You search for your main keyword, see Google's AI Overview, and check whether your site is in the sources. It is, or it is not. Most teams stop there. One look tells you about one day, and the next day often looks different.
We ran 660 US English Google searches in six industries and recorded every site the AI Overviews cited. We then repeated the searches every day for 15 days. Only 11.3% of AI Overviews cited exactly the same sites as the day before. This post explains what that means for AI Overview tracking, what to record, and how to pick the searches you track. The full study, with downloadable data, is at AI Overviews statistics 2026.
What is AI Overview tracking?
AI Overview tracking is recording, day after day, whether Google's AI Overview for a search cites your site. A useful record holds more than a yes or no. It notes where your site sits in the list of sources, which other sites are cited with it, and whether an AI Overview appeared at all. Over many days those records turn into a rate: the share of days you were cited, for each search.
The reason to track at all is how often AI Overviews now appear. On 26 September, 541 of the 600 generic searches in our study (90.2%) showed one. Tracking AI Overviews is one part of what AI search visibility measures, which also covers ChatGPT, Perplexity and the other assistants.
How we tracked 660 Google searches for 15 days
We collected the dataset between 26 September and 12 October 2026, using 660 US English Google searches. 600 were generic searches, 100 in each of six industries: B2B software, finance, health, local services, online shopping and travel. A fixed rule picked them from the highest-volume searches, not a person. The other 60 were "[brand] reviews" searches for the brands named most often.
On day one we recorded the organic top 10, the AI Overview and the AI Mode answer for every search, and measured the pages they pointed to. The 601 searches that showed an AI Overview were then run again every day for 15 days, at the same time each day. These are high-volume commercial US searches, not a random sample of everything people search. The method section of the study has the full detail.
How often do AI Overviews appear on commercial searches?
AI Overviews appeared on 90% of the 600 generic searches we tracked, and on 79% to 98% depending on the industry. Searches that showed an AI Overview kept showing one: on each of the following 15 days, 93% to 98% of them did again.
| Industry | Searches showing an AI Overview |
|---|---|
| Finance | 98% |
| Health | 98% |
| B2B software | 96% |
| Online shopping | 86% |
| Travel | 84% |
| Local services | 79% |
600 generic searches, 100 per industry, 26 September 2026. Source: Babel42 AI Overviews study.
Whether an AI Overview appears is fairly stable from day to day. Which sites it cites is not, and that is where a single check goes wrong.
Why does a single AI Overview check mislead?
A single check misleads because the sources in an AI Overview change most days. Across 15 days of the same 601 searches, only 11.3% of AI Overviews cited exactly the same sites as the day before. The first-listed source, the site Google shows at the top of the source list, changed in 42.3% of searches from one day to the next.
| Day-to-day measure | Result |
|---|---|
| AI Overviews citing exactly the same sites as the day before | 11.3% |
| First-listed source changed from the day before | 42.3% |
| Searches with the same set of sites on all 15 days | 0 of 418 |
601 searches run daily for 15 days. Day-to-day figures compare searches whose AI Overview cited at least one site on both days (530 to 582 a day). The last row covers the 418 searches with a cited AI Overview on all 15 days.
Suppose you check on a Monday and you are listed first. On the figures above, there is a 42.3% chance that a different site is listed first on Tuesday. Report Monday's result as "we are the top source" and you have reported a fact about one day.
AI Overview sources change, but a core of sites stays
Sources in AI Overviews change a lot, but they do not change completely. Each AI Overview cited about 7 sites on a given day (6.7 on average). Over the 15 days, the same search cited about 18 different sites (17.8). The sites you see on one day are only part of the list a search draws on.
| Measure over 15 days | Result |
|---|---|
| Sites cited per AI Overview on one day | 6.7 |
| Different sites cited by the same search over 15 days | 17.8 |
| Searches that kept at least one site on every day | 70% |
| Sites cited on all 15 days, per search | 1.3 |
| Searches that kept the same first source all 15 days | 14% |
| Day-one sites still cited on the last day | 62% |
| AI Overviews identical on the first and last day | 0.9% |
601 searches run daily from 28 September to 12 October 2026. Source: Babel42 AI Overviews study, daily change.
