AI Citations Fade in Days. Breadth Is What Lasts

You get cited in an AI answer on a Monday. You check again two weeks later and the citation is gone. Nothing on your site changed, and nobody sent a notice.

That is the normal pattern, not a glitch, if one vendor's data is representative. It also changes what a sensible small-business effort looks like.

What the study measured

Profound looked at daily citation data for 883,000 pages, which produced about 1.19 million page-and-engine pairs. For each pair it found the peak share of citations, then counted how long until the share dropped to half.

The headline: the median page hits half its peak within 11 days. From the first citation to that halfway point, the median is 42 days.

The more useful split is between two kinds of pages:

• Burst pages (78%). Half-life of 14 days or less. One spike of citations, then gone.
• Sustained pages (22%). A median half-life of roughly 27 to 29 days depending on the engine. Slower fade, same direction.

Even the winners fade. The difference is how fast, which is the core of Answer Engine Optimization as an ongoing job rather than a one-time setup.

One caution before building a strategy on this: Profound sells citation-tracking software, the metric is its own, and I could not verify the study independently. Read it as a strong signal about shape, not a law about exact days.

Why one engine is a thin bet

The part that matters for a small business is the breadth finding. Profound reports the share of pages still above half their peak at day 56, grouped by how many engines cited them:

• Cited by 1 engine: 1.2%
• Cited by 2 engines: 2.0%
• Cited by 3 to 4 engines: 3.5%
• Cited by 5 or more engines: 6.8%

All of those numbers are small. Still, the pattern is steady: each additional engine that cites a page goes with a longer life.

The study also found the engines mostly don't move together. Cross-engine correlation in citation duration was 0.03 to 0.09 for most pairs, which the authors read as each engine deciding on its own. If that holds, getting cited by ChatGPT tells you very little about whether Perplexity or Google's AI Mode will follow.

Correlation isn't cause. A page cited by five engines is probably a better page to begin with, so breadth may be a symptom of quality rather than a lever. Don't read this as "go get listed everywhere." Read it as "pages that earn citations in several places seem to be the sturdy ones, so build the kind of page that would."

What kind of page lasts longer

Profound sorted pages by intent. Reporting pages (news, announcements, research) faded fastest, with a median half-life of 27 days and 7.0% still above half-peak at day 56. Pages that help someone choose, explain something, or sell something held at 9.8% to 11.2%, which is roughly 1.4 to 1.6 times the rate of reporting pages.

For a local service business, that points away from news-style posts and toward the pages you already need: what you do, who it's for, how it works, what it costs to start, how to choose between options. Those are the pages that answer the question a stranger asks right before hiring someone.

Our earlier post on why AI search forgets your best page after a few months covers the refresh side of this: AirOps found most commercial-query citations come from pages updated within a year. The two datasets measure different things and shouldn't be merged into one number. Together they say the same practical thing: a page you wrote once and left alone is a page with a short shelf life in AI answers.

Spreading the bet without spamming it

If citations are short-lived and engine-specific, the goal is to keep your business eligible over time, not to win one moment. A few moves that fit a one-person team:

1. Keep the core pages current. Services, pricing approach, hours, service area. Put a real "updated" date on pages when something genuinely changed.
2. Answer the question in the first paragraph. Engines lift short, direct answers. Our notes on question-style headings cover the mechanics.
3. Get mentioned somewhere other than your own site. Brand mentions on other sites give an engine a second reason to trust you, which is the closest thing to breadth you can influence.
4. Check a few times a month, not daily. If an 11-day half-life is typical, a daily check will mostly show noise. Look for trends over weeks.

Four ways to spend effort on this and get nothing

• Chasing every new engine. There are seven in this study alone. You can't tailor a page to each, and the study doesn't suggest you should.
• Publishing news-style posts to "stay fresh." Reporting pages faded fastest in this data. A post about an industry announcement is a poor substitute for updating the page that sells your service.
• Treating one citation as a result. A single appearance that disappears in two weeks is the median outcome. Judge by whether you keep showing up across a month or two.
• Paying for a tool before you have pages worth tracking. Citation-decay dashboards are real products, but a tracker on thin pages only measures the thinness.

FAQ

How long does an AI citation last?

In Profound's data, the median page loses half its peak citation share within 11 days, and 78% of pages do so within 14 days. A minority of sustained pages fade over roughly four weeks. These are one vendor's figures across seven engines, so treat them as a rough shape.

Is a lost AI citation a sign something is wrong with my site?

Not by itself. Short citation lives are the norm in this study, so a citation disappearing doesn't signal a penalty. A change in your own rankings, indexing, or traffic would be a separate thing to check.

Does being cited by more AI engines help?

Pages cited by more engines were likelier to still hold half their peak share at day 56 (6.8% for five or more engines versus 1.2% for one). That's a correlation. Better pages likely earn more citations everywhere, so breadth may reflect quality rather than cause it.

Which pages hold their citations longest?

In the same study, pages that help a reader choose, explain, or buy outlasted news and announcement-style pages. For a small business, that means service, process and comparison pages over timely commentary.

Do I need a tool to track citation decay?

Not to start. A few manual checks a month on the questions your customers ask will show the trend. A paid tracker makes sense once you have pages worth watching and the budget to act on what it shows.

Should I update a page every week to stay cited?

No. The data points to keeping core pages accurate and current, not to constant rewrites. Change a page when something real changed, and record that date honestly.

If you'd rather someone keep those core pages current and watch the trend for you, that's what we do on Get found on Google and AI.