How to track Google AI Overviews and AI Mode

Compare manual checks, tracker exports and Search Console impressions with a fixed query set and a clear way to record missing data.

Sonar examines three cards marked with a ring, a link and a dash to classify search observations.

A Google result can show an AI Overview that links to a competitor. A tracker can also return an empty row because it never collected the result. Both can look like a zero in a spreadsheet. Only the first tells you anything about which sources appeared. Keep the status of the search separate from the status of your site before you start counting citations.

Use a fixed set of searches to see how often an AI Overview appears and which URLs it links to. Run a separate set of observations for AI Mode. Then use Google Search Console’s Generative AI performance report to see the impressions Google recorded for your property in eligible AI features. Each source answers a different question, and the counts belong in separate columns.

Start with a decision your team can make

A health check begins with a single question. If you want to know whether your migration left old URLs in AI answers, select queries for pages that moved and record the exact destination of each source link. If you want to evaluate a tracking tool, select the searches your team actually needs to monitor. A generic database of other keywords can be useful for discovery, but your purchase decision depends on coverage of that smaller set.

Choose twelve searches for a first pass. For a publication, a useful starting mix is four diagnostic questions, four comparisons, and four practical tasks that match pages you maintain. Weight that mix to your own audience. Include a branded query as a check on brand recognition if that is part of the assignment. Mark it as branded before calculating a combined rate, since a brand-dominated sample can make coverage appear stronger.

Write down the expected URL for each task before collecting results. If two pages could reasonably answer the same query, record both and note which one you would prefer to maintain. A source link to an unexpected but relevant page is a finding to investigate. Check intent, canonical URL, and the passage that supports the answer before calling it a mistake.

Capture each Google surface separately

For AI Overviews, start from an ordinary Google results page and record whether a generated summary actually appeared. Open its visible source links and record the observed destination URLs. Keep ordinary organic links below the feature in their own column. A page that ranks organically and sits below an AI Overview should count in an organic-rank report.

For AI Mode, start a fresh conversation for the initial query. Decide in advance whether your study includes follow-up prompts. A follow-up inherits context, so record the preceding messages when they are in scope. A fresh answer and a follow-up are two different observations. Google’s generative AI documentation explains how its Search features retrieve information and can issue related queries behind the scenes.

Give every row a timestamp, country, interface language, device, account state, query ID, and collection method. Use the same market and device for tools you want to compare. Preserve the original evidence when you can. A fresh browser session reduces account-history effects, but location, experiments, and changing responses can still move the result. Treat the manual capture as the reference for that run.

Keep a missing observation separate from a missing citation

How to classify an AI answer observation
State Evidence you observed How to count it
Cited The answer showed a source link to the property Include in the completed-check and feature-shown totals
Feature shown, property absent A captured answer showed sources and none resolved to the property Include in the completed-check and feature-shown totals
Feature absent A completed ordinary Search result showed no AI Overview Include in the completed-check total
Unobserved The provider lacked a row, the page failed to load, or the sources were indeterminate Log the reason and report the missing share separately

Suppose a tracker had twelve scheduled checks. Ten completed, six showed an AI Overview, and two of those linked to the property. In this example, the feature appeared in 6 of 10 completed checks. The property was cited in 2 of 6 answers that appeared. The tracker left 2 of 12 scheduled checks unobserved. Dividing every number by twelve would obscure the difference between search behavior and collection coverage.

Add a separate text field for brand mentions. A publisher can appear by name while the link points to a third party. Keep both the displayed URL and its resolved canonical when available. Deduplicate the same property within one answer when calculating a property-level citation rate. Keep each distinct source URL if the purpose is to improve individual pages.

Choose a tool by the evidence it can return

Semrush Position Tracking documentation describes filtering for AI Overviews and inspecting domain appearances on tracked keywords. Ahrefs’ public AI Overviews tracker offers a brand lookup over a vendor-collected index. Its AI Mode tracker has a separate lookup. Before comparing results, establish whether each tool runs your query now, returns a stored observation, or looks for a brand across a different set of queries. The vendors’ public documentation can establish their stated coverage. A purchase decision needs a trial with your own test set.

On November 21, SearchEngineAnswer entered its own domain into both free Ahrefs lookups. The AI Overviews and AI Mode previews each showed no mentions. Both screens gave two possible explanations: the brand was absent from the tool’s observations or data collection was still in progress. Neither preview supplied the same fixed query set as a controlled manual run. That result belongs in a vendor-coverage note. It cannot be counted as a zero for every Google query. AI Overviews lookup AI Mode lookup

Ask for sample exports at the query level. Look for the run time, market, device, surface, feature status, and sources. If a vendor only shows a score and a chart, use it as a discovery signal and keep your own captures for decisions. A manual browser lookup gives you one visible result. A vendor snapshot covers its own sample and may use a different collection time. Two outputs can differ while both accurately describe their observations.

Set up a simple trial with the same fixed list in each tool. Report the number of usable records before comparing how often the tools agree. Annotate a disagreement when you can identify a timing difference, unsupported locale, incomplete citation list, or redirect. Leave unresolved disagreements visible and repeat the check. SearchEngineAnswer’s SERP Test can help set up a consistent market and device for a manual search.

Reconcile the results with Search Console

Google’s report records eligible AI-feature impressions for a verified property. It combines supported features such as AI Overviews and AI Mode and gives you dimensions including page, country, and device. A property report cannot tell you how many answers include your domain when it never appeared. The Search Console report guide explains what gets counted.

One result might show the same URL across two days of manual checks while Google reports much deeper impressions. Your fixed sample may capture a few long-tail searches while the page appears in many other answers. The opposite can also happen. A few carefully chosen sample queries may show links on a day when the monitored pages have very little overall impression activity.

Review any material change in three steps. Read the query log to see which searches changed. Open the cited page and the current source to check what each answers. Then inspect the corresponding pages in Google’s property report, allowing for data delay and aggregation. If the issue is business impact, examine attributable visits and outcomes in a separate report. Our AI visibility measurement crosswalk maps those data sources to the questions they can answer.

For a first weekly report, keep the raw evidence for one query per intent, the scheduled and completed counts, and one specific action. A change in source URL might warrant a redirect check. A repeated zero from an unobserved query might warrant a tool support question. A visible citation to the intended page might need no change at all.

Keep learning

Continue this topic

Community discussion

Discuss: How to track Google AI Overviews and AI Mode

Have a question, a useful example, or a different perspective? Join the discussion, share evidence, and help other readers reach a better answer.

0 replies Moderated
No replies yet.

Be the first to ask a focused question, share a practical example, or add useful evidence.

Ask a question or join the discussion

Share evidence, a useful example, or a clear question. Be specific, stay on topic, and challenge ideas without attacking people. First-time replies may be held for moderation.