HomeResourcesCitation stability

Citation stability

The Half-Life of an AI Citation: How Fast AI Answers Change Their Sources

Getting cited once is not the same as staying cited. New data on weekly citation drift and content freshness shows how quickly AI answers move on.

By MentionMethod 6 min read
74%of ChatGPT’s sources change each week
1.4%of cited news articles remain permanently
25.7%fresher content cited by AI than by Google search
Key takeaways
  • The specific pages AI engines cite change fast: about 74% of ChatGPT’s sources and 56% of Google AI Mode’s change week to week. AI Overviews are steadier at 5%.
  • The websites behind those pages are far more stable. BrightEdge found 96.8% of cited domains saw no change week over week. The site stays; the page rotates.
  • News is the most short-lived source: only 1.4% of cited news articles stayed cited permanently. Video was the stickiest format.
  • AI assistants favor fresher content, citing pages 25.7% newer on average than those in Google’s organic results.

Citations are not a one-time win

A brand gets cited in an AI answer, someone screenshots it, and it goes into a slide as proof of success. A week later, the same question may return a different set of sources. That is not a glitch. It is how AI answers work, and new data shows just how quickly the sources behind them move.

SISTRIX tracked 82,619 prompts across Google AI Overviews, Google AI Mode, and ChatGPT search for 17 weeks, capturing more than 1.5 million snapshots in six countries.1 It measured how many cited sources were exchanged from one week to the next.

Figure 1Share of cited sources replaced each weekAverage weekly citation drift by platform.
Source: SISTRIX, AI Citation drift (May 2026), 82,619 prompts, 1.5M+ snapshots, December 2025 to April 2026.
View the data
PlatformWeekly drift
ChatGPT search74%
Google AI Mode56%
Google AI Overviews5%

In ChatGPT search, roughly three out of four cited sources are different from one week to the next. Google’s AI Mode replaces more than half. AI Overviews are the outlier, changing only about 5% of sources weekly, likely because they stay closer to Google’s regular search results.

See what weekly drift does over a few months

SISTRIX also found that 86% of prompts keep a stable core of sources that persists, while the periphery around it rotates. The simulator below models that: an answer with 10 citation slots, two of which (outlined) belong to a stable core. Pick a platform and press play to run 17 weeks at its average drift rate.

Figure 2Citation drift simulator10 citation slots, 17 weeks. Outlined slots are the stable core.
0Weeks elapsed
10 of 10Original sources still cited
0 of 10Sources replaced this week
Illustrative simulation. Average weekly drift rates from SISTRIX; the two-slot stable core is a simplification of its core and periphery finding. Blue = original source, gray = replacement.

Run ChatGPT and the original periphery is usually gone within two or three weeks, with only the core surviving. Run AI Overviews and most sources are still there months later. Same idea, very different maintenance requirements.

The site stays, the page rotates

If citations churn this much, why do some reports describe AI search as stable? Because they measure something different. BrightEdge looked at citations at the domain level across ChatGPT, Gemini, Google AI Mode, AI Overviews, and Perplexity in early February 2026. It found that 96.8% of cited domains saw zero change week over week, and 97.2% of brands saw no change in mentions.2

Put the two studies together and a clear picture emerges. Engines keep returning to the same trusted websites, but they swap which pages from those sites they cite. For a brand, that means being covered by the right publication once is not enough. The engine may still cite that publication next month, just a different article on it.

BrightEdge also found that when domains did move, 87% of the changes were declines. Visibility tends to erode quietly, rather than grow on its own.

Which kinds of sources stick around

Not every format decays at the same speed. SISTRIX measured how often different kinds of domains became part of an answer’s stable core in Google AI Mode.

Figure 3Core rate by source type, Google AI ModeHigher means the source type is more likely to become a stable, recurring citation.
Source: SISTRIX (May 2026). Separately, only 1.4% of cited news articles remained cited permanently.
View the data
Source type (AI Mode)Core rate
Video and YouTube24%
Big Tech platforms16%
Wikipedia12%
Marketplaces10%
Forums and UGC3%

Video was the stickiest format, with a 24% core rate. Forums and user-generated content were among the least sticky at 3%, and news was the most fleeting of all: only 1.4% of cited news articles stayed cited permanently. That makes sense, since news is time-bound by nature. It also means a single press hit can lift visibility briefly, then fade.

AI prefers fresher content than Google search does

Churn is one half of the story. The other is what replaces the old sources. Ahrefs analyzed nearly 17 million cited URLs across seven platforms and compared the age of the content each one cites.3

Figure 4Average age of cited content, in daysLower means fresher. The dashed line marks Google’s organic results (1,416 days).
Source: Ahrefs, “Do AI Assistants Prefer to Cite Fresh Content?” (July 2025), 16.975M cited URLs. ChatGPT figure is for in-line citations.
View the data
PlatformAvg. content age (days)
ChatGPT (citations)958
Copilot1,056
Gemini1,118
Perplexity1,166
Google AI Overviews1,432
Organic search results1,416

Overall, AI assistants cited content that was 25.7% fresher than what appears in organic results. ChatGPT was the freshest at 958 days, while Google AI Overviews matched the age of regular search. Muck Rack has seen the same tilt in journalism: more than half of the news articles AI cites were published within the past 12 months.4

What brands and agencies should do with this
  1. Plan for a steady cadence, not a launch spike. With most cited pages rotating within weeks, a continuous flow of new mentions keeps you in answers better than one big burst.
  2. Aim for the trusted domains, repeatedly. Engines keep returning to the same sites but swap the pages. Recurring coverage on the publications in your category beats a single feature.
  3. Mix in durable formats. Video and reference-style content become stable citations far more often than news. Pair timely coverage with assets built to last.
  4. Refresh your evergreen pages. AI favors fresher content, so updating key pages and earning new write-ups keeps you eligible.
  5. Measure trends, not snapshots. Track visibility over weeks and months. A single check tells you little when sources change this fast.

Frequently asked questions

How often do AI answers change their sources?

SISTRIX measured average weekly drift of 74% for ChatGPT search, 56% for Google AI Mode, and 5% for Google AI Overviews.

If sources change so much, why do some studies say AI search is stable?

They measure domains rather than pages. BrightEdge found 96.8% of cited domains did not change week over week, even though the specific pages cited from those domains rotate.

Does AI prefer newer content?

Yes, on average. Ahrefs found AI assistants cite content 25.7% fresher than organic search results, with ChatGPT citing the freshest content of the platforms studied.

Which content stays cited the longest?

In SISTRIX’s data, video had the highest core rate at 24%, while only 1.4% of cited news articles remained cited permanently.

Sources

  1. Johannes Beus, “AI Citation drift: How stable are sources in AI search results?” SISTRIX, May 2026. sistrix.com
  2. BrightEdge, “AI Search Citations: How Much Do They Really Change Week to Week?” February 6, 2026. brightedge.com
  3. Ryan Law and Xibeijia Guan, “Do AI Assistants Prefer to Cite Fresh Content? (17 Million Citations Analyzed),” Ahrefs, July 28, 2025. ahrefs.com
  4. Muck Rack, What Is AI Reading? report, May 2026. Press release
Written byMentionMethodMentionMethod helps brands understand and grow how they show up in AI answers. We publish data-driven research on AI visibility, citations, and brand mentions.
Learn more about MentionMethod

Convert ad spend to mentions in AI answers.

Built by ex-Facebook, LinkedIn, and IBM engineers
Influence answers across ChatGPTGeminiPerplexityClaudeCopilot

© 2026 MentionMethod. A product of OptalX, LLC. All rights reserved.