How often do AI citations change?

In our 13-day test, 47.5% of cited sources changed from one day to the next, running the identical questions. Only 4.3% of domains stayed cited on 12 or more of the 13 days, and nearly half appeared on a single day before vanishing. Day-to-day churn is the norm, not the exception.

Why did Perplexity have the most stable citations?

It was the surprise of the experiment. Perplexity changed just 28.9% of its citations day to day, versus 60.7% for ChatGPT and 54.9% for Gemini. One likely reason is that Perplexity re-derives sources from a live search each time rather than a cached index, but our data shows the pattern, not the cause.

Does a one-time AI citation still count?

It counts, but weakly. Half the citations we tracked appeared on only one of 13 days, so a single sighting is often a fluke rather than a repeatable result. Being cited once tells you the engine can reach your page, not that it reliably will. Consistency over time is the stronger signal.

What is citation consistency and how do you measure it?

Citation consistency is the share of repeated AI answers, over a set window, in which your source gets cited. You measure it by running the same questions on a schedule and counting the percentage of runs that include your domain. A rolling week of daily checks separates durable visibility from one-day noise.

Do these citation-stability findings apply beyond sneakers?

The category was sneakers, but the mechanism is general: AI answers are generated fresh each time and lean on third-party sources that shift as indexes update. The specific stable domains change by topic, yet the pattern of high day-to-day churn and third-party dominance is likely to hold across categories.

How many times should I check whether AI cites my brand?

At least five to seven checks across a week, not one. Because roughly half of citations are one-day events, a single snapshot can easily mislead you in either direction. Several data points across a rolling window give you a consistency rate you can trust and compare over time.

Do AI Citations Change Over Time? A 13-Day GEO Experiment

Key Takeaways

Across 13 days, 47.5% of cited sources changed from one day to the next. A one-off citation is often noise, not a signal.
Only 4.3% of domains stayed cited on 12 or more of the 13 days. Half appeared on a single day, then disappeared.
Perplexity, which searches the web live, was the most stable engine at 28.9% churn, not the least.
Every source cited on 12+ days was a third-party site (RunRepeat, YouTube, Reddit). No brand’s own website made the stable list.

How did we measure AI citation stability?

We ran the same 5 buyer questions in one product category through ChatGPT, Gemini, and Perplexity once a day for 13 straight days, then recorded every source domain each engine cited each day. Instead of only counting citations, we compared the set of cited domains against the previous day to see how many changed. Citation stability is how consistently the same sources get cited when you ask the same question over time.

A citation is a link an AI engine uses as a source behind its answer, separate from a mention, which is just your brand name appearing in the text. Most GEO advice stops at “did I get cited?” We wanted the harder number: if you got cited today, are you still cited tomorrow? Running the identical prompts daily is the only way to separate a durable citation from a one-day coincidence (leadrescue.app analysis, 2026).

How much do AI citations actually change from day to day?

A lot. Across the 13 days, 47.5% of cited sources changed between one day and the next. Only 4.3% of all cited domains held their spot on 12 or more of the 13 days. Nearly half, 49.7%, appeared on exactly one day and then vanished. The average cited source lasted just 3.2 of the 13 days, and the median was 2.

That churn reframes what a citation is worth. If half of everything you observe is a one-day appearance, then seeing your page cited once tells you almost nothing about whether it will be there tomorrow. The engines are not slightly noisy at the edges. The instability sits right in the middle of the results, where most brands are fighting to appear (leadrescue.app analysis, 2026).

47.5% of the sources AI engines cited changed from one day to the next, running the identical questions across 13 days. leadrescue.app analysis, 2026

Which AI engine has the most stable citations?

Perplexity, by a wide margin, which was the biggest surprise in the data. Because Perplexity runs a live web search on every query, we expected it to be the most volatile. Instead it changed the fewest citations day to day, at 28.9%. ChatGPT churned the most at 60.7%, and Gemini sat in the middle at 54.9%. The engine that constantly re-checks the web gave the steadiest answers.

Engine	Citations changed day to day	Read
Perplexity	28.9%	Most stable — live web search on every query
Gemini	54.9%	Moderate churn
ChatGPT	60.7%	Most volatile — over half changed daily

The live-search engine was the most stable, not the least. Source: leadrescue.app analysis, 2026.

Why the live-search engine is the most stable is not something our data can prove. One plausible explanation is that Perplexity re-derives its sources from a fresh search each time, while engines leaning more on a cached index shift as that index updates. We treat that as a hypothesis, not a fact. The observed pattern held across all 13 days regardless of the cause.

What kind of sources stay cited over time?

Third-party sources, without exception. Every domain cited on 12 or more of the 13 days was a reviewer, publisher, or community: RunRepeat, YouTube, Who What Wear, Reddit, Runner’s World, and Outdoor Gear Lab all held 13 of 13 days, with Tom’s Guide at 12. Not one brand’s own website made the stable list. The sources that flickered in and out were mostly smaller sites competing for the same query.

The sources that never move are all third-party. Brand-owned sites were less consistent, and nike.com barely appeared. Source: leadrescue.app analysis, 2026.

This lines up with what makes a source easy for an engine to trust. A Princeton-led study found that clear structure, statistics, and citations lifted a source’s visibility in AI answers by up to 40% (Aggarwal et al., Princeton, 2023). Established review sites tend to have exactly that profile, which may be why they hold their citations while newer, thinner pages flicker. It is also the same pattern we found when we looked at what sources AI engines cite when recommending brands.

Why does a single AI citation not mean much?

Because a single citation is a coin flip, not a trend. In our data, 49.7% of cited sources appeared on exactly one of the 13 days. If you check once and see your page cited, there is roughly a coin-flip chance it was a one-day event that will not repeat. A one-day citation is a single observation of a noisy signal, and one observation cannot tell you whether the signal is real.

This matters more as buyers move real research into AI. Pew Research found that 34% of US adults had used ChatGPT by mid-2025, roughly double two years earlier, with product research among the top uses (Pew Research Center, 2025). If a meaningful slice of buyers is asking AI for recommendations, a citation that only shows up one day in ten is not really visibility. It is a lottery ticket.

How should you measure AI citations instead?

Measure consistency over time, not a single snapshot. The useful question is not “was I cited?” but “what share of the time am I cited when someone asks?” In practice that means running your key questions on a schedule and tracking how often your domain appears across the whole window, so a one-day fluke never gets mistaken for durable visibility.

Use a rolling window, not one check. Half of what we saw was a one-day appearance, so you need several data points across a week before trusting any result.
Track a consistency rate. Count the percentage of daily runs in which your source is cited, not whether it was cited once.
Expect churn, and treat stability as the goal. One citation that holds across many runs is worth more than ten that each show up once.

This is the metric Lead Rescue calls Domain Coverage: the percentage of repeated AI answers where a domain gets cited over time, measured across ChatGPT, Gemini, and Perplexity. For the exact formula, see the metrics glossary, and for the full tracking method, read how to track AI search citations.

Lead Rescue runs your prompts daily and reports your citation consistency per engine, so you can tell a durable citation from a one-day fluke. See how consistently AI cites you →

Do AI Citations Change from Day to Day?