The Missing AEO Metric: Citation Half-Life

TL;DR

Most answer engine optimization (AEO) programs count citations the way they count backlinks: cumulative, permanent, always going up. On a recent Distribution Podcast episode, Ravish Agrawal of Gamma floated a more uncomfortable idea — that AI citations might decay, behaving less like a monument and more like a half-life. This is a hypothesis, not a measured fact, and Ravish framed it that way. But if it holds even partly true, it reframes AEO as an always-on engine rather than a one-time publishing task. This piece introduces two metrics worth tracking now — time-to-citation and citation half-life — and how to measure and act on them.

Why citations aren't permanent

Most teams carry a backlink-era assumption into AEO: publish a great page, earn a citation in ChatGPT or Perplexity, and that citation stays put. Links are durable, so surely mentions in AI answers are too.

Ravish Agrawal of Gamma questioned that with a carbon-dating analogy. Carbon-14 doesn't disappear on a fixed date; it decays at a predictable rate, so you measure how much is left to estimate age. He suggested AI citations might behave similarly — a piece of content contributes strongly to model answers at first, then less and less over time as fresher material, new model versions, and shifting retrieval indexes crowd it out. Framed as a hypothesis, not established data, the implication is that a citation you earned three months ago may be quietly fading right now, and dashboards built to count rather than track decay would never show it.

To make the idea concrete, Ravish reached for a number: a roughly three-month citation half-life. Be precise here — that figure was a hypothetical example to illustrate the concept, not a benchmark Gamma measured [VERIFY — ~3-month half-life was explicitly a hypothetical, per the source ledger; never present as data]. No one has published a validated half-life for AI citations. The value isn't the number. It's the shift from treating citations as a stock you accumulate to a flow you have to keep feeding.

That shift is why Ravish's team treats AEO as an always-on engine: because citation decay isn't well understood, the safe assumption is that visibility erodes unless actively maintained.

Two metrics you're not tracking: time-to-citation and citation half-life

If citations both take time to appear and fade over time, then two variables matter that almost nobody instruments today.

1. Time-to-citation — how long between publishing (or redistributing) a piece of content and its first appearance as a citation in a given AI surface.

This is not uniform across platforms, and the gaps are large. From Ravish's own observations at Gamma:

• New Reddit content took roughly 30–45 days to surface as a ChatGPT citation [attributed observation; a range, not a guarantee].
• Perplexity picked up a specific new phrase as fast as the next day [anecdotal, single observation].
• In a separate anecdote, the host described a niche Reddit query appearing in a Google AI Overview within one day [single observation].

The takeaway isn't a formula — it's that latency is platform-specific and wide-ranging. ChatGPT may sit weeks behind Perplexity for the same source. Judge a campaign "failed" at day 20 because ChatGPT hasn't cited it, and you may simply be measuring before ChatGPT's pickup window has opened.

2. Citation half-life — how long a piece of content keeps contributing citations before its citation share drops by half.

This is the decay side. Even if you can't yet compute a precise half-life, you can watch the shape: is a page's citation share holding, plateauing, or sliding? A page that earned a steady share of your AI mentions for weeks and then trends down is telling you something a cumulative counter never will.

Notably, Ravish mentioned that Gamma was tracking this manually in a spreadsheet — logging citations by hand — because the tooling they used (he referenced Profound) didn't surface a half-life view [attributed; verify exact tool capabilities before repeating as a product claim]. That's the current state of the art for a lot of serious AEO teams: the metric matters enough to track by hand, but isn't yet a standard column in most dashboards.

How to measure them

You don't need a research lab to start. You need disciplined logging and the willingness to look at curves instead of weekly totals.

Log four fields per cited asset:

• Publish date — when the content (or a redistribution of it, like a Reddit post or a third-party mention) went live.
• First-cited date — the first date you observed it cited on each AI surface, tracked per platform (ChatGPT, Perplexity, Gemini, Google AI Overviews, Copilot). The gap between fields 1 and 2 is your time-to-citation for that surface.
• Last-seen date — the most recent date you observed it still being cited. The span between first-cited and last-seen, watched over time, is how you approximate citation half-life.
• Source URL — the exact page or post being cited, so you can distinguish your owned content from third-party coverage and from social surfaces like Reddit.

