The Citation Half-Life: Why Your AI Mentions Decay

Here is the direct answer: an AI citation is not a trophy you win once — it is a position you rent, and the rent comes due constantly. Every time someone asks ChatGPT, Perplexity, or Google's AI Mode the question you currently answer, the engine re-runs its retrieval and re-decides who to name. Your slot from last month is not grandfathered in. I call the rate at which your mentions fade the citation half-life: the time it takes for half the AI answers that once named you to stop naming you, with no new work on your part. For most businesses in most categories, that half-life is far shorter than they assume — weeks to a few months, not years. If you treat AEO as a project you finish, your visibility is already decaying while you celebrate.

This is my read after two years of watching client mentions rise and fall at AIrecommend.ai, and after the pattern showed up again and again in the data behind our State of AI Search 2026 research. It is not a law of physics. But it is the most useful mental model I have for why "we got recommended by ChatGPT once" almost never turns into "we get recommended by ChatGPT reliably."

Why a citation decays at all

The intuition most people carry over from SEO is wrong here. In classic search, a page that earned a top ranking tended to hold that ranking for a while. Rankings had inertia. There was an index, your page sat in it, and unless something changed, your position was roughly stable day to day. You could rank once and coast for a quarter.

Answer engines do not work that way, and the reason is structural. When you ask an answer engine a question, it does not look up a saved ranking for that query. It performs a fresh retrieval, pulls back a candidate set of pages, reads a subset, and composes an answer grounded in whatever it just read. The citations you see are a byproduct of that retrieval run, at that moment. Nothing about the process is designed to keep naming the same source it named yesterday. Consistency, when it happens, is an emergent property of your sources being repeatedly the best available — not a memory the engine keeps of you.

So decay is the default and durability is the exception. A citation is a snapshot of the engine's judgment under conditions that are quietly shifting underneath you. When enough of those conditions move, your snapshot expires.

Half-life is not the same as volatility

I want to draw a sharp line between two things that get conflated. Answer volatility is the day-to-day flicker — ask the same question three times and you get three slightly different shortlists. That is noise around a stable mean, and I have written about it elsewhere. Citation half-life is different: it is the trend, not the flicker. It describes the slow, directional loss of your position over weeks. Volatility is the weather; half-life is the climate.

You can be low-volatility and short half-life at the same time — steadily named this month, quietly gone by next quarter. That combination is the dangerous one, because the steadiness lulls you. Everything looks fine right up until it isn't. Measuring only whether you appear today tells you nothing about the slope you are on.

The three forces that shorten your half-life

When I dig into why a client's mentions have faded, the cause almost always lands in one of three buckets. They compound, which is why decay often feels sudden even though it was gradual.

1. Freshness decay

Answer engines lean toward recency, especially for anything that could plausibly have changed — pricing, availability, "best of" lists, rankings, comparisons, anything with a year in the query. A page that was clearly current when it was written starts reading as stale the moment a fresher alternative exists that says roughly the same thing. You did not get worse. The field around you got newer. If your cornerstone content carries a 2024 feel and a competitor publishes a genuinely updated 2026 version, the engine has a cheap reason to prefer them, and it will take it.

2. Competitive displacement

Citation slots are scarce by design — an answer names a handful of sources, not forty. That makes AEO a game of relative position, not absolute quality. Every time a competitor earns a new review, gets covered by a publication the engine trusts, tightens their entity data, or publishes a cleaner answer to the exact question, they are not just helping themselves. They are pushing someone out of the small set of named sources, and sometimes that someone is you. You can do nothing wrong and still lose the slot because the person next to you did something right.

3. Model and index turnover

The engines themselves change under you. Models get retrained. Retrieval indexes get refreshed. The rules for which sources clear the trust bar get quietly adjusted. A model update can reshuffle who gets named for a whole category overnight, and you will get no notice and no changelog entry with your name in it. This is the force you cannot control at all — which is exactly why the other two matter so much. The parts you can influence are the ones worth compounding.

How to measure your own half-life

You cannot manage what you refuse to look at, and "do we show up in ChatGPT?" checked once a month is not looking. Here is the measurement discipline I actually use.

