Article's Content
Article's Content
Welcome back to What Matters This Week.
Last week, we covered SE Ranking’s analysis showing Reddit’s #1 rankings jumped 54% after Google’s May Core Update. This week, we’re going one layer deeper. Past the SERP, past the citation, into the question almost nobody is measuring: how long does an AI citation actually last?
The 11-Day Lease: New Data on How Quickly AI Citations Disappear
Here’s the TL;DR
- A new Writesonic study tracked 23 million cited sources across AI platforms from April to June 2026. 44% of cited pages appeared exactly once before disappearing from the answer set entirely.
- The typical AI citation lifespan is 11 to 15 days. Perplexity holds sources about twice as long as ChatGPT.
- Citation share is only part of the story. Citation durability is closer to what actually matters.
What’s Happening
At a recent masterclass with Writesonic’s CEO Samanyou Garg, Ross and Sam shared seven studies pulled from Writesonic’s index of 2 billion AI conversations. One stood out as the first of its kind we’ve seen published.
In it, the Writesonic team tracked 23 million cited sources across ChatGPT, Gemini, Perplexity, and Google AI Overviews over a two-month window, from April to June. They were trying to answer a question most GEO discussions miss: when a brand gets cited, how long does that citation actually hold?
According to Writesonic’s data, 44% of cited pages disappeared from AI answers after a single appearance. They didn’t come back the next day. They didn’t turn up in a related query. They were there, and then they were gone.
For the citations that stuck around, the median lifespan was 11 to 15 days. Perplexity retained sources roughly twice as long as ChatGPT, making it the most “loyal” of the major engines. ChatGPT was the most volatile.
The study lines up with two other findings from the same session that reinforce why this matters:
- After the GPT-5.5 model update in May, ChatGPT’s share of citations going to third-party sites jumped to 96%, with the average vendor site capturing just 4.3%.
- In 90% of cases, when a new brand enters an AI answer, another brand gets dropped. AI answers operate with fixed seats, not expanding lists.
Read together, the picture is clearer than most GEO conversations have made it. Getting cited once is not the same as being in the answer. And being in the answer this week is not the same as being in the answer next month.
Why It Matters
Citation share is fluid. Citation durability is the asset.
Most GEO tools report citation share as a point-in-time number: out of X queries this week, you appeared in Y percent of answers. That metric has the same problem as measuring a stock portfolio on a single trading day. The number tells you where you stood. It doesn’t tell you whether you’re holding ground.
If 44% of citations appear once and then disappear, what looks like a healthy-looking citation share could actually be a rare spike amidst constant citation churn. You might be cycling in and out of answers at the same rate competitors are. Or you might be in the small group of sources the models keep returning to. The point-in-time number doesn’t distinguish between them.
The two-week window changes how often you need to refresh.
If the average citation half-life is around two weeks, then the content you published last quarter is no longer guaranteed to be doing work for you: freshness makes a big difference. AI visibility is closer to a paid media motion than an SEO one. It requires ongoing investment, not a one-time push.
This also explains a pattern Ross has been calling out for the last year. Brands celebrate when they show up in ChatGPT. The next week, they’re gone, and they don’t know why. The data now suggests the answer is straightforward. They were never anchored in the first place. They were in the rotation, and the rotation moved.
The replacement game makes durability harder than it sounds.
The 11–15 day lifespan isn’t the whole story. Layered on top is the structural reality that AI answers have a fixed number of seats. When a new brand enters an answer, another brand exits. There’s no growing pie. There’s a fixed set of slots that brands trade among each other.
To stay cited, you have to do two things at once: keep the model trusting your existing content enough to retrieve it again, and prevent a competitor’s newer or better signal from displacing you. Brands tracking only citation appearance are measuring the first half. The second half is invisible to them.
What to Do About It
Audit citation durability, not just citation frequency. Pull your top 20 priority prompts and run them weekly for 30 days. Track which sources from your domain appear once and never return, which appear intermittently, and which hold across multiple runs. That third group is your actual GEO asset base.
Figure out why your durable content holds. Across the pages that get cited and stay cited, look for the common signals. In our experience, the durable citations almost always carry one or more of: original data, named-expert commentary, deep specificity (use cases, industries, company sizes), or strong third-party validation. Once you know what’s making it stick, you have a creative brief for what to do more of.
Build the off-site layer that backstops your durability. Per the same Writesonic data, 96% of ChatGPT’s citations now point to third-party sites. If your only durability play is on your own domain, you’re optimizing 4% of the surface area. The brands holding their citations longest tend to also have a presence on YouTube, Reddit, G2, and industry publications. When the model rotates sources, those off-site signals keep the brand in the answer even when your owned page drops out.
Treat GEO measurement like portfolio management. Citation share is one input. Citation durability, citation share of voice against named competitors, and the gap between citations and mentions (Writesonic found that 52% of brands cited by Perplexity aren’t actually mentioned in the answer) are all separate signals. Each tells you something different about whether your AI visibility work is compounding or evaporating.
🎙 New Masterclass Recording: What Still Works in Content, Search & Distribution
The data above came from a live session Ross ran with Samanyou Garg, Founder and CEO of Writesonic. The full hour covers seven Writesonic studies on AI citation behavior, the fundamentals Ross thinks marketers have forgotten, and a working framework for citation durability.
Other things worth queuing up the recording for:
- Why YouTube is the most underestimated channel for AI visibility right now
- The three Reddit accounts every B2B brand should run, and how to set them up
- How to reverse-engineer a successful Reddit post for your own category
- The “take a competitor’s seat” mental model for AI answer optimization
Watch the full session now on YouTube.
Go Deeper on GEO Measurement
→ GEO Metrics: How to Measure Visibility, Trust, and Brand Presence in AI Search. Our full framework for tracking the metrics that matter in a zero-click world. Covers the three-pillar model of visibility, citations, and sentiment, and includes the questions to ask before you trust any single GEO number. The piece to read alongside the Writesonic data if you’re rebuilding your reporting. [Foundation Labs]
→ How Content Freshness Affects AI Citations in 2026. Writesonic’s Rohit Mishra goes deep on the mechanics behind the durability problem: which freshness signals LLMs actually parse, the difference between cosmetic and substantive updates, and a tiered refresh cadence by content type. The practical companion to this week’s data if you want to start pulling the durability lever on your own content. [Writesonic]
→ GPT-5.3 Instant Is Citing Fewer Sources. If You’re Not One of Them, You’re Invisible. Our earlier breakdown of how OpenAI’s model update compressed citation counts in a single release. Pair it with this week’s durability data and the picture is bleak for brands counting on volume: fewer citations issued, shorter lifespans on the ones that get issued. [Foundation Labs]
That’s it for this week.
If something landed, tell us. If something felt off, tell us that too. Reply to this email or DM me on LinkedIn.
Have a great weekend,
Ethan Crump