The stable part is small. On average 1.3 sites per search were cited on all 15 days. A site in that core is cited every day. A site outside it is cited on some days and not others, and one check cannot tell you which group you are in.
How daily change in AI Overviews differs by industry
Daily change happens in every industry we tracked, but the size varies. B2B software was the most stable, and still only 21.3% of its AI Overviews cited the same sites as the day before. Health was the least stable at 4.0%.
| Industry | Same sites as the day before | First source changed |
|---|---|---|
| B2B software | 21.3% | 36.1% |
| Online shopping | 13.1% | 43.1% |
| Local services | 14.0% | 53.5% |
| Travel | 10.5% | 50.3% |
| Finance | 6.6% | 42.3% |
| Health | 4.0% | 32.1% |
601 searches, average of 14 day-to-day comparisons per industry.
The two columns do not move together. Health changed its full set of sites most often, but changed its first source least often (32.1%). Local services changed its first source in 53.5% of cases. Look at both measures for your own industry, because "are we cited" and "are we cited first" behave differently.
Measure AI Overview citations as a rate across many days
The figures above point to one conclusion: measure citation as a rate, not a yes or no. For each search you track, count the share of days your site was cited. Then record the position you held and the other sites cited with you. A result such as "cited on 9 of 15 days, usually fourth, alongside the same three review sites" tells you far more than "cited today".
A rate also makes changes easier to read. If your share of days drops after a site change, that means something. A single missing day does not, because the sources shift for most searches from one day to the next. The same thinking applies to AI citation rate across other AI assistants, and to measuring AI search visibility in general.
What to record for each search, each day
Record the same fields for every search, every day, at the same time of day. In the study, each search ran once a day at a fixed hour, so each day was collected the same way. Use one row per search per day.
| Field | Why it matters |
|---|---|
| AI Overview shown that day? | Days with no AI Overview are not days you lost. Count them apart. |
| Was your site cited? | The basis of your citation rate. |
| Your position in the source list | The first-listed source changed in 42.3% of searches day to day. |
| Every other site cited | Shows who you share the answer with, and who replaces you. |
| Number of sites cited | Each AI Overview cited about 7 sites a day (6.7). |
Count a site once per answer, however many of its pages are cited. That is how our study counted, and it keeps the rate from rising just because Google lists two of your pages.
How many days of tracking before a number means something?
One day is not enough, and our data shows why. The same search cited 6.7 sites on an average day and 17.8 different sites over 15 days. Even comparing the first day with the last, only 62% of day-one sites were still cited.
The study does not set an exact minimum number of days. What it does show is that 15 daily checks gave a far fuller picture than one, and that no set of sources stayed fixed for that long. Treat a rate built on a handful of days with care. Run each search on as many days as you can, and judge changes on the rate over that period, not on one day.
How to pick the searches to track
Pick searches by a fixed rule, not by hand. The study started from a fixed set of topics in each industry, kept searches with a buying or research word, and dropped navigational searches, "near me" searches and near-duplicates. It then took the highest-volume searches that were left. A fixed rule stops you picking only the searches where you already do well.
Add searches for your own brand. The study included 60 "[brand] reviews" searches, because buyers search for reviews of the brands they are considering. Check whether each search shows an AI Overview before you commit to it. In our data local services showed one on 79% of searches, against 98% in finance and health, so some industries will lose more searches at this step. The study's most-cited sites show which other sites you are likely to share answers with in your industry.
Track AI Mode as a separate result
AI Mode cites different sources from AI Overviews, so a rate for one does not stand in for the other. For the same search on the same day, the two shared 32.5% of their cited sites on average in our study. Our guide to AI Mode vs AI Overviews covers the differences in more depth.
Tracking AI Overviews with Babel42 AI Visibility
Babel42's AI visibility tool checks whether Google's AI Overviews and AI Mode, ChatGPT, Claude, Perplexity, Gemini and Grok cite and recommend your brand, and lists the sources they used. Checks repeat on a schedule set by your plan, from weekly on the free plan to daily on Pro, so you build a record over many days rather than relying on one look. Each run also records when Google showed no AI Overview for a search.
The free plan runs one AI Buyer across any two of the seven assistants each week, with no card required. To see the full numbers behind this post, including the sources cited on each day, download the data from the AI Overviews study.
