# How often do the sources cited in AI Overviews change?

> Google AI Overview citations can change between consecutive observations, and published studies commonly find source turnover within a few days.

Source: https://bestaioverviewtrackers.com/ai-overview-citation-volatility · Publisher: AI Overview Trackers · Updated: 2026-11-15

## TL;DR
- AI Overview citations can change between consecutive checks, and Semrush observed URL changes on 55.99% of desktop and 54.77% of mobile observation days.[8][1]
- The strongest practical default is daily collection with weekly reporting for important queries, because average consecutive URL persistence in Semrush’s study was about three to four days.[4]
- A missing citation is not necessarily permanent loss: 43% of removed desktop URLs and 41% of removed mobile URLs later reappeared.[8]
- Stable domains can hide changing pages and source sets, so trackers should report cited URLs, domains, source order and AIO presence separately.[4]
- No universal AIO refresh interval or validated minimum sample size is established; buyers should record the collection configuration and preserve repeated observations.[4]

## What the evidence shows

Google AI Overview citations can change between consecutive observations, and published studies commonly find source turnover within a few days. The strongest daily study found URL changes on 55.99% of desktop observation days and 54.77% of mobile observation days, while a separate Ahrefs study found 54.5% average URL overlap between consecutive observations.[1]

There is no published universal replacement rate or fixed Google refresh schedule. Google explains that AI Overviews may use query fan-out, different models and Search systems, while Search results can also vary by time, location, language, device and data-center context.[2][3] Finseo's own [study of query fan-out](https://www.finseo.ai/blog/query-fan-out) found that AI Mode rewrites one prompt into about nine synthetic searches. That measurement covers AI Mode, not AI Overviews.

The practical answer for buyers is **daily collection for important queries and weekly interpretation for reporting**. Daily capture preserves short-lived citations and returning sources; weekly summaries make the resulting data easier to interpret.[4]

### Why “volatility” needs a precise definition

Studies use several different measures:

| Measure | What it counts | What it does not prove |
|---|---|---|
| **AIO presence** | Whether an AI Overview appears for a query and configuration.[5][6] | Whether citations changed inside an Overview that appeared. |
| **Citation URL change** | Whether cited page URLs differ between observations.[7] | Whether the same publisher remains visible through another page. |
| **Source-order change** | Whether cited sources move, arrive or disappear in the ranked list.[4][7] | A literal percentage of URLs replaced unless the study says so. |
| **Answer-content change** | Differences in wording or semantic similarity between answers.[1] | That the answer’s meaning or recommendation changed. |
| **Domain-count stability** | Whether aggregate cited-domain counts change across periods.[5] | Whether individual queries return the same URLs. |

These measures should not be blended into one score. A stable leading domain can coexist with changing pages, source order and answer wording.[7]

## What studies actually measured

The table below focuses on studies that measured AI Overview presence, citations, answer changes or closely related repeatability. The samples and metrics are not interchangeable.

| Study | Publisher | Date | Sample | What was measured | Headline finding |
|---|---|---:|---|---|---|
| *Exploring URL Volatility in Google’s AI Overviews* | Semrush | 20 November 2024 | 1,500 US desktop keywords and 1,500 US mobile keywords; each produced an AIO on at least 20 of October’s 31 days | Daily cited-URL changes and persistence | URLs changed on 55.99% of desktop and 54.77% of mobile observation days; average consecutive persistence was 3.87 and 3.33 days.[8] |
| *AI Overviews Change Every 2 Days (But Never Change Their Mind)* | Ahrefs | 11 November 2025 | More than 43,000 keywords, each with at least 16 recorded AIOs over one month | Changes in wording, citations, entities and semantic similarity between consecutive observations | Content changed 70% of the time, average reported content persistence was 2.15 days, and citation URLs had 54.5% average overlap.[1] |
| *AI Search Citations: How Much Do They Really Change Week to Week?* | BrightEdge | 6 February 2026 | Thousands of prompts across nine industries and five engines, including AIO | Weekly changes in aggregate cited-domain counts | 96.8% of cited domains showed zero week-over-week change; this was not a per-query AIO URL-retention measure.[9] |
| *How Volatile Are AI Responses?* | Detailed | 23 September 2026 | More than 70,000 responses; over 1,300 prompts daily for 28 days; US, non-logged-in collection | Repeated answer, page and domain patterns across ChatGPT and AIO | About half of cited pages recurred the following day; the leading cited domain appeared on 89% of days.[10] |
| *How Stable Are AI Citations?* | Ahrefs | 29 October 2026 | 563 AIO questions and 4,830 day-one cited websites within a 20-day study | Continuous survival of the initial citation cohort | 50% of initial AIO citations were gone by day two, 18% lasted at least a week and 6.6% survived all 20 days.[11] |
| *AI Overviews & SERP Volatility* | Authoritas | Not published | 11,203 US desktop keywords observed on 23 August 2024, 17 October 2024 and 17 January 2025 | Position-weighted AIO source-list volatility | Published AIO volatility scores were 0.68 and 0.73; these are scores, not percentages of replaced citations.[12][8] |
| *Auditing Google’s AI Overviews and Featured Snippets* | Academic study | 17 November 2025 | 528 queries and variations across five US cities, plus a broader baby-care and pregnancy audit | Geographic differences in AIO appearance and content | The geographic AIO-content test found no significant difference, with p = 0.33; it did not measure URL retention over time.[13] |
| *Who Anchors AI Overviews in Health?* | Academic study | 6 September 2026 | 1,920 results across 12 countries and four languages | Geographic authority and repeated-response robustness | The authors reported that repeated responses did not vary meaningfully, but did not publish a numerical URL-overlap coefficient.[14] |

