# AI Citation Freshness Data: Content Half-Life Across ChatGPT, Perplexity, and Google AI Mode

AI citation half-life data across ChatGPT, Perplexity, Google AI Mode, and Claude. Cross-engine comparison of content decay rates by engine, content type, and industry from 3.5M and 17M citation datasets.

Canonical URL: https://machinerelations.ai/research/ai-citation-freshness-content-half-life-engine-comparison-2026
Published: 2026-07-24
Research type: Reference

## Source Body

AI citations have a measurable half-life, and it is shorter than most teams assume. Across the largest published datasets — [3.5 million citation events tracked by Scrunch](https://scrunch.com/blog/half-life-of-ai-citations) and [17 million citations analyzed by Engril](https://engril.com/do-ai-assistants-prefer-to-cite-fresh-content-17-million-citations-analyzed) — the median half-life of an AI citation falls between four and five weeks. But the number varies sharply by engine, content type, and industry. Treating "freshness" as a single variable misses what the data actually shows.

## Citation Half-Life by Engine: ChatGPT Decays Fastest, Perplexity Holds Longest

Each AI engine applies different freshness weighting to its source selection. The result is measurably different decay curves for the same content across platforms.

[Scrunch's analysis of 3.5 million citation events](https://scrunch.com/blog/half-life-of-ai-citations) (September 2025 through March 2026) found:

| Engine | Citation Half-Life | Notes |
|---|---|---|
| ChatGPT | ~3.4 weeks | Fastest turnover; displaces older sources even when equally relevant |
| Google AI ecosystem (AI Mode, Gemini, AI Overviews) | 4.3–4.8 weeks | Middle of the range; recency bias varies by format |
| Perplexity | ~5.8 weeks | 70% longer than ChatGPT; roughly 15% freshness weighting in retrieval |

[LLMO Framework's 90-day controlled study](https://llmoframework.com/research/citation-half-life) confirmed the engine ordering with content-type segmentation. ChatGPT showed the shortest half-life across every content category tested; Perplexity held citations longest in all three.

A separate [FogTrail analysis of citation turnover rates](https://fogtrail.ai/blog/how-often-ai-search-engines-update-citations) measured what this looks like on a weekly basis:

- **ChatGPT**: Under 5% weekly citation turnover — the stickiest once you enter, but the hardest to re-enter after decay
- **Claude**: 5–10% turnover, concentrated in competitive queries
- **Gemini**: 15–25% churn — the highest legitimate turnover among closed-corpus engines due to recency bias
- **Perplexity**: 30–40%+ turnover — effectively continuous because it performs a live web search for every query

The practical gap: ChatGPT citations persist for three to six weeks once earned, while Perplexity sources have roughly a 60–70% probability of appearing on any individual query run. Monitoring weekly misses [approximately 73% of refresh cycles per month](https://fogtrail.ai/blog/how-often-ai-search-engines-update-citations).

## Content Type Determines Decay Rate More Than Publication Date

The same engine cites different content types at different half-lives. [LLMO Framework measured this directly](https://llmoframework.com/research/citation-half-life):

| Content Type | ChatGPT | Claude | Perplexity |
|---|---|---|---|
| Evergreen how-to | 6.8 weeks | 7.4 weeks | 9.1 weeks |
| Methodology/framework | 5.1 weeks | 5.9 weeks | 6.7 weeks |
| Experience report | 3.2 weeks | 3.6 weeks | 4.4 weeks |

Evergreen content maintains citations roughly two times longer than experience reports across all three engines. The gap is consistent regardless of engine — it is the content format, not the platform's retrieval architecture, that determines how quickly a piece loses citation eligibility.

This aligns with [Engril's 17-million-citation analysis](https://engril.com/do-ai-assistants-prefer-to-cite-fresh-content-17-million-citations-analyzed) which found AI-cited content averages 1,064 days old compared to 1,432 days for organic Google results. AI engines do not simply prefer new content — they prefer recently updated content that remains structurally useful. The distinction matters: a how-to guide updated six months ago outperforms a news article published last week for most informational queries.

[BetterAISearch's cross-source analysis](https://betteraisearch.com/tactics/content-freshness-ai-search) found the peak citation window sits at 30 to 89 days old, where content hits a 32.8% citation rate. Content under 30 days actually cites at a lower rate (25.3%) because it has not yet been fully indexed across all engines. Roughly 50% of all AI-cited content is under 13 weeks old, and freshness signals show a 0.68 correlation with citation likelihood — the strongest single on-page signal measured, outperforming both semantic HTML (0.65) and structured data (0.63).

