Around three in four AI citation drops were temporary in Ahrefs' study of five AI products over 20 days (Search Engine Journal's write-up of the study). So when a tracker flags a lost citation and someone wants the page rewritten by lunch, the better first move is to wait and check.
This guide gives you a four-check triage, a simple log with a seven-day count, and a script for explaining the result to a client or boss. It won't tell you how to win a citation back, because the study did not test that.
Why a single lost citation tells you so little
The study followed about 37,500 first-day citations from Google's AI Mode and AI Overviews, ChatGPT, Gemini and Copilot. By day two, about half of those references had left AI Overviews, and even more had left on the other tools. Across the five tools, about half of the citations that dropped came back within three days.
That means one daily check can't separate a real loss from normal churn. It also means an edit made on the day of a drop can pick up credit for a comeback it did not cause.
Step 1: Build the daily log
Start the log before the next drop happens. Record one row per question, per tool, per day. Use these columns:
| A: Date | B: Question | C: Tool | D: Cited (Y/N) | E: Who was cited instead | F: Brand named (Y/N) | G: Ranks in Google (Y/N) | H: Days cited, last 7 | |---|---|---|---|---|---|---|---| | 2026-11-01 | example question | AI Mode | Y | | Y | Y | |
In column H, count the days your site was cited for that question on that tool over the last seven days, including today:
=COUNTIFS($B:$B,B2,$C:$C,C2,$D:$D,"Y",$A:$A,">"&A2-7,$A:$A,"<="&A2)
Drag it down. Use real dates in column A, not text, or the date comparison will fail. Leave column E empty when the answer cited nobody. That blank is useful on its own.
The study's authors recommend counting how many of the last seven days a site was cited. A count like "5 of 7" tells a manager far more than a yes or no from a single check. It shows how often the site turns up, though not how long any one citation held.
Step 2: Run the four-check triage
Work through the checks in this order, and write the answer to each in the log or a note.
- Read the answer. Did the tool cite anyone? Open the actual response. If it has no links at all, the loss reflects that day's tool behavior, not your page. In the study, Gemini answered without citations about 64% of the time and Copilot about 53%. AI Overviews, AI Mode and ChatGPT almost always cited sources (fewer than 1% of answers had none), so on those three a lost citation almost always means another site took the spot. If you see a competitor link, record the domain in column E.
- Check the history. How many of the last seven days were you cited? Look at column H. In the study, citations that lasted one day before dropping failed to return 34% to 41% of the time on AI Overviews, AI Mode and ChatGPT. For those that held eight to 14 days, the figure was 4% or less. A site with a long run that vanished yesterday is in a different position than one that appeared once.
- Check the brand. Is it still named? A citation is a link. A mention is your name in the answer text, linked or not. The study found 50% to 70% of cited sites showed up on only one day per tool, compared with roughly one in five named brands. Mentions moved far less than citations. Scan the answer for variations of your name too, since tools shorten or lengthen brand names from answer to answer. Record the result in column F.
- Check rankings. Does the page rank in Google for the question? On AI Overviews and AI Mode, cited pages that also ranked in Google for the question held their citations longer. That finding came from an analysis of 230 questions, and it is a correlation. Ranking was not required: most pages cited by AI Mode and ChatGPT did not rank for the question. Record yes or no in column G.
Telling the three look-alikes apart
A no-citation answer shows N in column D, an empty column E and no links in the response. A competitor swap shows N in column D and a domain in column E. A lost citation with the brand still named shows N in column D and Y in column F. Each calls for a different response, which is why the log keeps them in separate columns. This pairs well with how to evaluate aeo research before changing your seo explained.
Step 3: Walk through a worked example
The log below is hypothetical and illustrative. It is not real data and does not reflect any actual site or tool.
Question: "how to audit a content calendar." Tool: AI Mode. The tracker flags a loss on Sunday.
| Day | Cited | Who was cited instead | Brand named | Ranks in Google | |---|---|---|---|---| | Mon | Y | | Y | Y | | Tue | Y | | Y | Y | | Wed | N | Competitor A | Y | Y | | Thu | Y | | Y | Y | | Fri | Y | | Y | Y | | Sat | Y | | Y | Y | | Sun | N | Competitor A | Y | Y |
The seven-day count is 5 of 7. Now run the checks. AI Mode nearly always cites, so Sunday's gap is a swap, not an empty answer. The history shows a run of citations with one earlier miss. The brand is still named, and the page still ranks.
The decision is keep watching. Nothing on the page has been shown to be wrong, and the pattern looks like fluctuation. This pairs well with our guide to how to check core update impact during holiday traffic.
Step 4: Turn the checks into a decision
The study offers one timing rule. Its write-up, by Ahrefs' Mateusz Makosiewicz, treats one missing citation as no reason to change a page and suggests reviewing the page if the citation is still missing after two or three weeks. The four outcomes below build on that rule. The extra conditions are working rules, not findings from the study, so adjust them to your own question set.
Ignore. The answer cited nobody (common on Gemini and Copilot), or the count is still high and the brand is still named. Log it and move on.
Keep watching. You were cited recently, you have been missing for under about two weeks, and the page still ranks. Check the log daily and leave the page alone.
Review the page. The citation has stayed missing for two to three weeks. Start with how the page performs in Google for the question, then how well it answers that question. If another site is cited in your place, read that page too. If rankings dropped alongside the citation, start there.
Edit. Edit only when the review turns up a specific problem, such as a lost ranking with a clear cause or an answer that is thin or out of date. Write down the date and the change, and don't promise the citation will return.
Step 5: Avoid crediting a coincidental comeback
Most citations that disappeared in the study came back, so a comeback after your edit is weak evidence. Makosiewicz wrote, "Don't credit a page edit for a single comeback." The reverse holds too. Half or more of citations were gone by the next day in the study, so a drop the day after an edit is weak evidence that the edit caused harm.
To judge an edit fairly, compare the seven-day count for several weeks before the change with several weeks after. Look for a sustained increase, not one good day. Keeping a few unedited questions in the log as a comparison also helps.
Also read: related topic: block gptbot, allow oai-searchbot: a robots.txt audit
A short script for clients and bosses
Adapt this wording:
"The tracker shows we lost the citation for [question] on [tool] on [date]. We checked the answer: [it cited nobody / [competitor] was cited instead]. Over the last seven days we were cited [X] of 7 days, and our brand [is / is not] still named. The page [does / does not] rank in Google for this question. Research from Ahrefs found most dropped citations return, so we are watching it for two to three weeks before reviewing the page. If it is still missing then, we will check rankings and how well the page answers the question before proposing changes."
For the hypothetical example above, the short version is: "Cited 5 of 7 days, brand still named, page still ranks, so we are watching."
What this method cannot tell you
Keep these limits in mind before you quote the study to anyone:
- It is one organization's question set, answered daily by each tool over 20 days.
- It did not test whether editing a page restores a lost citation.
- It tracked only the first time each citation dropped, so it does not show how often citations vanished again after returning.
- It measured at site level, so a returning site may not be the same page that dropped.
The authors also note their return figures may be an underestimate, since citations lost late in the study had less time to come back. Model changes could shift these patterns over time. Ahrefs sells Brand Radar, and its own tracking there showed about two-thirds of vanished citations coming back, a lower rate than in the main study.
What to expect next
The first week of logging will feel slow, because the seven-day count only becomes useful once seven days exist. After two or three weeks, you should be able to see which questions are stable, which swing between tools, and which have been quietly missing for a while. That is when a page review has real evidence behind it, and when the next "why did we lose this?" gets a calm answer.



