Perishable, Not Permanent
Answer engines treat authority as a flow with a measurable half-life, and that turns content strategy from a problem of accumulation into a problem of maintenance.
In the first week of August, Gartner’s first Market Guide devoted to answer engine visibility tools went into wide circulation. The category now has a name, a list of requirements and a roster of vendors. Every marketing team that has spent the past year improvising will read that document as permission to finally buy something.
The obvious reading is that SEO has simply been renamed. Same discipline, new acronym, new invoices, and a fresh reason to move budget from one line to another.
The number that matters arrived a week later, and it had nothing to do with tools. Profound, one of the vendors named in that guide, published its citation decay data on the 13th. Half the content cited by answer engines is less than 13 weeks old. Authority inside these systems is not something that accumulates. It is something that runs out.
Where the word “evergreen” comes from
The word we use for durable content is borrowed, and it is worth knowing from whom. In newsrooms an evergreen was a piece with no time peg: an article written in advance, kept in a drawer, and dropped into the paper on a thin news day. It read as well in November as in March. Editors kept a small stock of them the way a kitchen keeps preserves. The term came from the botanical description of trees that hold their leaves through winter, and it described a property of the object itself. This thing does not wilt.
Content marketing took the word and quietly promoted it. Evergreen stopped meaning “still readable later” and started meaning “still profitable later”. The pillar page that compounds. The definitive guide that gathers links for a decade. The asset that sits on the balance sheet and works while you sleep. That promotion holds up everything else: almost every content strategy written in the past 15 years rests on the assumption that the right page, made once, keeps paying.
Answer engines do not honour that contract, and the citation decay data shows exactly how they break it. Every cited page now carries a first-citation date, a rise, a peak, a half-life and a last-citation date. That is the vocabulary of radioactive decay, not of real estate. One example in the data describes a page that climbed for 63 days and then halved in 7: a slow ascent followed by a collapse. Some formats halve within two weeks of publication. The object did not change. The index moved on.
Three ways this breaks current practice
The stock illusion. Most organisations still model content authority as capital. You invest in a page, it returns, and the returns accumulate. Under that model the correct strategy is to build fewer things, better, more permanent. Under a decay model the correct strategy looks far more like running a warehouse of perishable goods: you need to know what is fresh, what is moving and what has already gone. The two strategies look similar on an editorial plan and produce opposite behaviour the moment a page starts to fade. The first says wait, it is an investment that matures. The second says act now, the curve is falling.
The calendar refresh. Teams that have accepted the need to update almost always do it on a fixed cadence: the annual audit, the quarterly review, the rule that anything past 18 months gets looked at. That cadence is administratively convenient and analytically arbitrary. It updates pages that were still climbing and ignores pages that aged three weeks after publication. Ordering the update queue by half-life rather than by publication date is a small operational change with a large consequence, because it replaces a proxy with the actual signal.
The owned-media reflex. The instinct, on hearing all this, is to publish more on your own domain. Sometimes that is right. Often it is not. Profound’s analysis of 11.84 billion citations across 29 sectors and 8 models found that roughly 57% of citations globally point back to brand-owned sites, which sounds like a mandate until you look at the spread. ChatGPT sits at 47%; Gemini at 69%. Cybersecurity reaches 74% median share for brand sites; the public and non-profit sector stays under 16%. Pharmaceuticals take most of their citations from earned media, software almost none. There is no universal playbook, only a sector-specific and model-specific answer that almost no organisation has ever calculated.
What the sources agree on
Gartner’s January note on optimising enterprise applications for agentic AI, the March market guide, Profound’s citation corpus, the Genesys data on the state of customer experience, the maturity study from ti&m and the University of Lucerne: different methods, different questions, different continents, and one shape underneath. The systems that now sit between organisations and their customers do not reward the object. They reward the pipeline that keeps that object current.
The hardest impact falls on customer experience, where the same logic applies to material nobody thinks of as content. 95% of consumers expect to be remembered across channels; 48% of organisations do not automatically pass information between virtual agents and human ones. An outdated support article is not merely a poor answer to one customer. It is the source an answer engine cites when somebody asks about your brand, long after your team stopped thinking about it.
The Swiss and Italian markets I work in are particularly exposed here. Adoption is broad and thin: around three quarters of Swiss organisations report productivity gains from AI, revenue gains remain rare, and more than 70% invest less than 5% of their IT budget in the technology. Usage concentrates in marketing, customer service and text production — exactly the functions now sitting on the decay curve. That is an enormous volume of published material, held up by teams sized to produce rather than to maintain.
The uncomfortable arithmetic
If citations have a half-life, then the cost of visibility is recurring rather than capital. Nobody has priced it yet. The content operations budgets I see are built around production: briefs, drafts, reviews, publication. Maintenance appears as a line for occasional audits, when it appears at all. Under a decay model maintenance is not the tail of the process. It is the process, and production is what feeds it.
This is not a comfortable message for teams already stretched, and I would treat the 13-week figure as provisional until somebody independent replicates it. But the direction is not really in question. Gartner’s own forward view in that market guide describes a seismic shift under way towards agentic AI, and agents that execute tasks will be even less sentimental than answer engines about when a page was published.
The next article will look at the second half of the problem: not when to update, but where to publish, given that the citation mix varies so sharply by sector and by model that most organisations are optimising the wrong surface.
Stay close.
— Stefano
Sources
- Joey Adelman and Matthew Huo, “Introducing Citation Decay in Profound”, Profound, 13 August 2026. https://www.tryprofound.com/blog/citation-decay
- Jasman Singh, “Where do AI citations come from?”, Profound, 30 July 2026. https://www.tryprofound.com/blog/where-do-ai-citations-come-from
- Noam Dorros, Isoke Mitchell and Serena Philip, “Market Guide for Answer Engine Visibility Tools”, Gartner, 9 March 2026, circulated 6 August 2026. https://www.gartner.com/en/documents/7559273
- Irina Guseva, “Optimize Enterprise Apps for Agentic AI’s GEO and AEO”, Gartner, 7 January 2026, ID G00841018.
- Francesca Roche, “Genesys Report Reveals CX Loyalty Is Now At Make-or-Break”, CX Today, 14 July 2026, reporting the Genesys 2026 State of Customer Experience. https://www.cxtoday.com/contact-center/genesys-state-of-cx-2026-ai-customer-experience/
- “AI Maturity Study 2026”, ti&m AG with the Lucerne University of Applied Sciences and Arts, Zurich, February 2026. https://www.ti8m.com/en/insights/media/ai-maturity-study-2026