1. Fix a prompt set. Write down the 15 to 30 real questions a customer would ask an assistant in your category and city. Freeze the list. This is your instrument, and an instrument you keep changing measures nothing.
2. Sample on a schedule. Run that set across the engines that matter to you — for most of my clients that is ChatGPT, Perplexity, and Google's AI Mode — on a fixed cadence. Weekly is plenty for most; daily if you are in a fast-moving or contested category.
3. Record presence, not vibes. For each run, log a simple binary per prompt per engine: were you named or not. Optionally note where you landed in the answer, because slipping from first-named to last-named is an early warning that a full drop is coming.
4. Watch the trend line, then find the slope. Plot the share of prompts that name you over time. Your citation half-life is roughly the interval over which that share falls by half. When you see the slope turning down, you are watching decay in progress — while there is still time to act.

You do not need a fancy platform to start. A spreadsheet and a standing calendar reminder will surface the slope. The point is to convert "I think we're still showing up" into a number you can watch move.

What actually lengthens a half-life

The good news is that decay is a rate, and rates can be slowed. The businesses whose mentions persist are not the ones who did one brilliant thing. They are the ones who kept a few boring things alive.

Keep your cornerstone answers genuinely current. Not a cosmetic date change — a real refresh of the numbers, examples, and claims, on a cadence, so the freshest correct answer in your category is repeatedly yours. This single habit fights freshness decay and competitive displacement at the same time.

Feed corroboration continuously. The engines trust what multiple independent sources agree on. A steady drip of reviews, mentions, and third-party references keeps the web's consensus about you fresh, which is the thing an engine can verify cheaply. Corroboration is not a one-time campaign; it is a metabolism.

Own the exact question. Displacement usually happens at the level of a specific query. If you can see which questions you are losing, you can publish the cleaner, more directly-answering version of that specific answer and take the slot back. Vague "content" does not defend a slot. A page that answers one real question better than anyone else does.

Shore up your entity. Consistent name, category, and location data across the sources the engines read makes you cheaper to trust and harder to confuse for a competitor — which raises the bar a challenger has to clear to displace you.

Why the half-life is getting shorter, not longer

If anything, I think the pace of decay is accelerating, and the reason is that the engines themselves are changing faster than they used to. Through 2026, Google has been pushing its AI Mode from an experiment toward the center of search, the assistants keep tightening how they weigh freshness and trust, and the whole retrieval layer underneath is being rebuilt more or less continuously. Each of those shifts is a reshuffle. A category that felt settled in the spring can look unrecognizable by the fall, not because anyone in it got worse, but because the machine deciding who to name changed its mind.

This is the part that frustrates the "set it and forget it" crowd, and I understand why. You did the work, you earned the mentions, and then the ground moved. But I would reframe it: if the engines are re-deciding more often, then the businesses that keep showing up are precisely the ones present at each re-decision. Frequent reshuffles punish the absent and reward the consistent. The faster the churn, the more a steady tending habit compounds — because your competitors, most of whom did the work once and walked away, are decaying while you are refreshing.

A quick illustration

Picture two businesses that both got named by ChatGPT in July. The first treated it as a win and moved on. The second kept publishing a genuinely updated answer every few weeks, kept its reviews flowing, and kept an eye on its prompt set. By September, a model refresh and two competitor moves have quietly rewritten the shortlist. The first business is gone and does not know it — nobody sent a notice. The second slipped for a week, saw the slope turn on its own tracking, published the sharper answer to the query it was losing, and took the slot back. Same starting point. Completely different outcome. The only variable was whether anyone was watching the slope.

The honest hedge

Everything above is a working model, not a measured constant. "Citation half-life" is a lens I find useful, not a number I can hand you for your industry — it varies wildly by category, by how contested your space is, and by how often the engines you care about refresh. My read is that competitive categories decay in weeks and sleepy ones in months, but that is a pattern I have watched, not a guarantee I can make. The mechanism, though, I am confident about: answer engines re-decide constantly, slots are scarce, and standing still is a slow slide. Treat your AI visibility like a garden, not a monument. The monument crowd wonders where their mentions went. The garden crowd is still being named, because they never stopped tending.

About the author

Scott Tischler is the Founder & Chairman of AIrecommend.ai and a practitioner-authority on AI search and Answer Engine Optimization. With 20+ years in marketing technology — including American Express, MetLife, and UBS — and executive and professional study at Wharton, Harvard, and Oxford, he helps businesses become the ones AI recommends.