreshed rather than genuinely new.
If you are optimizing for Gemini, comparison and marketplace presence is where freshness pays off fastest.
ChatGPT: the blogs, guides, and brand-pages engine
ChatGPT sits in the middle at 73% updated in the last year, 87% within two, and it leans hardest on editorial: blogs and guides are 46% of its citations, nearly half.
ChatGPT cites brand and corporate pages more than the others (10%), and its industry mix is the most balanced (energy 45%, pet 20%, banking 18%, travel 16%).
If your own pages are going to earn citations anywhere, ChatGPT is the most likely place, so this is where owned-content freshness matters most.
Perplexity: forgiving, only 65% updated in the last year
And 83% of Perplexity citations referencing content updated within the last two years. It leans on blogs and guides (44%) and comparison (22%), but it reaches further into the long tail and pulls older reference material the others skip.
Perplexity is the one engine where strong, older evergreen content still earns citations, rewarding depth and authority, not just recency.
Industry doesn't really change the “freshness rule”, but the type of citations by industry does
The recency rule is held in all four categories, but the shape of each landscape is different, and each engine behaves differently inside each one.
In Gemini content recency’s impact varies the most, from 82% in energy down to 63% in banking, because freshness rides on how much comparison and marketplace content a category has.
ChatGPT is the most stable across categories. Perplexity is consistently the oldest, whatever the vertical.
The other big variable is concentration: in some categories a small set of sites captures most of the citations, in others the citations are spread thin. I am measuring this as a share, the percentage of citations held by the top 10% of sites
Retail energy: 80% of content cited is updated in the last year
The freshest and most ownable industry in the study. It is also the most concentrated: the top 10% of sites capture 78% of all citations. The whole industry category is carried by far fewer sites (around 300, versus 600 to 865 elsewhere). That makes it a small, mappable shelf. Old, established content is the most refreshed (37% fresh‑from‑old), and its owned pages run fresh (82%). About 37% of its cited pages are “always on,” meaning cited in all four months.
Commercial banking: opportunity to win citations with structured guides
75% of commercial banking citations were content updated in the last year, and the most consistently ‘dated’ category because financial sites publish clean structured dates. Citations spread across about 700 sites at roughly 10 each.
Guides and explainers dominate fresh banking content LLM citations, and half of its cited pages are always‑on, the highest share of any vertical.
Travel:new content outperforms refreshed content in LLM citations
72% of cited content was updated in the last year. Citations are overwhelmingly editorial, destination and planning guides are nearly two thirds of what gets cited, and it has the lowest fresh‑from‑old rate (18%), so its fresh content is more often new.
This is also where I saw the clearest warning sign:
One travel brand we looked at had its own pages running older than the third-party pages written about it. That is a refresh flag the brand controls directly and probably did not know it had.
In pet retail, breadth of content wins, recency matters the least
69% of pet retail content was updated in the last year, the lowest of the four and the most fragmented. Citations scatter across more than 860 sites with no small set carrying the category. Blogs and guides are more than half of cited pages, and that evergreen care content is the slowest to be refreshed, which is why this vertical lags.
Freshness fades the longer a page stays cited
Recency as a long‑term strategy is not about a single citation spike. You want pages that get referenced month after month. So that “always-on” referenced above, is looking at how many of the four months each page showed up in and how fresh those pages were.
The more consistently a page is cited, the older it tends to be. Pages cited in all four months (the “always-on” set, 45% of dated pages) are 68% fresh with a median age around six months. One‑month spikes are the newest of all (86% fresh, median about two months). Freshness keeps climbing as consistency drops.
So here is how I read it. A very fresh update earns the first pickup, that is your spike. But the pages that stay cited every month are established pages that are kept reasonably current, not the newest things on the internet. They are still maintained (median under six months since an update), just not bleeding edge. They are also more editorial (52% blogs and guides) and far more likely to be cited across all three engines (42% of them). The always‑on share runs 37% in energy, around 50% in banking and pet, and 48% in travel.
