AI CITATIONS HAVE AN 11-DAY HALF-LIFE: AEO FOR EXECUTIVE SEARCH FIRMS

By

Most search firms treat a well-received article or a strong practice page like a placement that stuck. It got written, the partners signed off, and it’ll keep representing the firm until somebody gets around to replacing it.

That isn’t a silly way to think. For most of the web’s history, a decent page could hold its ground for years, and the numbers further down bear that out: the average page in Google’s organic results is almost four years old.

The research published so far on AI answer engines describes something else. A citation in an AI answer isn’t something your firm owns. It’s a position, held against an enormous and shifting pool of other sources, and in most cases it’s held only briefly.

Picture a radio station’s playlist rather than a library shelf. A new track goes into heavy rotation, gets played constantly for a short stretch, then drops back as other tracks take its slot. It can come back later. The station never promises it a permanent place. A book on a library shelf stays exactly where it was put until someone pulls it. In my experience, most firms manage their content like the book. The citation data describes the playlist.

When did you last check whether an answer engine still names the piece you were proudest of last spring?

It’s an odd thing to look up. Nobody mentioned it could change.

This piece covers what the decay data shows (and what it doesn’t), then what that changes about which pages your firm keeps working on. It’s about time.

What the decay data shows

Two of the larger vendor studies have measured how long a citation lasts. An academic study points the same way.

Profound tracked 883,000 pages across seven AI engines (ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Gemini, and Microsoft Copilot) between September 2025 and September 2026. It measured each page’s share of citations within its topic category, then counted the days from that page’s peak until its share fell to half and stayed there. The headline finding: “The median page loses half its citation share within 11 days of peaking.”

Eleven days. Not eleven months.

Seventy-eight per cent of cited pages fell to half their peak share within two weeks. The other 22%, which Profound calls sustained pages, lasted longer, with a median half-life of 27 to 29 days. Measured from a page’s first citation rather than its peak, the median time to half-peak was 42 days.

Half-life, since everything here hangs on it, is the time a page’s share takes to fall to half its highest point. If your page accounted for 2% of the citations in its topic at its peak and behaved like the median page, it’d be down to 1% within 11 days.

Scrunch, working with Stacker, came at the question differently. It analyzed about 3.5 million citation events across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode between September 2025 and March 2026, grouped sources by the week they were cited, and tracked how many were still being cited in later weeks. “For the average source, citation activity drops by half in approximately 4.5 weeks.” ChatGPT turned sources over fastest, at about 3.4 weeks. Perplexity held on longest, at about 5.8 weeks.

The two numbers don’t match, and there’s no reason they would. Profound measures one page’s slide from its own peak. Scrunch measures how fast a weekly group of sources thins out. What they agree on is the unit. Citation positions are measured in days and weeks, not years.

Both of those companies sell tools for tracking AI citations, which is why the third study matters. Mahe Chen and Nick Koudas of the University of Toronto, in a paper posted to arXiv in October 2026, collected answers from ChatGPT, Gemini, and Perplexity every day for 90 days, mostly to questions written for one US fibre internet provider. They measured how consistently the same pages within a website got cited from one two-week window to the next. For two windows ten weeks apart, the share of that variation they had in common was 14% for ChatGPT, 29% for Gemini, and 0% for Perplexity.

That study measures something different from a half-life. It looks at which of one site’s pages an engine prefers, not at how long a single page’s share lasts. But it lands somewhere compatible, and it does so from a separate dataset, gathered by university researchers on their own monitoring platform. The paper does thank ktau.ai for help with the data, and ktau.ai is a Toronto company that sells AI-search optimization, so it isn’t entirely clear of the industry either. Which of your pages an engine favoured ten weeks ago tells you very little about which it favours today.

Profound and the Toronto paper measure pages, and that matters for a firm. BrightEdge, which also sells AI-search tracking, looked at whole domains in a single week of February 2026 and found 96.8% of cited domains saw zero change week over week. Your firm’s name can hold steady while the individual pages behind it turn over. The pages are the part that moves, and they’re what this piece is about.

Why citations fade

Profound puts the short half-life down to “the breadth of the source pool.” An answer engine draws on a far wider set of sources than a page of ten search results, so the odds of any one page being cited again are low. Most short-lived pages, in Profound’s words, “appear in a single burst of citations and are never cited at that rate again.”

Back to the station. One with a few hundred tracks replays each of them often. One drawing on tens of thousands replays any single track rarely, however good it is. The track’s quality isn’t the main variable. The size of the library is.

That’s a rough thing to read if you’ve put real thought into a piece. It doesn’t say the piece was weak.