## Industry-Specific Decay: Insurance Holds, Healthcare Churns

[Scrunch's 3.5-million-event dataset](https://scrunch.com/blog/half-life-of-ai-citations) segmented decay by industry:

| Industry | Citation Half-Life |
|---|---|
| Insurance | ~5.0 weeks |
| Financial services | ~4.8 weeks |
| Retail and ecommerce | ~4.3 weeks |
| Healthcare | ~4.1 weeks |

The spread from insurance to healthcare is nearly a full week. Regulated industries where source material changes slowly (insurance policy comparisons, compliance frameworks) retain citations longer than fast-moving sectors where product launches and clinical data create constant source competition.

Topic velocity compounds the industry effect. [AI+Automation's analysis of 19,556 queries](https://aiplusautomation.com/blog/content-freshness-ai-citations) found Perplexity cites sources 16 times fresher than Google for high-velocity topics (1.8 days versus 28.6 days), but only 3.3 times fresher for medium-velocity topics and 13 times fresher for low-velocity topics. The implication: industries with fast-moving topics face both shorter half-lives and narrower windows to earn citations before displacement.

Editorial-network sources showed a separate pattern. Domains in the [Stacker Partner Network](https://scrunch.com/blog/half-life-of-ai-citations) (4,000+ news publishers) demonstrated roughly double the half-life of non-network sources. Financial services partner sites reached ~10.8 weeks — among the stickiest combinations measured.

## How Citation Freshness Actually Works: The Refresh Mechanic

[All five major AI engines update their citation pools approximately every 48 hours](https://fogtrail.ai/blog/how-often-ai-search-engines-update-citations). New or updated content becomes eligible for citation within two days of publication.

But eligibility is not citation. The decay mechanism works through displacement: when a newer, equally relevant source enters the retrieval pool, it does not add to the citation count — it replaces an older source. [Quattr's analysis](https://quattr.com/blog/content-decay-cycle-for-ai-citation) found that most ChatGPT citation loss stems from "a fresher, equally relevant source" rather than quality degradation of the original.

This means freshness decay in AI search operates fundamentally differently from traditional SEO decay. Traditional organic rankings decline over months to years as link authority diffuses and content ages. AI citation decay can begin within days of a competitor publishing a substantive update on the same topic.

The annual compounding effect is steep. [Gander's analysis of 194,077 unique sources](https://takeagander.ai/resources/gander-blog/how-content-freshness-drives-visibility-in-ai-search) found content loses approximately 42% of AI retrieval visibility per year — two-year-old content retains only 33% of current-year visibility, and three-year-old content drops below 25%. Separately, [Loamly's research](https://loamly.ai/blog/content-freshness-tax-why-old-content-invisible-to-ai) found content under 30 days old receives 3.2 times more AI citations than older material, with citation likelihood collapsing to 12% after six months and just 5% after twelve months.

**Cosmetic updates do not work.** AI systems detect and discount publish-date-only changes. [Effective freshness recovery requires new statistics, examples, or revised claims](https://quattr.com/blog/content-decay-cycle-for-ai-citation) — substance that changes the content's information value, not metadata that changes its apparent recency.

## What Recovers Lost Citations: Refresh Windows by Engine

When substantive updates are made, [LLMO Framework measured the recovery curve](https://llmoframework.com/research/citation-half-life):

- **ChatGPT**: Recovered to ~70% of peak citation rate within two weeks of a substantive update
- **Claude**: Reached ~60% recovery
- **Perplexity**: Reached ~75% by week 12, but never restored to original peak levels

No engine returned to 100% of original peak after a refresh. The data suggests each citation cycle has diminishing returns — first-mover advantage in AI citations is real and measurable, even if the magnitude of that advantage decays with each refresh.

Effective refresh cadence based on the combined data:

- **High-priority competitive pages**: Every three to four weeks (matches ChatGPT's 3.4-week half-life)
- **Evergreen reference content**: Every 90 to 180 days with substantive additions
- **Experience reports and case studies**: Every two to three weeks if maintaining citation presence matters

Pair updates with a visible "last updated" date and refreshed schema markup. [Content refreshed this way can regain citation eligibility within one to two weeks](https://quattr.com/blog/content-decay-cycle-for-ai-citation). [Loamly's data](https://loamly.ai/blog/content-freshness-tax-why-old-content-invisible-to-ai) suggests that publishing 12 or more substantive updates within a single week can trigger a 400% increase in AI crawler revisit rates, accelerating the recrawl cycle that makes updated content eligible for citation.