### Semrush provides the clearest daily URL evidence

Semrush’s study is the most direct evidence for day-to-day AIO citation turnover because it observed cited URLs daily for a month.[4] The study found that the average consecutive URL run lasted 3.87 days on desktop and 3.33 days on mobile.[8]

It also found that no sampled keyword retained exactly the same URL set throughout the observation period, and 96% experienced a domain change.[8] However, the study selected keywords that produced an AIO on at least 20 of 31 days, so it describes citation volatility among frequently appearing AIOs rather than all Google searches.[8]

Removed citations also returned: 43% of removed desktop URLs and 41% of removed mobile URLs reappeared later.[8] A single disappearance therefore should not be treated as proof of permanent visibility loss.

### Ahrefs separates wording change from meaning change

Ahrefs reported 70% content change between observations but a semantic similarity score of 0.95.[1] That means answer wording often changed while the underlying meaning remained highly similar; it does not mean that 70% of answers changed their factual conclusion.[1]

The Ahrefs observations were not necessarily daily, so the reported 2.15-day content persistence should not be presented as Google’s regeneration interval.[5] It is an observation from that study’s collection process, not a published Google refresh schedule.

### The 2026 studies show short-lived sources and persistent leaders

Ahrefs’ 20-day study followed sources cited on day one and measured continuous survival.[11] Half of the initial AIO citation cohort disappeared by day two, while only 6.6% remained continuously cited for all 20 days.[1] In German-language (DACH) AI search, [Finseo's own study of AI citation stability](https://www.finseo.ai/blog/ai-citation-stability) found that only 31.6% of URLs cited in Google AI Overviews were still cited the next day for the same prompt.

Continuous survival is stricter than eventual visibility. A source that disappears and later returns fails the continuous-survival test even though it regains citation visibility.[15][10]

Detailed found that about half of cited pages recurred the following day, while the leading cited domain appeared on 89% of days and occupied first position on 28% of days.[10] This indicates that a persistent publisher can remain visible while the individual pages and source order change.

## How should buyers compare AI visibility studies across AI Overviews and AI Mode?

Study results should be compared only after checking four details:

- **Product:** AI Overviews, AI Mode and other generative-search products are not interchangeable.[4]
- **Unit:** A cited URL, domain, answer, source position and citation count measure different things.[1][7]
- **Interval:** Same-day repeats, daily checks, weekly checks and irregular observations answer different questions.[16][10]
- **Denominator:** A percentage may refer to all queries, only queries with an AIO, only day-one citations or only domains that moved.[15]

For example, BrightEdge’s 96.8% week-over-week stability concerns aggregate cited-domain counts across several engines.[9][7] It cannot establish that individual AIO queries retained the same URLs.

Likewise, SE Ranking’s 9.2% exact-URL overlap and 14.7% domain overlap came from three same-day AI Mode tests, not from repeated AI Overview observations.[15] Recasting those results as an AIO citation replacement rate would change both the product and the research question.[4]

The academic paper *Don’t Measure Once* supports repeated measurement and recommends seven runs for brand visibility and eight for source coverage in its tested systems.[4] Because AI Overviews were excluded, those numbers should not be advertised as validated AIO sample thresholds.[4]

## How often should an AI Overview visibility tracker check Google AI Overviews?

**Daily tracking is the strongest default for commercially important queries.** The direct AIO evidence shows that cited URLs often persist for only several days, and that removed sources can return later.[15][8]

Weekly tracking can still be useful when the goal is a broad directional trend rather than a complete citation history. It should be described as a sparse sample, because weekly checks can miss appearances and disappearances between observation points; the reviewed studies do not quantify exactly how many events a weekly schedule misses.

### How to choose a tracking frequency

1. **Track daily when a query affects revenue, brand protection or a major content decision.** Daily collection captures short-lived citations and makes it possible to distinguish disappearance from later reappearance.[4]

2. **Use weekly reporting even when collection is daily.** A weekly report can summarize visibility, cited URLs, source changes and recurring domains without discarding the underlying daily observations.[4][5]

3. **Use weekly collection only for exploratory monitoring.** This reduces cost and may reveal broad direction, but it cannot represent a complete record of intermediate citation changes.[4]

4. **Increase observation detail during an investigation.** When a high-value URL disappears, repeat the check under the same configuration and preserve the answer text, cited URLs and timestamp before declaring a sustained loss.[4][17]

5. **Separate AIO presence from citation visibility.** Record “AIO absent,” “AIO present but target URL absent” and “target URL cited” as different outcomes.[4][5]

6. **Keep device, location and language consistent.** Google identifies time, location, language, device, recent searches and data-center context as factors that can affect Search results.[2]

7. **Use matched series for device or location comparisons.** Compare the same queries with explicitly configured devices and locations rather than mixing observations from different settings.[4]

8. **Choose enough keywords to represent the decision.** Published studies range from 563 AIO questions to more than 43,000 keywords, but no universal AIO minimum is established by this evidence.[4]

9. **Preserve raw evidence separately from summary metrics.** Historical visibility charts do not necessarily provide replayable answer text, HTML or screenshots.[17]

10. **Report uncertainty instead of overinterpreting a single change.** A missing citation in one observation establishes absence in that configured check, not permanent loss or zero future probability.[17]

### What should a tracker store?

A buyer evaluating AI Overview tracking should ask whether the product records:

- Query and query version
- Collection date and time
- Country, city or location
- Language
- Device type
- Login and personalization state
- Whether an AIO appeared
- Full answer text
- Every cited URL and cited domain
- Source order
- Historical reappearances
- Collection failures separately from “no AIO”
- Retention period for raw answers and URLs
- Export or API access

Collection frequency and reporting frequency are different product capabilities. Semrush, for example, describes Position Tracking updates as generally occurring within 24–48 hours, even though its AIO monitoring is described as daily.[4][16] A “daily” label therefore does not necessarily mean that a completed observation is immediately visible.

SE Ranking illustrates another distinction: its documentation specifies 30-day full-text retention and one-year presence-metric retention.[18][4] A buyer may therefore have a year of visibility metrics without a year of replayable answer text.[17]

## Why one snapshot can mislead


Why can a single AI Overview snapshot give the wrong impression?

A snapshot records one answer for one query, time and search configuration. It cannot show whether a cited URL appeared yesterday, disappeared temporarily, returned tomorrow or was replaced by another page from the same domain.[8]

A missing AIO is also different from an AIO that appears without a particular citation. Google says AIOs do not trigger for every search, and Search results can vary with context such as location, language, device and time.[2]

A reliable interpretation therefore needs repeated comparable observations, separate denominators and preserved answer evidence. “Target URL cited in one snapshot” is an observed state, not a stable visibility rate.[1]



## What should buyers look for in an AI Overview visibility tracking tool?

The evidence supports a layered tracking model:

| Tracking layer | Recommended use | Why it matters |
|---|---|---|
| **Daily raw capture** | Important queries and citation investigations | Captures short runs, disappearances and returns.[4] |
| **Weekly reporting** | Management decisions and trend summaries | Reduces noise while retaining the daily evidence underneath.[4] |
| **Monthly review** | Query-set quality, recurring domains and source strategy | Helps distinguish persistent authority from temporary URL turnover.[4] |
| **Same-configuration repeats** | Anomalies, high-value losses and research validation | Separates temporal changes from context differences.[4] |
| **Long-term retention** | Audits, content decisions and competitive analysis | Allows later review of answer text and citation history rather than only summary metrics.[17] |

The strongest synthesis is not that every AIO changes every day. It is that **source instability and stable authority can exist at the same time**.[4] A leading domain may recur frequently while its cited pages, neighboring sources and ranking positions change.

For most buyers, daily evidence capture with weekly interpretation is the most defensible balance. Weekly-only tracking is appropriate when directional monitoring is sufficient, but it should not be described as a complete record of citation behavior.

## FAQ
### How often do AI Overview citations change?
Published studies show that cited URLs can change between consecutive observations and often turn over within several days.[4] Semrush observed URL changes on 55.99% of desktop observation days and 54.77% of mobile observation days, while average consecutive URL persistence was 3.87 and 3.33 days.[8]

### Do AI Overviews change every day?
Not necessarily. Some studies observe substantial daily turnover, while academic audits found no significant geographic differences in particular health-related samples.[13] The evidence supports query- and context-dependent volatility, not a universal daily-change rule.

