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.