6 October 2026 · SEO Dispatch
One AI citation is not a lasting win
Profound's September 2026 study of AI citation half life, read carefully: what citation share means, what the numbers show, and why one screenshot of a citation is not a strategy.
The short answer
Probably not. A single citation in ChatGPT, an AI Overview or another AI answer is worth noticing, but recent research suggests it is often a moment rather than a position you hold.
The research is a study published on 30 September 2026 by Profound, a company that sells AI search tracking. Its title says it plainly: "The half-life of an AI citation is 11 days". Below is what it measured, what it found, and how far you can take it.
What the study measured
Profound tracked how long citations lasted for 883,000 pages across seven AI answer engines: ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Google Gemini and Microsoft Copilot. The window ran from 8 September 2025 to 8 September 2026. Pages came from a fixed 1 in 128 sample of cited URLs.
Each page was measured separately on each engine that cited it. After Profound's qualifying rules, that gave 1,188,816 page and engine pairings.
Two terms do a lot of work here:
- Citation share is how often a page was cited, compared with all the pages cited in its category on that engine, using the prompts Profound runs. It is not clicks, visits or sales.
- Half life is the number of days after a page's peak until its citation share falls to half of that peak and stays there for 14 days in a row.
Profound used a Kaplan Meier estimate, a standard method from survival analysis. Its useful feature is that pages still being cited when the study window closed are kept in the count as "still holding", rather than dropped. Statisticians call that censoring.
What it found
The headline figures, as reported on Profound's page:
- The median page lost half its citation share within 11 days of peaking.
- Of the 1.11 million pairings observed for at least two weeks past their peak, 78% had a half life of 14 days or less.
- Breadth went with staying power. Among pages cited by just one engine (679,077 in Profound's table), 1.2% were still above half their peak share at day 56. Among pages cited by five or more engines (43,770), 6.8% were. Profound describes that as 5.5 times as likely.
- Engines did not move together. How long a page lasted on one engine said very little about how long it lasted on another. Profound reports correlations of 0.03 to 0.09, apart from Google's three surfaces, which correlated with each other a little more.
One method detail is worth knowing. Profound smooths daily citation share with a 14 day rolling average. Its own write up notes that this makes two weeks the longest half life a page cited only once can have. So the 78% figure partly reflects how often pages were cited in one short burst, and partly reflects how the measure is built.
What it does not tell you
This is a large study with a clearly published method, which is more than most AI search claims offer. It still has limits you should keep in view:
- It is observational. It describes what happened to citations. It does not test what causes a citation to last or fade.
- It measures citation share, not traffic. A page can lose share and still get visits, or hold share and get few.
- The prompts are Profound's. Their prompt set and categories are not the questions your customers ask.
- It comes from a vendor. Profound sells a product that tracks citation decay, and the post ends with a demo offer. That does not make the data wrong. It is context.
- Medians are not targets. An 11 day median across hundreds of thousands of mixed pages is not a benchmark for your site, and beating it would not mean much on its own.
Profound's own conclusion is that content now needs to arrive on a regular cadence to stay eligible. That is their interpretation. The study did not test publishing schedules, so it does not prove that posting more often stops citations fading.
What to do with this on your own site
Treat an AI citation as a data point with a date on it, not a trophy.
- Write it down properly. Note the engine, the question you asked, the date and the page cited. A screenshot without that context is hard to learn from.
- Do not rebuild your plan around one sighting. If a citation disappears next week, that is normal according to this data, not a sign you broke something.
- Keep the basics strong. Google says its AI features are rooted in its core ranking and quality systems, so crawlable, accurate, clearly written pages remain the base. We cover that in What Google's May 2026 AI optimization guide actually says.
- Update pages when the facts change, not just to refresh a date.
- Measure what you can see for your own site. Search Console shows how your pages do in Google over time. Compare like with like, and record what you changed and when. How to check a live SEO change explains how.
For how we weigh vendor studies, case studies and official guidance, see Most SEO evidence is uncontrolled.
Where SEO Dispatch fits
SEO Dispatch does not measure AI citation half life, and it does not promise AI citations. It works on the parts you control: whether your pages can be found, crawled and understood, and what Search Console says about your own site over time.
Checklist
- Read one AI citation as a single observation, not a lasting win.
- Record the engine, question, date and page each time.
- Remember the study measured citation share on a vendor's prompts, not clicks.
- Do not use Profound's medians as a goal for your site.
- Put your effort into pages that are accurate, crawlable and worth citing.
- Judge progress with your own data over comparable periods.
Sources
- Davis McCain, Profound, "The half-life of an AI citation is 11 days", 30 September 2026 (figures and methodology): https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days
- Google Search Central, "Optimizing your website for generative AI features on Google Search": https://developers.google.com/search/docs/fundamentals/ai-optimization-guide