The Median AI Citation Lasts 11 Days

Profound tracked 883,000 pages across 7 AI engines for 12 months and found the median AI citation has an 11-day half-life from peak. Not from first citation — from peak. The page that gets cited today will, on median, lose half that citation share within 11 days. Published September 30, 2026, with a dataset of 1.19 million page/engine lifecycles, this is the most rigorous citation longevity study to date — and it arrived the same week that DemandSphere data showed Google AI Overviews appearing for 80%+ of branded queries. These are different findings. They confirm the same thing: the GEO question is not "can we get cited?" It is "how do we stay cited?"

What the Profound data shows

Profound's analysis covered 7 AI engines (ChatGPT, Perplexity, Google AI Overviews, Gemini, Microsoft Copilot, Claude, and one additional engine) with 12 months of longitudinal citation tracking. Key findings:

• Median half-life: 11 days from peak — the median cited page loses half its peak citation share within 11 days
• 78% of cited pages are single-event citations — they reach a peak and fall to below half-peak within 2 weeks
• 22% are sustained citations — these have a median half-life of 27–29 days
• Claude citations persist 2.8× longer than ChatGPT among pages in the sustained 22%
• Cross-engine breadth = 5.5× durability — pages cited by 5+ engines are 5.5× more likely to stay above half-peak at day 56 vs. single-engine pages
• Date-tied content decays at 1.4–1.6× the baseline rate — event-anchored content falls faster than evergreen content

Corroboration exists from two independent studies. An SSRN working paper by Sangra (August 2026) put the median at 4.5 weeks using a different methodology — measuring from first citation to half-peak rather than from peak — and the direction is consistent. A Writesonic study (June 2026) independently found an 11–15 day median lifespan. Three separate research teams, three separate methodologies, same conclusion.

The six predictors that separate the 22% from the 78%

The Sangra SSRN paper cross-referenced citation persistence with page and brand characteristics and found six variables with statistically significant correlation to sustained citations:

1. Brand search volume— 0.334 correlation coefficient, the single strongest predictor. Pages from brands people actively search for stay cited longer. This is not a content quality signal — it is an authority signal. Brand search volume determines whether an AI engine's retrieval system treats a citation as reliable or experimental.
2. Original research content — 82% citation rate among sustained pages vs. the baseline. Pages that contain proprietary data, original studies, or first-party benchmarks hold citations because there is no substitute source. The AI engine has no alternative — it cites or omits the data point entirely.
3. Third-party attribution — 6.5× more citations than self-referential content. When other credible sources cite you before the AI engine does, the AI engine is more likely to treat you as a primary rather than self-promotional source. Earned media compounds citation durability.
4. Content freshness — 3.2× citation lift on Perplexity for pages updated within 30 days. Not published within 30 days — updated within 30 days. AI engines weight freshness signals differently from traditional SEO. A two-year-old page updated last week outperforms a two-week-old page not touched since publish.
5. Schema markup completeness — correlated with sustained citation probability. Pages with complete entity markup (Organization, Article, Author, FAQPage where appropriate) are structurally more citable. The AI engine can parse and attribute them with confidence.
6. Bing indexation— a proxy for cross-engine coverage. Bing's index feeds Copilot, the second-highest retention platform at 34%. Pages not indexed on Bing are excluded from Copilot citations regardless of their quality on other platforms. Cross-engine breadth is the 5.5× durability lever — Bing indexation is a prerequisite for accessing it.

The branded query problem

Separate from the half-life research: DemandSphere tracking showed Google AI Overviews appearing for over 80% of tracked branded queries by late September 2026, up from approximately 26% at the start of the month. Search Engine Roundtable testing by Chris Long found AI Overviews on 93 of 100 major brands tested (Reddit, Salesforce, Amazon, Adobe, Apple, Samsung, and others). The surge peaked at 90.48% on September 27.

This matters for the half-life argument. For any brand large enough to trigger a branded AI Overview — which is now nearly any recognized business — the question is not whether an AIO appears when someone searches for them. It is who gets cited in that AIO. A competitor's review site, a third-party directory, a news article critical of your product — all of these are citation-eligible. Your own site is citation-eligible too, but only if your GEO score clears the bar.

The two findings compound: your brand queries are almost certainly triggering AI Overviews now. The citations in those Overviews decay at an 11-day median half-life. The sources that replace them after they decay are determined by whoever wins the next retrieval cycle. If you are not actively optimizing for citation, someone else fills the gap.