What the studies don’t establish is why a particular page fades, whether because it aged, because a competitor published something newer, or because it fell out of date. They do establish that citation share, once gained, usually slips away quickly.

Old pages still get cited

Read the decay figures on their own and you’d conclude anything older than a few weeks is dead weight. Other datasets rule that out.

Ahrefs, in a study published in July 2025, analyzed about 17 million citations. Across ChatGPT, Perplexity, Copilot, and Gemini, the average cited page had been published about 2.9 years earlier. That’s younger than the average page in Google’s organic results (about 3.9 years), but it’s a long way from new.

Search Atlas looked at 150,000 citations from OpenAI, Gemini, and Perplexity in late 2025. With web search switched on, the ages of roughly 10,000 dated URLs were “strongly right-skewed” toward recent pages, with a long tail of sources that were “several years old.”

None of this contradicts the half-life studies. Those describe how long a burst of citations lasts once it peaks. These describe how old the cited pages are. My take here is that an older page can be cited, fade, and come back in a new burst, the way an old track returns to rotation. What it doesn’t do, on the evidence, is hold a steady share of answers on its own.

Pages tied to a date fade first

Profound sorted pages into four content families. At 56 days after peak, the share of sustained pages still holding at least half their peak citation share was 11.2% for pages that help a reader choose, 10.2% for pages that explain, 9.8% for pages that sell, and 7.0% for “Report” pages: news, announcements, and research tied to a date or event. Report pages came last on all seven engines. Profound’s summary: “a page tied to a date fades before one that is not.”

Two caveats, and they matter. The gap is modest, and every family has a low survival rate at 56 days. The data supports a ranking. It doesn’t support the idea that any kind of page holds its position indefinitely.

Pages cited in more places hold on longer

Profound also found that breadth across engines went with persistence. Of pages cited by five or more engines, 6.8% were still above half their peak at day 56. For pages cited by a single engine, it was 1.2%.

Stacker, which worked with Scrunch on that analysis, reports that sources in its own publisher network kept their citations about twice as long as other domains. Read that one carefully. It’s about Stacker’s network, Stacker sells distribution through that network, and the author says plainly that causation can’t be claimed. It’s an association, not a general rule.

A hypothetical example

The following example is invented to illustrate the argument. It doesn’t describe a real firm or a measured result.

Say a boutique firm published two pieces in 2024. One announced a new partner joining to lead its industrial practice. The other was a substantial article on how boards should prepare for CEO succession.

The announcement is a Report page in Profound’s terms. It answered a question that mattered for a few weeks: who joined the firm. If it got cited at all, it’d be cited briefly, and then it’d fade. That’s the page doing exactly what it was built to do. Nothing to fix. You can let announcements fade on purpose, the way a station retires a news bulletin once the news has passed.

The succession article is a different animal. Boards keep asking that question, and the page sits in the families that hold up somewhat better. If you want your firm named when a board chair asks an answer engine how to prepare for succession, that’s a page worth keeping in contention.

Which means real revision when your view or the underlying data changes. Not a new date on an unchanged page, which is the content version of repainting the front door and calling it a renovation.

What this changes for your firm

No source cited here covers executive search. What follows is my reading of data from other categories, applied to yours.

The first change is deciding which pages you maintain. Most firm websites hold both kinds of content. A small number of pages answer the recurring questions a hiring client would put to an answer engine, such as what a CFO search in a given sector involves, or how your firm approaches board succession. Those are the pages worth keeping in contention, through regular, substantive updates and through presence beyond your own site. Announcements and dated commentary can be allowed to fade. Keeping the few pages that answer standing buyer questions accurate and current is the one part of that process you control. The data suggests it’s worth controlling.

The second change is timing. A content review held once a year runs on a clock measured in months. The decay data runs on days and weeks. A page can win and lose a citation position several times between annual reviews without anyone at your firm seeing it happen.

That’s an argument for treating content as an ongoing engagement rather than a one-off project, and for keeping your content library organized so every page that matters has an owner and a review schedule.

Executive Search Marketing Best Practice Test

Here’s a test you can run this week. List the pages on your site that answer a question a board will still be asking next year. Then find the last time anyone changed something substantive on each one. If the answer is “when the site launched,” that page is the library book.


I write high-performance marketing copy for executive search firms. You shouldn’t rely on generic, entry-level marketing content. You speak a complex language of organizational design, corporate life cycles, and high-stakes transitions. You deserve copy that showcases your track record and expertise. I can help.

Send me a sample of your current outreach assets, and I’ll give you my professional assessment.


Executive Writers Room specializes in writing high-performance executive recruitment copy that consistently places top leaders

Executive Writers Room specializes in writing high-performance executive recruitment copy that consistently places top leaders.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *