Google AIO SERP Tracking: How to Track AI Overviews for Your Keywords

A weekly log of trigger, citation and winning domains, kept beside the results page it describes.

There is something faintly comic about a modern rank report. It tells you, with real precision, where your page sits among the blue links, while just above them a Google AI Overview answers the question in its own words and cites somebody else. The report is accurate. It is simply describing the wrong layer of the page.

AI Overview tracking is a measurement practice that records, for each keyword, whether Google shows an Overview, whether your page is cited inside it, and which domains are cited instead. The tooling has started to arrive: rank trackers now flag AIO mentions and AIO rank as two separate signals and let you compare dates to see how the cited domains shift. That changes the decision in front of a marketing director this week. Keep counting positions, or start counting citations.

In our audit work we check whether ChatGPT, Claude, Perplexity and Google AI Overviews name our clients, and Google is the odd one out. Its answer sits on top of a page people thought they already understood. I will confess a small prejudice: I trust a hand-checked sheet of a few dozen keywords more than a dashboard of thousands, though the dashboard has far better stamina. Most teams will need both. What follows is the method, in the order I would run it.

What should you record for every keyword you track?

For each tracked keyword, log whether an AI Overview appears, whether your page is cited, whether that citation is linked and visible, and which competing domains win instead.

1. Trigger: did this keyword show an AI Overview on the day you checked?
2. Citation: is your page cited or mentioned inside the Overview?
3. Visibility and link: how prominent is that inclusion, and does it carry a clickable link?
4. Competitors: which domains does the Overview cite in your place?
5. Date: when was the check run, so this week can be laid against last month?

In 2024, Mike King of iPullRank told The Agency Blueprint podcast that a page can rank around 70th organically and still appear in the AI Overview, because authority matters little there and the relevance of the on-page passage matters most. He added that the top three sources an Overview cites are often not the top three organic results. The common assumption is that a strong blue-link position carries over into the Overview. So give citation its own column, and never fill it in from the rank column beside it. The Overview behaves less like the top of the list and more like a separate organism living on the list, feeding on whichever passage suits it.

A year later, the host of the Crawling Mondays podcast laid out what actually decides a site's exposure. Impact depends on the share of traffic coming from affected queries, on shifts in impressions or search popularity, on whether the site is included in the AI Overview, how visible that inclusion is, and whether it carries a link. Three of those five factors live inside the Overview itself. A rank tracker built for blue links reports none of them.

Tooling has started to catch up. At least one commercial rank tracker now records two separate signals per keyword, a mention inside the Overview and a rank inside it, lighting an icon green when your site is included and leaving it gray when it is not. It also lists every domain sitting inside the AI box, which is the field most teams forget. Knowing you lost matters less than knowing who won.

In our own audit work we track whether ChatGPT, Claude, Perplexity and Google AI Overviews name our clients, scoring each site against 53 criteria across 5 pillars. It seems to me the same discipline belongs in any team's keyword sheet. A review of 4 sources suggests one consistent shape: the Overview is its own surface, with its own winners and its own rules of admission. If your reporting still infers citation from position, our notes on research-driven content built on evidence explain why the passage, not the page, is the unit that gets chosen.

Position tells you where you stand in the queue. The five fields tell you whether anyone was listening, and they also expose something odd the moment you run them twice in the same week: the answers do not hold still.

Why do AI Overview numbers swing so much, and when is a trend trustworthy?

AI Overviews come and go for the same query, so trust only keywords that triggered consistently across your whole window, then compare them with a period before Overviews appeared.

Run your five fields on Monday and again on Thursday, and you will probably find the sheet has changed while your site has not. The host of the Crawling Mondays podcast put it plainly in 2025: AI Overviews are volatile, and they have so far touched mostly informational queries. Commercial terms often show no Overview at all. A daily count of how many keywords triggered one is therefore measuring Google's mood as much as your progress.

The fix described on that episode is a discipline rather than a tool. Use only queries where an Overview appeared consistently across the analysis period. Then compare those queries against a period when they had no Overview. Everything else gets set aside, however tempting it looks on a chart.

• Keep keywords that showed an Overview on every check in the window.
• Park keywords that flickered in and out; they add noise, not signal.
• Compare the kept set against the same queries in a window before Overviews were shown.

How many queries is enough? The honest answer is that the evidence does not give a minimum number. What it does show is the failure case. On the demo site in that 2025 walkthrough, filtering for Overview keywords returned only a couple of keywords, which the host called not enough data for the analysis. A cohort that small cannot carry a conclusion. Build a set large enough that one keyword dropping out does not swing the result, and say so in the report.

It is a little like counting a population of cells under a microscope. Any single cell is doing something odd at any given moment, dividing or resting or simply drifting out of frame. Only the population tells you whether the tissue is healthy. I think of our own records the same way. We have run 26,577 real AI-visibility audits and scored 11,000+ domains across 15 sectors and 28 categories, and in our experience a number that size earns its keep only as a population, never as one reading. Your keyword set deserves the same humility.

Two consequences follow. Weekly snapshots are fine for spotting who won a citation. They are poor evidence for telling a CFO that traffic fell because of AI Overviews. If your team needs a primer on why volatility matters to AI answers generally, our answers on AI search optimization cover the basics. The harder work is turning a stable cohort into a routine that someone actually runs every week, and ties to clicks, impressions and pipeline.

How do you set up a repeatable AI Overview tracking routine?

Fix one keyword set, log trigger, citation and winning domains on the same day each week, then report citation share beside clicks, impressions and pipeline every month.

The payoff from doing this patiently can be large. Our client Understood Care reached 3.9K monthly organic visits and 132 Google AI Overview citations through the content we produced, and 132 is a number you only know if someone is counting citations rather than positions. Here is the routine I would hand a marketing team on a Monday morning.

1. Build the set. Export your tracked keywords, keep the informational ones that trigger an Overview on every check, and freeze the list for the quarter.
2. Pick a cadence. Check on the same weekday, in the same market, with the same tool, so changes reflect Google and not your method.
3. Log the five fields. Trigger, citation, visibility and link, winning domains, and the date, one row per keyword per check.
4. Calculate citation share. Divide the keywords where you are cited by the keywords that triggered an Overview, for the frozen set only.
5. Track displacement. When you lose a citation, record which domain took it. When you win one, record who you replaced.
6. Join it to revenue. Once a month, lay citation share next to Search Console clicks and impressions for the same keywords, and next to booked calls or opportunities in your CRM.

Step six is where most programs quietly stop. It is also the step that leadership cares about. Citation share on its own is a vital sign, like a pulse; it tells you the organism is alive, not whether it is thriving. One of our clients went from a quiet site to 10 booked calls in 3 weeks. Booked calls are the unit a CFO reads.

Give the routine enough time to mean something. We guarantee our clients are named in AI answers across their target question set within 100 days, or our team keeps working for free, and that window is a sensible floor for judging any tracking program. A month of data is mostly weather. A quarter begins to look like climate.

The displacement column tends to surprise people. Sometimes the domain that took your citation is not a direct competitor at all but a listing, directory or comparison page. If the same aggregator keeps winning, it may be faster to appear on it than to outrank it, and our page on sponsored listings AI engines cite covers that route. Either way, the column tells you where to spend next.

A routine like this produces a tidy monthly chart. Before you present it to leadership, ask one question of every row: which domain won this keyword, and has that domain held it for the whole quarter?

What should you change about AI Overview tracking this week?

This week, add a winning-domain column to the tracker you already run, choose the weekday you will check, and decide which keywords stay frozen until the quarter ends.

Position will keep mattering. It is just no longer the whole animal. In 2025, the Crawling Mondays host noted that the industry had asked Google for an AI Overviews filter in Search Console. Until one arrives, the sheet you build is the record.

One small pilot of ours keeps coming back to me. Our study step read a product's repositories and finished a two-topic study in about two minutes, with each topic tied to one to four specific files and line ranges, and it dropped a topic whose controls were not yet wired up in the code. The habit is the one Overviews seem to reward: point at the exact passage, and say nothing you cannot point to. It is why we are content to have our own work judged on whether clients get named in AI answers, inside a fixed window, and not on where a blue link happens to sit on a Tuesday.