### Is daily AI Overview tracking better than weekly tracking?
Daily tracking is better when the buyer needs to detect short-lived citations, temporary losses or returning URLs.[4] Weekly reporting can still be appropriate if the underlying data was collected daily; weekly-only collection is a sparser trend sample.

### How many times should the same AI Overview query be checked?
No universal AIO repeat count is established in the reviewed evidence. A study recommending seven or eight runs excluded AI Overviews, so those figures should not be treated as validated AIO thresholds.[4]

### Can one missing citation prove that a page lost AI Overview visibility?
No. Semrush found that 43% of removed desktop URLs and 41% of removed mobile URLs later reappeared.[8] A missing citation proves only that the page was not cited in that particular configured observation.

### Why do AI Overview citations change between devices or locations?
Google says Search results can vary with time, location, language, device, recent searches and the responding data center.[2] Google also explains that AI Overviews may use query fan-out and different models or Search techniques, but it does not publish a universal AIO-specific geographic turnover rate.[3][15]

### How do Semrush and SE Ranking handle AI Overview tracking data?
Semrush describes its AIO monitoring as daily, but it says Position Tracking updates generally occur within 24 to 48 hours.[4][16] SE Ranking's documentation specifies 30-day full-text retention and one-year presence-metric retention.[18] A buyer may therefore have a year of visibility metrics without a year of replayable answer text.[17]

### Do AI Mode citations behave the same as AI Overview citations?
They should not be treated as interchangeable, because AI Overviews, AI Mode and other generative-search products are different products.[4] SE Ranking's 9.2% exact-URL overlap and 14.7% domain overlap came from three same-day AI Mode tests, not from repeated AI Overview observations.[15]

### What should an AI Overview tracking tool record?
A tracker should record AIO absent, AIO present but target URL absent, and target URL cited as separate outcomes.[4][15] It should also keep device, location and language consistent, and it should preserve raw answer evidence separately from summary metrics. Historical visibility charts do not necessarily provide replayable answer text, HTML or screenshots.[17]

## Sources
[1] AI Overviews Change Every 2 Days (But Never Change Their Mind) — https://ahrefs.com/blog/ai-overview-change
[2] Why your Google Search results differ from others - Google Search Help — https://support.google.com/websearch/answer/12412910?hl=en
[3] AI Features and Your Website | Google Search Central  |  Documentation  |  Google for Developers — https://developers.google.com/search/docs/appearance/ai-features
[4] Don't Measure Once: Measuring Visibility in AI Search (GEO) — https://arxiv.org/pdf/2604.07585
[5] Quantifying Uncertainty in AI Visibility A Statistical Framework for Generative Search Measurement — https://arxiv.org/html/2603.08924v2
[6] How to use AI Results Tracker: Rankings – Knowledge Base — https://help.seranking.com/hc/en-us/articles/16335399186460-How-to-use-AI-Results-Tracker-Rankings
[7] AI Results Tracker FAQ – Knowledge Base — https://help.seranking.com/hc/en-us/articles/29392630403356-AI-Results-Tracker-FAQ
[8] Exploring URL Volatility in Google‘s AI Overviews — https://www.semrush.com/blog/url-volatility-ai-overviews
[9] AI Search Citations: How Much Do They Really Change Week to Week? | BrightEdge — https://help.brightedge.com/resources/weekly-ai-search-insights/ai-search-citations-week-to-week-changes
[10] How Volatile Are AI Responses? (70K Answer, 28 Day Study) — https://detailed.com/ai-volatility
[11] How Stable Are AI Citations? 10 Things We Learned Tracking 37,500 Citations — https://ahrefs.com/blog/how-stable-are-ai-citations/
[12] Google AI Overviews more volatile than organic rankings: Report — https://searchengineland.com/google-ai-overviews-organic-rankings-volatile-452255
[13] Auditing Google’s AI Overviews and Featured Snippets: A Case Study on Baby Care and Pregnancy — https://arxiv.org/html/2511.12920v1
[14] Who Anchors AI Overviews in Health? Baidu, Google, and the Geography of Authority — https://arxiv.org/html/2609.06798v1
[15] AI Mode Research: Sources, Volatility, & Differences between AIO and Organic Search — https://seranking.com/blog/ai-mode-research
[16] How often does Semrush update its Position Tracking data? — https://www.semrush.com/kb/326-how-often-does-the-position-tracking-update-its-data
[17] Quantifying Uncertainty in AI Visibility A Statistical Framework for Generative Search Measurement — https://arxiv.org/html/2603.08924v3
[18] AI Results Tracker - SE Ranking API Documentation — https://seranking.com/api/project/ai-result-tracker
