Most advice on AI mode tracking boils down to “buy a tool that gives you an AI Mode rank.” That gets the order wrong. An AI Mode answer is a conversation that changes from run to run, so any single “position” is a sample. Start with the data Google now gives you: since June 2026, Search Console has a dedicated report for generative AI impressions. It shows which pages appeared in AI features, but not which questions put them there. A prompt sample you design fills part of that gap, as long as you know how sampling misleads.
What AI Mode Tracking Can and Can’t Tell You
Split the problem into three layers: eligibility (is your content allowed to appear?), exposure (do your pages get shown and cited?) and outcome (does that exposure turn into clicks and revenue?). Each layer has its own data source, and most confusion comes from using one layer’s number to answer another’s question.
- Eligibility comes from Search Console settings, your robots.txt rules for Googlebot and your snippet controls. You can check it exactly.
- Exposure comes from Search Console’s generative AI report (first-party, page-level, no queries) and from prompt sampling (query-level, but a sample).
- Outcome comes from Search Console clicks and GA4. It’s real data, but it’s mixed in with classic organic search.
Nothing here gives a complete “AI Mode rank” for every query. Anyone selling one is selling an estimate.
How Search Console Counts AI Mode Traffic
Google’s Performance report help page sets the rules. A click on an external link in AI Mode counts as a click, standard impression rules apply, and position “follows the same methodology as a Google Search results page.” The rule with the biggest consequence: a follow-up question in AI Mode is treated as a new query, and the impressions, positions and clicks in that response belong to it.
Google’s AI features documentation adds that this traffic is reported under the “Web” search type, and the Performance report help page notes that Search Labs experiments are excluded. Glenn Gabe at GSQi tested it in practice. On desktop, links in the answer module can be reported as position 1 or after the blue links, depending on layout. On mobile, the module is a carousel, so its links share one position. Follow-up answers restart positions, and nothing in GSC marks a query as a follow-up.
So your average position now blends classic results, AI Overviews and AI Mode conversations, plus a long tail of queries that are really second and third turns. Treat a sudden shift in position or query count as a prompt to check AI Mode, not as a ranking verdict.
The Generative AI Performance Report: Your First-Party Baseline
On June 3, 2026, Google’s Search Central blog announced Search Generative AI performance reports in Search Console, starting with a test group of sites; the help documentation says they reached all websites worldwide as of August 31, 2026. The Search report covers AI Overviews and AI Mode. Its one metric is impressions, broken down by pages, countries, devices and dates (by day, week or month), with a filter for text-based versus multimodal searches such as Lens.
Three details change how you read it:
- AI Overviews and AI Mode are combined. Google’s documentation doesn’t describe a way to separate the two, so this report can’t tell you which feature showed your link.
- No query dimension is listed. You see which pages showed up, not which questions triggered them.
- Aggregation differs by view. The chart counts by property, so two of your URLs in one answer count as one impression. The page table counts by page. That’s why the chart and table totals don’t match.
The most useful number you can build here is a page-level AI exposure ratio: a page’s generative AI impressions divided by its total Web impressions in the main Performance report, over the same dates. Both tables count by page, so the comparison holds. A page where AI features drive a large share of impressions while clicks fall is your first rewrite candidate. A page with strong classic rankings and no generative AI impressions has a content-shape problem, not a ranking problem.
Check the Switch Before You Measure Anything
Alongside the reports, Google added the Search generative AI control (Settings in Search Console). Include is the default; Exclude removes your links and content from AI Overviews, AI Mode and Discover’s generative AI features, along with their impressions and traffic. According to Google’s help page, the control isn’t used as a ranking or inclusion signal for the rest of Search. It’s also separate from Google-Extended, which governs AI training. Exclusion usually takes effect within one to two days, though caching can make it take longer.
Check it first in any audit: a setting someone flipped months ago explains “we vanished from AI Mode” faster than any tracker. Then check the older levers in Google’s AI features guide. A page needs to be indexed and eligible for a snippet to be a supporting link, so nosnippet, max-snippet and data-nosnippet all limit what AI Mode can show. If Googlebot is blocked, nothing else matters. Our free AI Crawler Robots.txt Checker shows which AI bots your robots.txt allows. Remember that Google’s AI features use Googlebot, so blocking GPTBot or ClaudeBot has no effect on AI Mode.
Why AI Mode Rank Tracking Is a Different Measurement
Classic rank tracking assumes a stable results page per keyword, location and device. AI mode rank tracking breaks that in three ways.
Fan-out. Google’s help page says AI Mode splits a question into subtopics and searches for each one at the same time. Your citation depends on ranking for those hidden sub-queries, not just the prompt you typed. Our guide to query fan-out explains how to infer them.
Personalization. Google says AI Mode personalization is available to users 18 and over who have Search history and personalized recommendations turned on. A signed-out tracker measures a clean answer many real users never see.
Conversation. Each follow-up is a new query with new links. A tracker that checks only the first answer misses the turns where buying decisions often happen.
None of this makes trackers useless. Their output is a rate across a sample, and should be reported as one.
Build a Prompt Panel That Samples Honestly
Semrush’s Position Tracking, for example, lets you pick Google AI Mode as the engine and reports mentions, cited pages and an average AI Mode position; our roundup of AI visibility tracking tools compares the alternatives. Either way, design the panel yourself:
- Write prompts as people ask them. Use full questions and comparisons (“best X for Y under Z”), not two-word keywords. Mix informational, comparison and purchase intent.
