# Ahrefs tracks 37,500 AI citations: half vanish overnight, but most return A new Ahrefs 20-day study across five AI answer tools finds 50–83% of day-one citations disappear by the following day, yet roughly three-quarters of lost citations return. A single AI-visibility snapshot is therefore a weak basis for SEO decisions. Ahrefs tracked roughly 37,500 cited websites for 20 days and found high daily citation churn, frequent recoveries and major differences between AI tools. - Status: Active - Published: 2026-10-30T06:27:32+13:00 - Updated: 2026-10-30T06:27:32+13:00 - Categories: Marketing & Distribution, SEO & Search, GEO & AI Search, Conversion & Analytics - Tags: Ahrefs, AI search, citation analytics, generative engine optimization, SEO - Canonical HTML: https://beyondthe.news/dossiers/ahrefs-37500-ai-citation-volatility-20-day-study-october-2026 ## What changed On October 29, 2026, Ahrefs published a longitudinal analysis of approximately 37,500 website citations initially observed across five AI tools and tracked daily for 20 days. The day-one sample covered 1,996 Google AI Mode questions (18,009 cited websites), 3,173 ChatGPT questions (11,225), 563 Google AI Overview questions (4,830), 1,276 Gemini questions (1,994), and 823 Copilot questions (1,447). By the next day, 50% of AI Overview citations, 58% of AI Mode citations, 59% of ChatGPT citations, 81% of Gemini citations and 83% of Copilot citations were absent. Yet approximately 75% of citations that disappeared at least once reappeared within the study window. Crucially, 64% of Gemini answers and 53% of Copilot answers contained no citations, inflating apparent disappearance; excluding those responses brings next-answer churn closer to 58–60%. ## Why it matters Builders and marketers using AI-citation reports to judge SEO/GEO changes can easily mistake ordinary answer variation for a lost placement or an optimisation win. Ahrefs' evidence suggests measuring recurring appearances over seven or more days, separating brand mentions from link citations, and treating a one-day disappearance as noise until it persists. For product teams buying AI-visibility dashboards, snapshot-only reports may be misleading. This is observational data from a vendor selling measurement tools, not proof that a specific SEO tactic causes citation retention. ## The study follows the same questions, not one-off screenshots Ahrefs' researcher Xibeijia Guan asked five AI tools the same questions daily for 20 days, tracked the cited websites from day one, and measured whether each persisted or returned. The sample contains roughly 37,500 day-one citations across 7,831 tracked questions. Ahrefs also cross-checked selected patterns against two Brand Radar question groups (25 and 69 questions), but those smaller checks are not a representative external replication. ## Most citations disappear quickly and many return Between 50% and 83% of day-one citations were missing the following day depending on platform. Roughly 75% of citations that vanished at least once reappeared within 20 days, around half within three days. These are different denominators: return rates describe citations that dropped, not all citations. A late disappearance has less time to return within the observation window. ## Empty answers distort Gemini and Copilot comparisons In this dataset, 64% of Gemini and 53% of Copilot answers had no source citations. Skipping such responses changes next-cited-answer disappearance from 81% to 58% for Gemini and from 83% to 58% for Copilot. Skipping empty answers also changes the time interval being compared, so it cannot be treated as a perfect apples-to-apples adjustment. ## Brand facts, search ranking and source selection For questions naming a brand, its own website was cited on 60–68% of days in Google AI tools and ChatGPT, versus 16–20% for other sources. Pages ranking for the same question in Google were more likely to retain AI Mode/Overview citations for at least a week (28–29% versus 4%); however the ranking analysis covered only 230 questions and is correlational, not evidence that changing rank alone causes citations to persist. ## What this adds to earlier citation-drift studies Earlier SISTRIX research tracked 82,619 prompts over 17 weeks and found high weekly source replacement, while other longitudinal studies examined citation-share decay after peak. Ahrefs adds a daily disappearance-and-return view. Weekly domain churn, page citation-share half-life and day-to-day citation survival are different metrics and should not be combined into a single universal retention rate. ## Key details - Published October 29, 2026; roughly 37,500 day-one citations across five AI tools and 20 daily observations. - Day-two disappearance: AI Overviews 50%, AI Mode 58%, ChatGPT 59%, Gemini 81%, Copilot 83%. - Approximately 75% of citations that disappeared at least once returned within the 20-day window. - Gemini 64% and Copilot 53% no-citation answers; skipping these lowers next-cited-answer disappearance to about 58% each. - Brand's own website is much more persistent for brand-named questions than third-party citations in Google's tools and ChatGPT. - Ranking-retention comparison used only 230 questions and cannot establish causation. - The sample is observational and vendor-produced; question mix and engine versions limit generalisation. ## Builder takeaways - Do not rewrite pages or buy new tools because one daily AI citation disappears. - Track repeated citation presence for a fixed set of questions over seven or more days, and distinguish domains from exact pages. - Separate no-citation responses from genuine source replacement, especially in Gemini and Copilot. - Track brand mentions and business outcomes alongside link citations, rather than treating citation counts as the only GEO KPI. - Compare measurements within each AI engine and keep methodology and query mix consistent. ## What to watch - Independent replication with representative industry and non-marketing question sets. - Effects of model updates, region and account settings on daily citation churn. - Whether multi-week citation frequency correlates with qualified referral traffic or revenue. - Changes in reporting products from snapshots to repeat-sampling and confidence intervals. ## Uncertainties - The question sample and day-one citation cohort are not a random sample of all AI use. - A 20-day window undercounts returns for citations that disappear near its end. - Excluding no-citation answers changes observation intervals. - Brand and organic-ranking correlations do not establish causal optimisation effects. - Exact model versions, geographies and personalisation may change results. ## Sources - [How Stable Are AI Citations? 10 Things We Learned Tracking 37,500 Citations](https://ahrefs.com/blog/how-stable-are-ai-citations/) — Ahrefs · primary_original_study · 2026-10-29T00:00:00+13:00. Research methodology, daily platform comparisons, return rates, no-citation caveats and subgroup limitations. - [AI Citation drift: How stable are sources in AI search results?](https://www.sistrix.com/blog/ai-citation-drift-how-stable-are-sources-in-ai-search-results/) — SISTRIX · independent_longitudinal_research · 2026-05-01T00:00:00+12:00. Independent 82,619-prompt weekly study; corroborates source turnover but uses different time intervals and units. - [Citation Drift: How Stable Are AEO Citation Sources?](https://alexbirkett.com/citation-drift/) — Alex Birkett · independent_comparative_analysis · 2026-08-01T00:00:00+12:00. Compares distinct source-churn datasets and cautions against conflating domain and page volatility.