TL;DR
Use Search Console to surface question queries, near-miss positions, and pages losing clicks to answer features. Use SEMrush to validate demand with keyword gaps, answer-feature filters, and competitor page teardowns. Configure GA4 with a custom channel group for assistant referral domains, then judge that traffic on engagement and conversions rather than volume. Merge the exports into topic clusters, add a manual citation test, score each cluster, and pick five per quarter. Report six lines monthly and keep the prompt list identical every time.
What can these three tools actually tell you about AI search?
None of them has a clean "AI search" report. That is the honest starting point. What they do have is enough signal to triangulate.
Think of it as three different views of the same funnel. Search Console shows demand and how search systems interpret your pages. SEMrush shows the competitive field and which result formats appear. GA4 shows what happens after someone arrives.
Put together, they answer four questions: which questions are people asking, which of those questions trigger answer-style results, whether AI surfaces are sending you anyone, and whether those visits are worth anything.
You will not get a precision dashboard. You will get a reliable shortlist of opportunities, which is what a roadmap actually needs.
How do you use Search Console to find answer opportunities?
Search Console is the most underused of the three for this purpose. Start with four moves.
Filter for question queries
In the performance report, filter queries containing what, how, why, does, should, is, can, and best. Sort by impressions. You now have a list of conversational queries your site is already visible for, ranked by demand.
Pages earning question-query impressions without matching question headings are the fastest wins in this whole process. The demand exists and the page is close. It just is not shaped like an answer.
Find high-impression, low-click pages
Sort pages by impressions and look for low click-through rates in decent positions. Some of that gap is an AI Overview or featured snippet absorbing the click. Some is a weak title. Either way it flags a query worth inspecting manually.
Watch position bands, not just averages
Terms sitting in positions eight through twenty are your realistic near-term targets. Export them, group them by topic, and look for clusters where three or four related queries all hover just off the first page. Clusters beat single terms because one restructured page can lift all of them.
Compare periods deliberately
Use a 28-day comparison rather than eyeballing the trend line. If impressions are rising while clicks stay flat on informational pages, that is a common signature of answer surfaces taking the click. That is not necessarily a failure. It may mean you are being read without being visited, which changes how you measure success.
What should you look for in SEMrush?
SEMrush is where you check the competitive shape of an opportunity before committing to it.
Keyword gap analysis. Compare your domain against three to five competitors and filter for terms where two or more of them rank and you do not. Shared gaps are stronger evidence than any single competitor's ranking.
SERP feature filters. Filter keyword lists by the presence of featured snippets, People Also Ask, and AI-style results. A query with an answer feature already showing is a query where structured, answer-first content has somewhere to land.
Question keyword reports. Pull question-form keywords for your topics and cross-reference them against your Search Console question list. Overlap means validated demand. Terms that appear in SEMrush but not Search Console are content gaps.
Competitor top pages. Look at which competitor pages earn the most estimated traffic, then look at how those pages are built. Format usually explains more of the result than volume does.
Position tracking with intent tags. Set up tracking for your priority cluster and tag by intent. Watching informational and commercial terms separately makes the reporting far more useful than one blended visibility score.
The output of this step should be a short list of clusters where demand is real, an answer feature exists, and the incumbents are beatable. That is the same filtering logic I use in competitor keyword analysis for SEO, AEO, and generative search.
How do you set up GA4 to see AI-driven traffic?
Out of the box, GA4 lumps assistant referrals into generic referral traffic. A little configuration fixes that.
Create a custom channel group, or an exploration with a filter, that isolates referrals from AI assistant domains. Include the chat and search domains for the assistants your audience uses, and keep the list in a documented place so it can be updated as new ones appear.
Then compare that segment against organic search on four measures: engagement rate, average engagement time, pages per session, and key event completion. Do not compare volume. AI referral volume is small by nature and comparing it to organic search will make it look irrelevant.
What I typically see is lower volume with noticeably higher engagement and a shorter path to conversion. A professional services client found AI referrals at roughly two percent of organic sessions, converting at close to three times the rate. Judged on sessions it looked like noise. Judged on outcomes it was one of the best performing sources on the site.
Two setup details make or break this. Key events need to be defined and marked as conversions, and landing page reports need to be usable so you can see which specific pages AI surfaces are sending people to. If either is missing, fix that before drawing conclusions. This is the groundwork behind my analytics and reporting work.
How do you combine all three into one opportunity list?
Here is the workflow I run, in order.
- Export question queries and near-miss positions from Search Console.
- Export keyword gaps and answer-feature keywords from SEMrush.
- Merge them into one sheet and group by topic cluster rather than individual keyword.
- For each cluster, note demand, whether an answer feature exists, which competitor currently wins, and which of your pages is closest.
- Run ten to fifteen of the cluster's questions through the assistants your buyers use and record who gets cited.
- Pull GA4 engagement and conversion data for the pages already in each cluster.
- Score each cluster on demand, difficulty, answer opportunity, and business value. Sort. Pick the top five.
Then decide the action for each cluster: restructure an existing page, consolidate two overlapping pages, or write something new. Restructuring wins more often than teams expect, because an existing page with impressions and links has a head start over a new URL.
Keep the list to five items per quarter. A long list of possibilities is easier to produce and much harder to execute.
What should the monthly report include?
Six lines, and they should fit on one screen.
- Impressions and clicks for your target clusters, from Search Console
- Position changes for tracked priority terms, from SEMrush
- Count of priority queries showing an answer feature, and whether you own it
- Citation rate from your fixed monthly prompt test
- AI referral sessions, engagement rate, and key events from GA4
- Organic conversions, so the foundation stays visible
Use the same prompt list every month. Changing the prompts changes the number, and then you are measuring your own inconsistency. My post on GEO vs AEO vs SEO explains why these three layers need separate lines rather than one blended score.
What pitfalls should you avoid?
Over-attributing. Some AI referral traffic never carries a referrer, so your reported numbers are a floor rather than a total. Say so in the report.
Comparing AI volume to organic volume. Different scale, different job. Compare quality instead.
Chasing every question. If a query has no commercial connection to your services, answering it well earns you a visit and nothing else.
Skipping the manual prompt test because it feels unscientific. It is the only direct look you have at generative visibility, and a consistent method makes it useful enough to act on. Pair it with the structural work in GEO and AEO optimization and you have a loop that improves month over month.
AI search measurement FAQ
Can you track AI search traffic in GA4?
Partially. Create a custom channel group or exploration filter that isolates referrals from AI assistant domains. Some AI referrals arrive without a referrer, so treat the number as a floor rather than a complete count.
How do you find answer engine opportunities in Search Console?
Filter the performance report for queries containing what, how, why, best, and similar question words, then sort by impressions. Pages earning question-query impressions without matching question-style headings are the fastest opportunities to restructure.
Which SEMrush reports help with AI search?
Keyword gap analysis against three to five competitors, keyword filters for featured snippets and other answer features, question keyword reports, competitor top pages, and position tracking segmented by search intent.
Should AI referral traffic be compared to organic traffic volume?
No. AI referral volume is small by nature. Compare engagement rate, average engagement time, pages per session, and key event completion instead, since AI-referred visitors often arrive with higher intent.
What should a monthly AI search report include?
Impressions and clicks for target clusters, position changes for tracked terms, the count of priority queries showing an answer feature and whether you own it, citation rate from a fixed prompt test, AI referral sessions with engagement and key events, and organic conversions.