- Anchor prompts in real demand. Start from your Search Console queries and keyword data, not a brainstorm, so the panel reflects questions people actually ask.
- Fix the conditions. Keep the country, language and device the same for every check. For manual checks, open google.com/ai in a signed-out window so your own history doesn’t skew the answer.
- Repeat each prompt. Run each prompt several times per cycle and record how often you appear, not just whether you appeared once. One run is an anecdote.
- Log the follow-up. For your highest-value prompts, ask the obvious next question and record citations on that turn too.
The panel answers what Search Console can’t: which questions show your brand, and who gets cited when you don’t. See LLM prompt tracking for panel size and refresh schedules.
The Metrics Worth Reporting
Drop “average AI Mode position” as a headline number. Positions reset on follow-ups and depend on layout, so the average reflects panel design as much as visibility. Report these instead:
- Inclusion rate: the share of prompt runs where your domain is cited as a link.
- Mention-without-link rate: runs where the answer names your brand but cites someone else. Each one is a citation outreach opportunity (see AI citation outreach).
- Cited-URL spread: which of your pages earn citations. If one page carries everything, you depend on it.
- Competitor co-citation: which domains appear when you do, and which appear when you don’t. This is the input for share of voice in AI answers.
- Generative AI impressions by page: the first-party number from Search Console, trended weekly.
When panel inclusion and Search Console’s generative AI impressions move together, trust the trend. When they diverge, your panel probably doesn’t match real searches, so fix the prompts.
Connecting AI Mode Exposure to Traffic
In May 2025, Search Engine Journal reported that AI Mode links briefly carried a noreferrer attribute, so GA4 logged the visits as direct traffic. Google’s John Mueller said “it looks like a bug on our side,” and it was fixed within days. AI Mode clicks now arrive as ordinary Google organic traffic, which GA4 can’t separate from classic search visits.
So triangulate instead of filtering. For the pages your panel shows being cited, and those with rising generative AI impressions, watch Search Console clicks, GA4 engaged sessions and organic landing-page conversions. One more proxy: because follow-ups count as new queries, a regex filter for long, conversational queries in the Performance report (for example ^(\S+\s+){7,}\S+$ for eight words or more) brings up the question-style searches that AI Mode tends to create. It’s a heuristic, not an AI Mode filter, so present it that way. Our guide to AI traffic attribution covers the non-Google engines, where referrers behave differently.
Where SEO Rocket Fits (and Where It Doesn’t)
To be straight about it: SEO Rocket doesn’t run an AI Mode-specific tracker. Its AI Visibility feature, built on Ahrefs Brand Radar data, counts how many AI responses across ChatGPT, Google AI Overviews, Gemini and Perplexity cite your site, and shows the real questions where you’re cited (or the questions people ask in your space if you aren’t yet). The AI Overviews and Gemini data sit usefully beside your AI Mode panel, and the question list is a ready source of prompts.
The rest of the routine is where SEO Rocket does the work directly. The Search Console integration lets you ask in chat which pages gained impressions while losing clicks. Rank tracking covers classic positions for the fan-out sub-queries your panel suggests. The real-crawler site audit catches robots.txt blocks and indexing problems that make a page ineligible, and client dashboards put first-party numbers and panel results side by side. It’s the discipline behind a playbook proven across 1,000,000+ ranking pages: trust first-party data, label every sample, act on trends rather than single runs.
A Monthly AI Mode Tracking Routine
A cadence that fits a small team or an agency retainer:
- Weekly: check the generative AI performance report for page-level impression shifts. Note any page that gains or loses a lot.
- Monthly: run the full prompt panel with repeats, log inclusion and co-citation, and recompute the AI exposure ratio for your top pages.
- Monthly: compare panel trends against Search Console clicks and GA4 conversions for the cited pages.
- Quarterly: refresh the prompt panel from new Search Console queries, re-check the generative AI control and snippet directives, and re-audit robots.txt.
Fold the results into your AI visibility reporting, next to the engines in AI visibility tracking and Gemini visibility. For what to change once the numbers arrive, see how to optimize for Google AI Mode.
AI Mode Tracking FAQ
Can I see AI Mode traffic separately in Search Console?
Not entirely. The generative AI performance report shows impressions for AI Overviews and AI Mode combined, by page, country, device and date. Clicks stay mixed into the Web search type of the main Performance report. Google’s documentation doesn’t describe a filter that separates AI Mode from AI Overviews.
Is Google AI Mode tracking accurate in third-party tools?
It’s accurate for what it samples: one location, device and usually signed-out session, at one moment. Real users get personalized, conversational answers a tracker doesn’t reproduce. Use tools for trends and competitor comparisons, not as a census.
Does blocking Google-Extended remove me from AI Mode?
No. Google’s crawler documentation says Google-Extended does not affect inclusion in Google Search; it governs Gemini training and grounding in some other products. For AI Mode, the levers are the Search generative AI setting in Search Console, snippet directives and noindex.
AI Mode measurement will never be as neat as a rank report. Start with Google’s own impressions, add a prompt panel you label honestly, and judge the result by clicks and conversions on the pages that get cited.