Newest gets you the spike, established plus maintained gets you the staying power. If you want durable AI visibility instead of a flash, you are playing the second game, and that is a maintenance game, not a publishing one.
Which content types have to be freshest, and which ones you actually control
Not every content type is held to the same bar. The types that win citations most reliably are also cited freshest, so they have to stay current to earn the spot. News and editorial is the clear outlier, the oldest content cited, because models reach for evergreen reference far more than the news cycle.
But freshness is only half the picture. The other half is whether a given type is even yours to refresh. So I have tagged each one as an owned lever (content you publish and can update yourself) or earned / third-party (content on sites you can influence but not directly control).
Put the two together and the strategy gets clearer. The types held to the highest freshness bar, marketplaces and comparison and reviews, are mostly earned, so you cannot just refresh your way onto them. The levers you fully own are your blogs and guides and your brand pages, and those are exactly where a refresh program pays off directly. Blogs and guides sit in both worlds: you own yours, but plenty of the cited ones belong to other people.
Keeping your own house current works, but it is nowhere near enough on its own. The guides, comparisons, and marketplaces that carry the volume are where the visibility actually lives, and most of that is earned, not owned.
“Keep your content fresh” is the most obvious advice in the world.
So obvious it is almost useless. The hard part was never knowing that updates help. The hard part is figuring out what to update, and when, without it turning into a manual slog nobody on your team has time for. So that is where I want to land this. Not “update your content,” but how to build a system that tells you what is worth updating.
Build the trigger from your own data
The instinct is to slap a refresh schedule on everything. Do not. Use the data you already have. Use an AI citation tracking tool (We use Scrunch) to see what's actually getting pulled into AI answers, then look specifically at the URLs that keep showing up month after month, not just the ones that spike once. Check how recently those always-on pages were last updated, and how often they get touched. The idea is that you’re not aiming for a one-off citation but consistency in citations as well. You are letting the citation data set the threshold instead of guessing at a cadence. That is the difference between a system and a chore.
Let content type set the pace
Not everything needs the same attention, and the data makes that clear. The page types that win citations (comparison and reviews, marketplaces, your own brand pages) are cited freshest, so keep those close to current. Blogs and guides carry the most volume but tolerate a little more age. Some evergreen genuinely does not need touching on a schedule. Content type is a useful lever: it tells you roughly how often a page is worth revisiting, so you stop refreshing things that do not need it and stop ignoring things that do. Sort your library by content type, blogs, product pages, comparison pages, marketplace listings, and by function. Let your citation data show which types are earning their freshness and which aren't. That pattern varies by industry, so check your own.
For evergreen content, I’d ask myself three questions before leaving it alone: Have the facts held up? Is it still winning citations at its old rate? Would being outdated actually cost you? Clean on all three, leave it alone.
Remember it is not just your own content
Your own pages are the freshest you have but you may only own about 2% of what gets cited. The other 98% is third-party, so recency is an earned-media question too. If you are investing in partnerships, add one more thing to how you vet a publisher: how often do they refresh their content? A site that lets its articles rot is a worse bet for AI visibility than one that keeps them current, even at similar authority. You are not just buying a placement, you are buying a page that either stays alive in the models or quietly ages out.
Where I land
Recency still matters, and it is the last update that counts, not the publish date. None of that is new. What is new is having the citation data to confirm it, the breakdowns to see how it plays out by engine and category, and a clearer view of what to do about it.
And what to do about it is not “publish more.” The brands that win in AI search are not the ones writing the most. They are the ones who build a system to keep their best content alive, who let their own data tell them what is worth updating and when, who understand that staying cited every month is a different game than catching a spike, and who remember that recency runs through the third-party pages they do not own too. Publish and forget loses. Publish and maintain wins. The real work is making “maintain” something your team can actually sustain.
Looking to improve your GEO strategy with a data driven approach? Learn more about our offering here or contact us.