Then change what you plot. Weekly citation counts are a vanity line — they go up as long as you publish anything. Instead, build survival curves: for a cohort of content published in a given month, plot the percentage still earning citations at week 4, 8, 12, and beyond. That's the view that reveals decay. It answers the question a counter can't: of everything we earned, how much is still working?

A few practical notes:

• Track per platform, not in aggregate. Given the latency spread, a blended number hides the signal — ChatGPT and Perplexity are different channels with different clocks.
• Separate owned, third-party, and social sources. Their decay may differ, and so does your response: you refresh an owned page directly but influence third-party and social surfaces indirectly.
• Date-stamp everything. These are observations in a fast-moving environment. A half-life estimate from this quarter may not hold next quarter.

What to do about decay

The point of measuring decay is to act before visibility is gone. If the hypothesis holds, AEO stops being publish-and-forget and becomes a maintenance discipline.

Refresh when citation share drops. When a page's share of your AI citations starts sliding on a survival curve, treat it as a re-optimization trigger. Update the content with current data, tighten the answer-shaped sections AI engines extract, and re-timestamp it — you're trying to re-enter the fresh-content window that earned the citation the first time.

Redistribute, don't just republish. Because surfaces have different pickup windows and favored sources, decay on one channel is a cue to seed the message on another. A fading owned-page citation might be reinforced by a new third-party mention or a community post that ChatGPT and Google AI Overviews pick up on their own schedules.

Run it as an always-on engine. Assume visibility erodes, and budget continuous effort to maintain it. Set a cadence — monthly cohort reviews of your survival curves — so refreshes and redistribution are scheduled, not reactive.

This is exactly the discipline Asva AI's citation intelligence and brand visibility tracker are built to support: monitoring where and when you're cited across AI surfaces, per platform, over time — so decay shows up as a trend you can act on rather than a surprise you find after the traffic is gone. Building an AEO program from scratch? Our guide to answer engine optimization covers the foundations these metrics sit on.

FAQs

What is citation half-life in AEO?

Citation half-life is a proposed metric for how long a piece of content keeps earning citations in AI answers before its citation share drops by half. It was raised as a hypothesis by Ravish Agrawal of Gamma on the Distribution Podcast — not as a measured benchmark. The idea borrows from radioactive decay: citations may fade at a rate rather than disappearing on a fixed date.

Is the three-month citation half-life a proven number?

No. The roughly three-month figure was offered as a hypothetical example to illustrate the concept, not data Gamma or anyone else has validated. Treat it as a thinking tool, not a target. The useful takeaway is the behavior — decay — not the specific duration.

What is time-to-citation?

Time-to-citation is the gap between publishing (or redistributing) content and its first appearance as a citation on a specific AI surface. It varies widely by platform. Ravish observed new Reddit content taking roughly 30–45 days to appear in ChatGPT, while Perplexity picked up a new phrase as fast as the next day.

How do I measure citation half-life?

Log publish date, first-cited date, last-seen date, and source URL for each cited asset, tracked per platform. Then plot survival curves — the percentage of a content cohort still earning citations at weeks 4, 8, and 12 — instead of cumulative weekly counts. The declining share over time approximates the half-life.

Why should AEO be treated as an always-on engine?

Because if citations decay, visibility erodes unless it's actively maintained. Ravish Agrawal framed AEO this way precisely because citation decay isn't yet well understood, so the safe assumption is that content needs ongoing refreshing and redistribution rather than a one-time publish.

Do all AI platforms cite content on the same timeline?

No. Observations shared on the podcast suggest large differences — ChatGPT lagging weeks behind Perplexity for the same source, and Google AI Overviews sometimes surfacing content within a day. This is why time-to-citation and citation share should be tracked per platform, not as a single blended number.

Close

Citation half-life is, for now, a hypothesis — one practitioner's carbon-dating analogy, not a law of AI search. But you don't have to believe the specific number to benefit from the shift in thinking. The moment you stop treating citations as permanent and start tracking time-to-citation and citation share over time, your AEO program gains an early-warning system instead of a rear-view mirror.

Want to see how your brand is cited across AI surfaces right now? Run a free AI Visibility Report to get a baseline, explore your citation trends in the Asva app, or book a 30-minute walkthrough to map an always-on AEO program. If you do one thing today: start with the free AI Visibility Report and log the first-cited date on everything you publish from here on.

---