Ahrefs has published research that should reshape how you think about content strategy. Ryan Law, their Director of Content Marketing, analysed 17 million citations across major AI platforms: ChatGPT, Gemini, Copilot, Perplexity, and AI Overviews. The headline finding: AI assistants cite content that is 25.7% fresher than what ranks in traditional organic search.
That’s not a marginal preference. It’s a measurable gap with direct implications for the large majority of New Zealand businesses sitting on content they haven’t meaningfully updated in two or three years.
Most businesses aren’t doing content refresh. Not poorly. Not at all. And whilst they’re ignoring that, they’re spending time on technical fixes that the same research suggests have no meaningful impact on AI visibility. The gap between what marketing teams are doing and what the evidence shows actually works is wider than it should be.
This piece looks at what content AI systems actually reward and what to prioritise. For the broader picture of how AI search works and why it matters, the earlier articles in this series are a good starting point, particularly why your AI search ranking changes every time you check and how AI search builds brands before anyone clicks.
The freshness problem nobody is talking about
The average age of URLs cited by AI assistants is 1,064 days. That’s already relatively fresh. But traditional organic search results average 1,432 days, making AI-cited content nearly 370 days newer. ChatGPT specifically cites URLs that are 458 days newer than Google’s organic results. The strongest freshness preference of any platform tested.
The practical test for whether content needs updating isn’t whether it feels old. It’s whether the claims still hold up. Are the statistics accurate? Has the situation shifted enough that the advice would lead someone in the wrong direction? Are there developments (tools, research, platform changes) that a reader would reasonably expect to see covered?
Feeling old is usually a design problem. Genuinely outdated is when the substance has been overtaken by events.
A piece from 2023 explaining what AI Overviews are and why they matter might read fine on the surface but be missing eighteen months of meaningful development that changes the practical implications entirely. That’s what AI systems are sensing when they deprioritise older content. Not the aesthetic of it, but the accuracy of it.
The mistake most teams make is treating refresh as a cosmetic exercise. Changing the date and moving on. That’s not what the research is rewarding. AI systems are picking up on whether the content actually reflects current reality, not just whether the timestamp says 2026.
For a stretched team, pick five to ten pieces that are genuinely worth saving and do those properly rather than trying to touch everything. The 25.7% freshness advantage isn’t about volume. It’s about recency of substance.
The strategic play most are missing
The words used near your brand across the web shape how AI systems understand what your brand is for. This is the co-mentions principle. Ahrefs uses the example of Patagonia. AI associates that brand with wildcamping and backpacking because those words consistently appear near it across independent sources.
If Patagonia only talked about those contexts on their own website, the association wouldn’t carry the same weight. The signal has to come from sources that have no obvious reason to say it on your behalf.
A business might want to be visible for “sustainable office fit-outs” but if the content surrounding their brand online never uses that language, AI has no basis for making that association. You can’t just claim relevance on your own website. The association has to exist independently, across multiple sources.
This connects directly to entity consistency, the principle that before AI recommends a brand it’s checking whether the signals about that brand are coherent and consistent. Inconsistent signals don’t average out to moderate confidence. They reduce AI’s willingness to recommend. We covered that dynamic in more detail in this earlier piece.
The practical starting point isn’t a PR campaign. It’s an audit. What topics do you actually want to be known for? Now search for your brand name alongside those topics with your own site filtered out. What you’re looking for is whether independent sources, in their own words, connect you to that topic in a meaningful way.
Review sites, industry publications, forum discussions, comparison pieces, podcast transcripts. For most businesses the answer is going to be thinner than expected. If the results are thin, or the language being used doesn’t match how you’d describe your own positioning, that’s the gap. That audit tells you where to point your communications effort. Which publications to target, which communities to show up in, which conversations you need to be part of.
The harder question is whether it’s worth closing.
If a competitor is consistently appearing in those contexts and you’re not, that’s a visibility gap with real commercial implications. AI systems are building an understanding of your category partly by observing who gets mentioned in which contexts. Right now that competitor is shaping the picture. That’s worth addressing.
