Why AI Citations Decay: The 4.5-Week Half-Life and the Freshness Program

AI citations are not permanent. A 2026 Profound analysis of 240M citations found a median cited-source half-life of roughly 4.5 weeks — 40-60% of the domains an AI cites for a given query change month-to-month, and 70-90% rotate over six months. Roughly half of all AI-cited content is under 13 weeks old. This makes GEO a continuity program, not a one-time campaign: earning a citation is the start, and holding it requires a sustained freshness cadence. The brands that treat GEO as "optimize once and benefit forever" lose their citations to fresher competitors within about a month.

For the first two years of GEO as a discipline, the implicit assumption was build-once-and-benefit: optimize a page, earn the citation, keep it. The 2026 research round overturned that assumption. AI citations decay — fast — and the brands winning AI search are not the ones who optimized best once, but the ones who maintain a continuity program.

This guide covers the citation decay data, why it happens, and the freshness program that holds citations once you earn them.

Last updated: May 2026

AI citations decay with a ~4.5-week median half-life. GEO is therefore a continuity program, not a campaign — the page you optimized two months ago is likely already losing citations to fresher competitors. The strategic shift this forces: build a freshness cadence into your GEO program from day one, and measure citation retention, not just citation acquisition.

The citation decay data

A 2026 Profound analysis of 240M citations found a median cited-source half-life of roughly 4.5 weeks, with 40-60% of cited domains rotating month-to-month for an identical query and 70-90% rotating over six months. Around half of all AI-cited content is under 13 weeks old.

The headline numbers from the 2026 research round on citation permanence:

• Median citation half-life ≈ 4.5 weeks. For a given query, the set of sources an AI cites turns over fast — half the citation "value" of a source erodes within roughly a month (Profound, 240M-citation analysis).
• 40-60% of cited domains rotate month-to-month. Ask an AI the same question 30 days apart and nearly half the cited domains will be different.
• 70-90% rotate over six months. Over a longer horizon, the citation set is almost entirely different.
• ~50% of AI-cited content is under 13 weeks old. Recency is not a tiebreaker — it is a dominant selection signal.
• Distributed, earned content decays slower — roughly 2x slower than single-site content, according to survival-analysis research. Citations spread across many third-party sources are more durable than citations concentrated on one domain.

Treat the exact figures as directional — citation-decay research is young, and several of these studies come from GEO-tool vendors with a commercial interest in the durability finding. But the phenomenon is cross-corroborated and clear: AI citations are not stable assets. They are more like a position in a fast-moving ranking that must be continuously defended.

The numbers vary by study, but the direction is unambiguous: AI citations decay on a roughly monthly timescale. A citation earned today is, statistically, more likely than not to be gone in two months unless the underlying content stays fresh and competitive.

Why AI citations decay

AI citations decay for four reasons: AI models and retrieval indexes refresh continuously, competitors publish fresher content, the retrieval pipeline favors recency as a selection signal, and AI search results are stochastic — the same query returns different sources across runs.

Decay is not a bug. It is the natural consequence of how AI search works.

Reason 1: Indexes and models refresh continuously

AI search engines re-crawl the web, re-embed content, and update their retrieval indexes constantly. Each refresh re-evaluates which sources best answer a query. A page that was the best available answer last month may be outranked this month — not because it got worse, but because the evaluation ran again with new candidates.

Reason 2: Competitors publish fresher content

Every week, competitors publish new content targeting the same queries. The retrieval pipeline does not grandfather in existing citations — it re-scores the full candidate pool. A competitor's newer, well-structured page directly displaces an older one. In a category where multiple brands are actively doing GEO, citations rotate among them continuously.

Reason 3: Recency is a selection signal

AI search engines explicitly favor recent content for a large share of queries — roughly half of cited content is under 13 weeks old. As your page ages past that window, its recency signal weakens even if nothing else changes. The page is being penalized for the calendar, not for quality.

Reason 4: AI search is stochastic

Identical queries return different citations across runs. Part of what looks like "decay" is actually variance — the citation set genuinely fluctuates query-to-query. This is why a single "are we still cited?" check is unreliable, and why decay must be measured as a trend across repeated samples, not a single observation.

Citation decay is structural, not a failure of your content. Continuous index refreshes, a constant stream of competitor content, an explicit recency preference, and inherent stochasticity all push citations to rotate. You cannot stop decay — you can only out-pace it with a freshness cadence.

The freshness program: how to hold citations

Holding AI citations requires a freshness program: review every important page on a ≤13-week cycle, make each refresh a genuine ≥20% substantive change, sync visible dates with structured-data dates, and sustain off-site activity at a steady cadence rather than in one-off bursts.

If decay is structural, the response is a program, not a project. Four components:

Component 1: The 13-week review cycle

Every page that matters for AI citation should be reviewed at least every 13 weeks — the working freshness window. Put your priority pages on a rolling schedule so each one is touched before its recency signal fully decays. This is the single most important operational change: GEO content needs a maintenance calendar, the way a software product needs a release cadence.

Component 2: Substantive refreshes, not date bumps

A refresh only earns freshness credit if it is a genuine ≥20% substantive change — updated statistics, a new section, revised analysis, a "what changed" note. Changing the visible date without changing the content does not work; AI systems (and Google) detect cosmetic-only updates. The refresh has to be real.

Component 3: Sync the dates

When you refresh, update the visible "Updated [Month Year]" date and the structured-data dateModified field together. A mismatch — a fresh-looking visible date with a stale dateModified, or vice versa — sends a confused signal. The two should always agree, and both should reflect a genuine update.

Component 4: Sustained off-site cadence

Off-site citations decay too — and a single earned-media placement decays within roughly 4-5 weeks. A sustainable off-site presence is a steady cadence (one to two placements a month, continuous community participation) rather than a one-off campaign. Distributed, earned content decays roughly 2x slower than single-site content, so a broad, continuously-replenished off-site footprint is the most durable citation asset you can build.

What a freshness program looks like in practice

• A maintenance calendar with every priority page on a 13-week rotation
• A "substantive refresh" standard — each refresh adds real new value
• Visible date and dateModified synced on every update
• A monthly off-site cadence (placements, community contributions) rather than sporadic bursts
• Citation retention tracked as a metric, not just citation acquisition

The freshness program has four parts: a 13-week review cycle, substantive ≥20% refreshes, synced dates, and a sustained off-site cadence. The mindset shift is from "launch and benefit" to "maintain and defend." GEO content is a living asset with a maintenance calendar, not a finished deliverable.

What citation decay means for GEO strategy

Citation decay reshapes GEO strategy in three ways: budget must cover ongoing maintenance not just creation, measurement must track retention over rolling windows, and content portfolios should favor fewer continuously-maintained pages over many neglected ones.

The decay finding has direct strategic consequences.

Budget for maintenance, not just creation

A GEO budget that funds content creation but not ongoing maintenance will see its citations decay away. Allocate explicitly for the refresh cycle — the maintenance line item is not optional, it is what protects the creation investment.

Measure retention, not just acquisition

Most GEO measurement asks "did we get cited?" The decay finding adds a second, equally important question: "are we still cited?" Track citation retention over rolling windows. A program that earns 20 citations a month but loses 18 is barely treading water; one that earns 12 and loses 4 is compounding. Acquisition without retention is a leaky bucket.

Favor depth over breadth in the content portfolio

If every page needs ongoing maintenance, a portfolio of 200 neglected pages is worse than 40 continuously-maintained ones. Decay rewards focus: a smaller set of pages that each stay fresh will out-cite a larger set that goes stale. Resist the urge to maximize page count beyond what you can realistically maintain on a 13-week cycle.

Use natural content churn as an advantage

Some businesses have genuine, ongoing data churn — prices change, inventory updates, new information arrives. That churn is a freshness advantage if you surface it: a regularly-updated "current state" page turns real business change into a recurring freshness signal that competitors with static content cannot match.

Citation decay turns GEO from a creation discipline into a maintenance discipline. Budget for the refresh cycle, measure retention alongside acquisition, keep the content portfolio small enough to actually maintain, and turn any natural data churn in your business into a freshness advantage.

Frequently asked questions

How fast do AI citations actually decay?

A 2026 Profound analysis of 240M citations found a median cited-source half-life of roughly 4.5 weeks. For a given query, 40-60% of cited domains change month-to-month, and 70-90% rotate over six months. Treat the exact figures as directional — the research is young — but the phenomenon of fast decay is well cross-corroborated.

Does this mean GEO doesn't work?

No — it means GEO is a continuity program rather than a one-time campaign. GEO absolutely works; the citations are real and valuable. But they must be defended. The brands winning AI search are not the ones who optimized best once — they are the ones who maintain a freshness cadence. Decay changes how you run GEO, not whether you do it.

Why do AI citations decay so quickly?

Four structural reasons: AI models and retrieval indexes refresh continuously and re-score the candidate pool; competitors constantly publish fresher content that displaces older pages; AI search engines explicitly favor recent content as a selection signal; and AI search is stochastic, so the citation set genuinely fluctuates between runs. Decay is built into how AI search works.

What is the 13-week freshness rule?

Roughly half of all AI-cited content is under 13 weeks old, which makes 13 weeks the working freshness window. The rule: review every important page at least every 13 weeks, and make each refresh a genuine ≥20% substantive change — not a date-stamp bump. Pages reviewed on this cycle hold their citations far better than pages left static.

Does a date change count as a refresh?

No. A refresh only earns freshness credit if it is a substantive change — updated statistics, a new section, revised analysis. AI systems and Google both detect cosmetic-only date bumps. The visible date and the structured-data dateModified field should be updated together, and both should reflect a real content change.

Is off-site content also affected by decay?

Yes — a single earned-media placement decays within roughly 4-5 weeks, just like on-site content. But distributed, earned content decays roughly 2x slower than single-site content. The implication: a broad off-site footprint, continuously replenished at a steady monthly cadence, is the most durable citation asset you can build.

How do I measure citation retention?

Track your citation rate for a fixed query set over rolling windows — measure the same queries every 2-4 weeks and watch the trend, not any single check (AI search is stochastic, so single checks are noise). Retention is the share of last period's citations you still hold this period. A healthy GEO program retains far more than it loses.

Should I publish fewer pages because of decay?

For most teams, yes — favor depth over breadth. If every page needs maintenance on a 13-week cycle, a portfolio of 40 continuously-fresh pages will out-cite 200 neglected ones. Publish only as many pages as you can realistically keep fresh. Decay rewards focus.

Which content decays slowest?

Distributed, earned content — citations spread across many third-party sources — decays roughly 2x slower than citations concentrated on a single domain. Original research also tends to hold citations longer, because the underlying data remains the canonical source even as it ages. Concentrated single-site citations decay fastest.

How does citation decay interact with the rest of GEO?

Decay does not replace the other GEO fundamentals — content structure, factual density, entity signals, earned media — it adds a time dimension to all of them. Every GEO tactic now has a "and keep it fresh" clause. The freshness program is the layer that protects every other investment you make in GEO.