Loudmink tracked 20 natural-language buyer questions across 5 AI search engines and 25 B2B SaaS brands, checking every week for a full quarter. Here is what we found: only about 9% of citations survived the entire quarter, roughly 1 in 10. The share that held steady within a single month fell too, from about 31% in April to 28% in May to 21% in June. A win you recorded in spring was, more often than not, gone by summer. This article explains what citation durability is, why AI citations decay, and what to do to keep yours from disappearing.
A one-time AI visibility audit captures one frame of a moving picture. When you check on a Tuesday and your brand shows up in ChatGPT, that tells you almost nothing about Wednesday. The rest of this article walks through the numbers, the engine-by-engine differences, and a practical fix for keeping citations that actually hold.
The Bottom Line
- Across the full quarter, only about 9% of citations survived every weekly check, roughly 1 in 10. The share that held steady inside a single month also dropped, from about 31% to 28% to 21%.
- Even the most-cited brands move. Salesforce, cited in roughly 93% of its category's questions, still swung between 3 and 10 citations from one week to the next. Amplitude, the analytics leader, ranged from 5 to 13.
- Only two question-to-brand pairings stayed rock solid all quarter: "Vercel vs Netlify" always returned Vercel, and "product analytics platform comparison" always returned Amplitude. Comparison-style questions produced the only durable citations we found.
What Citation Durability Means
Citation durability is the share of your AI citations that are still there when you check again later. Take every specific case where one AI search engine cited one brand for one question, then count how many of those exact citations were present in every single weekly check across the quarter. That share is durability, and ours was 9%. The other 91% appeared, disappeared, or flickered in and out.
This matters because most brands and most monitoring tools treat a citation as a thing you win once and keep. It is not. A citation is a position that has to be defended every week, because AI search engines rebuild their answers from scratch each time someone asks. The number tells you how exposed you are: at 9% durability, the typical citation is temporary by default.
What to do: Stop measuring AI visibility as a single snapshot. A one-time audit reports a number that is mostly noise, because it cannot tell a durable citation from one that vanishes next week. Track the same questions over time so you can see which of your citations actually hold and which are flickering.
AI Citations Decay Month Over Month
The downward trend is not just quarter-over-quarter, it shows up inside each month. The share of citations that held across every check within a single month fell steadily as the quarter went on. By June, only about 1 in 5 citations survived all of that month's checks, down from nearly 1 in 3 at the start.
| Month | Citations that held all month |
|---|---|
| April | ~31% |
| May | ~28% |
| June | ~21% |
The practical reading: the longer the window you measure, the fewer citations survive it, and the window is getting harsher over time. Part of this is freshness. AI search engines favor content updated in roughly the last 30 days, so a page that earned a citation in April slips out of the preferred window unless you touch it again. This is the same dynamic we covered in why publishing more content won't fix AI visibility: one-time content decays just as fast as it was published.
What to do: Refresh your key pages monthly to stay inside the 30-day window AI search engines prefer. Update the date, add a current data point, tighten the answer at the top. A page you published in spring and never touched is competing against pages updated last week.
Even the Category Leaders Don't Sit Still
Being the most-cited brand in your category does not make your citations stable. The leaders moved as much as anyone. PostHog, an analytics challenger, slid across June from 6 citations to 5 to 3 to 4 to 2. Mixpanel was cited in earlier checks and then dropped to zero in the final check of June. If the leaders swing this much, a smaller brand has no floor at all.
| Brand and question type | Behavior over the quarter |
|---|---|
| Vercel, on "Vercel vs Netlify" | Rock solid: cited every single check |
| Amplitude, on "product analytics platform comparison" | Rock solid: cited every single check |
| Salesforce, across its category | Volatile: swung between 3 and 10 citations |
| PostHog, across June | Declining: 6, 5, 3, 4, 2 |
| Mixpanel, across June | Collapsed: cited early, then zero |
The pattern is the divide between stable and volatile. The citations that never moved clustered in comparison and head-to-head questions, where the incumbent answer is hard to dislodge. Everything tied to broad category questions moved. This tracks with our earlier finding that AI search engines disagree on the top recommendation about half the time: when the engines cannot agree, no single brand's position is safe.
What to do: Own the durable question types. Comparison content and "alternative to X" pages produced the only citations that held all quarter, because they answer a specific head-to-head intent that AI search engines treat as editorial. Publish honest comparison content on your own domain that names competitors and includes pricing, then keep it current.
Which Engine Cited You Changes How Durable It Is
How durable your citation is depends partly on which AI search engine produced it. Over the quarter, Perplexity was the most stable, changing only about 12% of its cited brands from one check to the next. ChatGPT was the most volatile, changing roughly 32%. So a Perplexity citation is a more reliable asset than a ChatGPT one, even for the same brand and the same question.
This is why a single-engine view is misleading. If you only watch ChatGPT, you are watching the most volatile engine and concluding that AI search is chaos. If you only watch Perplexity, you are watching the calmest one and concluding it is settled. Neither is the full picture, and the engines move independently, often in different directions at the same time. We dig into that month-to-month divergence in our companion piece on why the AI search landscape keeps shifting.
What to do: Monitor each AI search engine separately rather than blending them into one score. A blended number hides the fact that your Perplexity citation is holding while your ChatGPT citation is flickering, which are two different problems with two different fixes.
Why AI Citations Don't Last
AI citations don't last because AI search engines rebuild every answer from scratch each time someone asks, and the sources they pull from keep changing underneath them. The same question can return different brands on Tuesday and Wednesday, a run-to-run variance we covered in why AI search results change every time you ask. On top of that short-term wobble sits the slower month-over-month drift: new pages get published, old pages fall out of the freshness window, and review threads and roundups get rewritten.
Most of the raw material is not even yours. Around 85% of AI citations come from third-party sites: Reddit threads, review platforms, listicles, and editorial coverage. When a roundup adds a competitor or a Reddit thread gets a new top comment, the answer shifts and your citation can evaporate, without you changing anything on your own site. You do not control most of the inputs, so you cannot expect the output to sit still.
What to do: Build presence across multiple third-party sources so one shift does not erase you. If your only citation comes from a single Reddit thread, you are one re-ranking away from invisible. Spread your footprint across reviews, community discussion, and editorial so the loss of any one source is survivable.
How to Earn Citations That Actually Hold
Durable AI visibility comes from continuous monitoring plus a few defensible content types, not from a one-time push. The 9% that survived our full quarter were concentrated in comparison answers and reinforced by fresh, multi-source presence. Everything else was temporary. Four habits separate citations that stick from citations that flicker:
- Track over time, not once. A single audit measures noise. Re-check the same questions on a regular cadence so you can tell a durable citation from a lucky one.
- Refresh monthly. Keep your key pages inside the roughly 30-day window AI search engines favor. Update dates, data, and the answer at the top of the page.
- Own comparison and "alternative to" content. These were the only question types that produced stable citations all quarter. Publish them on your own domain with competitor names and pricing.
- Build multi-source presence. With 85% of citations coming from third parties, spread your footprint so no single source's change wipes you out.
Loudmink is an AEO platform that re-checks AI search engines on 24-hour cycles and re-verifies after your content publishes, so you can see which citations hold and which decay. Plans from $99/mo.
Frequently Asked Questions
What is AI citation durability?
AI citation durability is the share of your AI citations that are still present when you check again later. In our quarter-long research, only about 9% of citations were present in every weekly check. The other 91% appeared and disappeared over time.
Why do my AI citations keep disappearing?
AI search engines rebuild every answer from scratch when someone asks, and most of what they cite comes from third-party sources that keep changing, around 85% of citations. When a review roundup, Reddit thread, or competitor page changes, your citation can vanish even though your own site did not change.
How often should I check my AI search visibility?
Check on a regular cadence rather than once, because a single snapshot cannot tell a durable citation from one that flickers. Monthly is a reasonable minimum for manual checks, and daily monitoring catches shifts as they happen. One-time audits measure noise.
What kind of content earns the most durable AI citations?
Comparison and "alternative to X" content earned the only citations that held all quarter in our research. These pages answer a specific head-to-head intent that AI search engines treat like editorial, which makes the incumbent answer hard to dislodge. Broad category visibility was far more volatile.
Does it matter which AI search engine cites me?
Yes. In our research, Perplexity changed only about 12% of its cited brands between checks, while ChatGPT changed roughly 32%. A Perplexity citation tends to be more durable than a ChatGPT one, so you should monitor each AI search engine separately instead of blending them.