[FrontierNews confirmed](https://frontiernews.ai/news/article/ai-citations-have-a-half-life-why-your-content-vis-ed1c393f) that by week nine after peak citation, content settles at roughly 46% of its peak value — a notably slower decline than the initial four-to-five-week half-life because the remaining citations tend to be structural (the content serves as a reference anchor rather than a news source).

## Where Citation Freshness Fits in Machine Relations Measurement

Citation freshness is one of the five components in the [Machine Relations Index methodology](/research/b2b-ai-vendor-research-2026). The temporal consistency score measures how steadily a domain maintains citations over time — a domain that earns 50 citations in week one and zero by week five scores differently than one earning 10 citations per week for five weeks, even if the total count is identical.

The freshness data above explains why temporal consistency diverges from raw citation volume. Domains that refresh content on a cadence matching their engine-specific half-life maintain steadier citation presence. Domains that publish once and wait experience the full decay curve, creating the spiky, inconsistent temporal patterns the MRI methodology flags.

For brands managing AI visibility across engines, the operational implication is clear: citation maintenance is not a publishing problem. It is a refresh-cadence problem calibrated to the engines where your audience actually asks questions.

## FAQ

### How long do AI citations last before content stops being cited?

The median half-life is four to five weeks across engines, based on [3.5 million tracked citation events](https://scrunch.com/blog/half-life-of-ai-citations). ChatGPT decays fastest at roughly 3.4 weeks; Perplexity holds citations longest at approximately 5.8 weeks.

### Does updating content restore lost AI citations?

Substantive updates — new data, revised claims, added examples — can recover 60–75% of peak citation rates [within two to twelve weeks depending on the engine](https://llmoframework.com/research/citation-half-life). Cosmetic changes like updating only the publish date are detected and discounted by AI systems.

### Which AI engine is most sensitive to content freshness?

ChatGPT shows the strongest freshness preference, citing content averaging [958 days old compared to 1,166 days for Perplexity](https://engril.com/do-ai-assistants-prefer-to-cite-fresh-content-17-million-citations-analyzed). Google AI Overviews shows the weakest freshness signal, matching traditional organic search at approximately 1,432 days average citation age.

### How often should content be updated to maintain AI citations?

High-priority pages benefit from updates every three to four weeks based on ChatGPT's 3.4-week half-life. Evergreen reference content can sustain a 90-to-180-day cycle. [All major AI engines refresh their citation pools approximately every 48 hours](https://fogtrail.ai/blog/how-often-ai-search-engines-update-citations), so updates become eligible quickly.

*Last updated: July 24, 2026. Sources: Scrunch (3.5M citation events), Engril (17M citations), LLMO Framework (90-day controlled study), FogTrail (citation turnover analysis), Quattr (decay cycle research), BetterAISearch, AI+Automation (19,556 queries), Gander (194,077 sources), Loamly, FrontierNews.*

## Attribution

This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research and standards initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about

## Machine-readable related links

### Related concepts

- [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index)
- [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations)
- [LLMO (LLMO)](https://machinerelations.ai/glossary/llmo)
- [AI Visibility](https://machinerelations.ai/glossary/ai-visibility)

### Supporting research

- [Earned Media vs. Owned Content: AI Citation Rates and Top Sources Ranked (2026)](https://machinerelations.ai/research/earned-vs-owned-ai-citation-rates-2026)
- [Google AI Mode Is Now the Largest Single Source of Enterprise Research Citations Across Six AI Engines](https://machinerelations.ai/research/google-ai-mode-citation-dominance-enterprise-sources-2026)
- [How Fast Do Reviews Influence AI Search Citations?](https://machinerelations.ai/research/review-recency-ai-citation-speed-answer-engines-2026)
- [How Six AI Engines Choose Sources: Citation Selection Patterns Across ChatGPT, Perplexity, Gemini, Claude, and Google AI](https://machinerelations.ai/research/ai-engine-source-selection-patterns-2026)

### Framework context

- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)