Why it matters

1. One-time GEO audits produce one-time citation spikes. If the median citation decays within 11 days of peaking, a technical audit that happened six months ago is not protecting your citation share today. The structural changes — schema, entity disambiguation, page completeness — hold. But the citation positions they unlock decay and refresh continuously. The audit sets the floor; ongoing content freshness, original research, and earned coverage raise the ceiling.
2. 78% of brands who "got cited" are already gone. A page that appeared in AI Overview citations last month is, with 78% probability, no longer appearing at the same rate. Most GEO practitioners celebrate citation events. The Profound data says the average citation event has a two-week lifespan. Celebrating it is fine. Assuming it persists is expensive.
3. Platform choice affects decay rate by 4×. Perplexity retains citations at 44% at 28 days. Gemini retains them at 11%. If you are measuring "AI citation rate" as an aggregate across all platforms, you are averaging a retention leader with a retention laggard and getting a number that describes neither. Platform-specific strategy matters: Perplexity rewards freshness (3.2× lift within 30 days). ChatGPT favors cross-engine authority. Gemini is the most volatile — high peak, fast decay.
4. The 22% who sustain citations are predictable. They are not luckier. They score higher on the six predictors. Brand search volume, original research, third-party attribution, freshness, schema completeness, Bing indexation — these are engineering targets, not aspirations. The path from 78% to 22% is auditable and executable.

What to do

1. Audit your position on each of the six persistence predictors. Score your brand and content against the six factors from the Sangra paper:
   • Brand search volume: are you tracking branded search volume monthly, and is it growing?
   • Original research: does your site contain proprietary data, studies, or benchmarks that no other source replicates?
   • Third-party attribution: are credible outlets citing your content before AI engines have to decide whether you are authoritative?
   • Content freshness: do your highest-value pages get substantive updates within 30-day windows?
   • Schema completeness: does every major page carry appropriate structured data with entity markup?
   • Bing indexation: are your priority pages in Bing's index? If not, you are invisible on Copilot regardless of other scores.
2. Set up a 30-day content refresh cadence for your highest-value pages. The 3.2× Perplexity freshness lift applies to pages updated within 30 days. "Updated" means substantive content change — not a metadata tweak. Target your five to ten highest-traffic or most commercially important pages. Add new data, update statistics, expand the "What to do" section with any new practitioner guidance from the past month. Monthly refresh is the minimum viable cadence.
3. Build for cross-engine breadth, not single-engine optimization. The 5.5× durability advantage of being cited by 5+ engines versus 1 engine is the most powerful lever in the dataset. Cross-engine presence requires being in both Google's and Bing's indexes, having complete entity schema that multiple retrieval systems can parse, and achieving the third-party attribution that lifts you above single-platform consideration.
   • Verify Bing Webmaster Tools indexation for all priority pages
   • Ensure your Organization schema includes consistent sameAs references across LinkedIn, Crunchbase, and relevant directories
   • Target press and earned coverage in publications indexed by multiple crawlers — not just Google News
4. Measure citation retention rate, not just citation events. A citation that lasts 11 days and a citation that lasts 90 days are not equal. If your GEO reporting tracks "times cited this month" without tracking how long each citation persisted, you are measuring noise. Set up monthly snapshots with tools like Profound, Qwairy, or BotRank. Flag pages that show high peak citations but rapid decay — these are the candidates for freshness refresh and third-party attribution investment.
5. For branded queries specifically: run a monthly brand AIO audit. With AI Overviews now appearing for 80%+ of branded queries, you need to know: what is the AI saying when someone searches for your brand? Who is getting cited? Is it your site, a competitor's review, or a third-party directory? A monthly manual check of 10–15 key branded queries takes 30 minutes and tells you whether your brand narrative is controlled or contested.

Chad's take

Here is the thing GEO practitioners do not want to hear: the AI citation you earned last month is almost certainly gone. Not diminished — gone. The Profound data does not soften this. 78% of citation events have a two-week shelf life. The practitioners who are ahead are not the ones who got cited. They are the ones who built the infrastructure — original research, earned coverage, fresh content, cross-engine presence — that generates new citation events every 30 days. The uncomfortable truth is that GEO is not a project with a finish line. It is a production function. You are either running it continuously, or you are ceding the citations you earned last month to whoever is.

Sources

• Profound — “The Half-Life of an AI Citation Is 11 Days” — longitudinal analysis of 883,000 pages across 7 AI engines and 12 months (1.19M page/engine lifecycles). September 30, 2026. tryprofound.com
• Sangra / SSRN — Working paper on AI citation persistence and brand authority predictors. Six-variable regression analysis across multiple AI engines. August 2026.
• Writesonic — Independent AI content lifespan study finding 11–15 day median citation window. Consistent with Profound methodology directionally. June 2026.
• DemandSphere — Google AI Overviews branded query tracking showing surge from 26% to 80%+ of branded queries in September 2026, peak 90.48% on September 27. Reported by Search Engine Land, October 1, 2026. searchengineland.com
• Chris Long / Search Engine Roundtable — Manual testing finding AI Overviews on 93 of 100 major brand queries tested. September 30, 2026.