Why 70% of AI Citations Vanish in 6 Months

What is citation drift?

Citation drift is the way AI engines keep changing which sources they cite for the same question over time. The page an engine quotes today may be gone from the answer next month. It means an AI visibility win is rarely permanent. Your brand can appear in ChatGPT one week and vanish the next.

How fast do AI citations change?

Faster than most teams expect. Profound's analysis of 240 million ChatGPT citations, reported via Machine Relations, found 40 to 60% of cited domains change month to month and 70 to 90% over six months. SISTRIX, studying tens of thousands of prompts, found ChatGPT can replace as much as 74% of sources weekly. Treat both as directional.

Why are citations so volatile?

Because AI engines re-retrieve sources constantly and weigh freshness, relevance, and trust differently each time. Small changes in the model, the index, or competing content can swap who gets cited. The system is not built to be stable. It is built to answer well right now, which means the cited set keeps shifting underneath you.

When should you measure AI visibility?

Repeatedly, not once. Because citations drift week to week, a single snapshot tells you little. Measure on a regular schedule, monthly at least, so you track a trend rather than a moment. The goal is to see your direction over time, since any one reading can swing sharply for reasons outside your control.

Where do the most stable citations come from?

From high-authority, frequently referenced sources. The 5W Citation Source Audit found Wikipedia at about 13.15% and Reddit at about 11.97% of US ChatGPT citations, together over 25%. Established, widely trusted destinations drift less than a single blog post. Being referenced across these durable sources is steadier than relying on one page you control.

Which platforms drift the most?

Evidence points to ChatGPT showing high turnover, with SISTRIX observing up to 74% weekly source replacement. Different engines behave differently, and the research here is still maturing, so treat platform comparisons as directional. The safe assumption is that every major AI engine drifts enough that no single citation should be considered locked in.

Should you chase citations or traffic?

Track both, but do not over-index on either alone. Citations show whether AI engines trust you; traffic and conversions show whether that attention pays off. Because citations drift, durable business value comes from broad visibility plus a site that converts the visitors you do get. Watch citations as a signal, not as the final scoreboard.

Will earned media protect your visibility?

It helps more than a single owned page. When you are referenced across many credible third-party sources, you are less exposed to any one citation dropping. As 5W's Ronn Torossian put it, the old PR tier hierarchy no longer reflects how influence works. Broad, genuine coverage is a sturdier foundation than one piece of content you own.

Can you track drift per platform?

Yes, with the right tools. Platforms like Profound, Semrush, and others now monitor citations across ChatGPT, Perplexity, and Google AI surfaces over time. Checking each engine separately on a schedule shows you where you are gaining or losing ground. Per-platform tracking turns drift from an invisible problem into something you can actually watch and respond to.

How often should you re-measure?

At least monthly, more often if AI visibility is core to your growth. Given that a large share of cited domains change within weeks, a quarterly check misses too much. A monthly cadence balances effort against signal, letting you spot a real decline early rather than discovering months later that your citations quietly disappeared.

Building durable AI visibility? Pair this with my piece on why ranking #1 no longer wins citations, the Reddit AI citation index, and the Conductor 2026 AEO benchmarks. Let's chat.