The New Zealand market advantage is real here. The competitive threshold for establishing a contextual association in a local category is much lower than it would be in a global one. You don’t need to dominate the conversation. You just need to be present in it consistently enough that the association starts to form.
A handful of genuinely useful pieces in the right places (industry publications, relevant local media, sector-specific communities) can establish an association that would take vastly more effort to achieve in the US or UK.
YouTube: where the evidence is pointing
YouTube mentions show the strongest correlation with AI visibility of any factor studied, approximately 0.737. YouTube is the most or second-most cited domain across every major AI platform. One reason: YouTube transcripts are a significant training data source.
But correlation tells you what’s associated with visibility, not necessarily what caused it. Brands that are already well known, well covered, and well established tend to have YouTube presence as part of a broader footprint, and that broader footprint is probably doing a lot of the work.
For a New Zealand marketing manager who’s still working out their content refresh programme and hasn’t audited their entity gaps yet, starting a YouTube channel feels like skipping several steps. Get the fundamentals right first: fresh, accurate content, coherent entity signals across independent sources, consistent brand language in the right contexts.
If you’ve done that work and you’re looking for the next lever, then YouTube becomes a more interesting conversation. Right now for most New Zealand businesses it’s useful to know the data points in that direction, without treating it as an urgent action item ahead of things that will have more immediate impact.
Two things you can safely ignore
When you’re faced with something as opaque as AI visibility (where the rules aren’t published, the signals aren’t fully understood, and the results aren’t cleanly measurable) there’s a natural pull towards anything that feels like a direct lever. llms.txt feels like one. It’s a file you create, you control it completely, and it has the aesthetic of doing something technical and deliberate.
The research on this is unambiguous. Ahrefs examined 300,000 domains and found no correlation between llms.txt and AI citation. The file has no confirmed support from major AI providers and no demonstrated impact on whether or how your content gets used.
Schema markup is the other area where the value proposition has split. Schema still does useful things. It helps Google understand the structure of your content, it can drive rich results, and it has genuine value in specific contexts like local business listings and product pages. But if a team is spending meaningful time on schema implementation expecting it to improve their AI citation rate, the Ahrefs research suggests that effort is probably misallocated.
That said, not all the research points in the same direction. A separate study by AirOps, analysing citation patterns across ChatGPT, found that rich schema was present in 60.5% of ChatGPT-cited pages compared to 24.5% of pages that ranked in Google but weren’t cited by AI. That’s a meaningful gap and it complicates the picture. The honest answer is that the evidence is mixed and the field is moving fast enough that today’s finding may look different in six months. Schema remains worth maintaining for traditional search. Whether it’s a meaningful lever for AI citation specifically is still an open question.
It’s not glamorous work but it’s the kind that compounds. The associations you build this quarter are still there next quarter, and the quarter after that.
Everything else (the technical markup questions, the llms.txt debate, the YouTube question) can wait until those two things are in reasonable shape. Not because they don’t matter, but because the evidence is clearest on these two and the opportunity cost of chasing the others first is real.
The businesses that build these associations now, whilst most of their competitors are still either ignoring AI search or chasing the wrong signals, are going to be in a significantly stronger position in twelve months. The compounding nature of it means early movers get a head start that’s genuinely hard to close later.
The question isn’t whether you can afford to invest in something that takes time. It’s whether you can afford to let competitors establish these associations first.
For plain-English definitions of entity gaps, co-mentions, content freshness, and the other key terms in this piece, the ADMATIC AEO/GEO Glossary is updated after each article in this series.
The next piece in this series goes deeper on citation mechanics, specifically where AI actually looks when forming its understanding of your brand and why your website is the last place it checks.
Free AEO Audit for Marketing Association Members
ADMATIC offers a free AEO audit for Marketing Association members. We’ll run a baseline assessment of how AI search engines currently see your business. That means checking how AI systems describe you when asked directly, whether you’re appearing in AI-generated answers for your key topics, where your entity presence is strong and where it’s inconsistent, and how you compare to competitors in AI recommendations. No jargon, no obligation. Find out more here.
Source: Rod Russell, Managing Partner, ADMATIC, 12th May 2026