How to Track AI Search Traffic

If you've searched Reddit or Google for "how do I track AI Overview visibility," you'll have noticed the same answer keeps coming up: there isn't a proper one yet. Google Search Console has no dedicated AI Overview report. Rank trackers weren't built for a search result with no fixed position. And most business owners are left wondering whether their AI search optimisation is actually working, or just a hopeful guess.

We get asked some version of this question every week by clients who've invested in AI/GEO-focused SEO and now want proof it's paying off. This guide covers what's genuinely possible to track right now, what still can't be measured precisely, and the process we use with clients to connect AI visibility to real business outcomes.

Why This Is Genuinely Difficult (Not Just a Tooling Gap)

Traditional SEO works because Google has a consistent, crawlable index and a search results page with fixed positions. You rank, or you don't, and Search Console tells you exactly how many impressions and clicks that ranking generated.

AI search doesn't work like that, for three structural reasons:

• No fixed output. Ask ChatGPT the same question twice and you can get two different answers, citing different sources. There's no "position 1" to track.
• No consistent referral data. When someone reads an AI-generated answer and later visits your site directly, that visit often lands in Analytics as "Direct" traffic, not as an AI referral — even though the AI Overview or ChatGPT response is what drove it.
• Every engine works differently. ChatGPT blends training data with live browsing. Perplexity searches the web in real time and shows its sources. Google's AI Overviews pull from the same index as regular Search but summarise rather than list. Microsoft Copilot combines Bing results with AI processing. A method that surfaces one won't necessarily surface another.

This means ranking well in Google tells you very little about your AI visibility, and the reverse is also true. You need separate tracking for each.

What You Can Actually Track in Google Search Console

Search Console still doesn't give you an "AI Overview" filter or a dedicated report — that's not a workaround you're missing, it genuinely isn't there yet. But there are two indirect signals worth watching:

1. Impressions-without-clicks patterns on informational queries. If a query starts generating a lot of impressions but a falling click-through rate, that's often a sign an AI Overview has started appearing above your listing and is satisfying the search without a click.
2. Position stability alongside CTR decline. If your ranking position for a query holds steady but CTR drops noticeably, cross-check that query in an incognito search to see if an AI Overview has appeared.

Neither confirms AI Overview presence with certainty, but tracked over several weeks across your core queries, the pattern becomes a reasonably reliable proxy.

Semrush and similar "AI visibility" platforms - if you have the money!

The tooling gap is closing faster than expected. Semrush now runs a dedicated AI Visibility Toolkit that tracks rankings on Google Search, appearances in AI Overviews, and mentions across AI platforms like ChatGPT and AI Mode for a custom set of keywords and prompts you define. It scores your brand's overall AI Visibility on a 0–100 benchmark, tracking mentions, cited pages and citations, plus which specific AI platforms reference you most and which regions drive your exposure. Coverage is still narrower than full SEO suites, though — most reviewers note the standalone toolkit tracks primarily ChatGPT and Google AI Overviews, with wider platform coverage only unlocked on higher-tier plans, and pricing sits around $99 a month per domain standalone. It's not the only option: Ahrefs runs a comparable Brand Radar feature, and newer specialists like Peec AI and Scrunch AI are worth a look if broader engine coverage or native GA4 attribution matters more to you than staying inside one platform.

Worth knowing too — Semrush's own research found 45% of marketing leaders currently can't accurately measure their brand's visibility within AI-generated answers, and only 9% have tools covering all the relevant metrics across platforms, which is a useful reality check on how early-stage this space still is, even with the new tooling

Manual Tracking: The Method That Actually Works Today - and is free

For most small and mid-sized businesses, structured manual testing is currently the most reliable way to measure AI citation frequency — not because it's cutting-edge, but because there's no mature third-party tool that covers every AI engine well yet. It's more manageable than it sounds once it's a weekly five-minute habit rather than an occasional deep dive.

Step 1: Build a query list.

Pick 10–15 queries that reflect how a real customer would ask about your services, covering:

• Direct service queries ("who provides [service] in [location]")
• Problem-based queries ("how do I fix [problem your service solves]")
• Comparison queries ("what should I look for in a [provider type]")
• Recommendation queries ("can you recommend a [service] company")

Step 2: Run each query across ChatGPT, Google AI Overviews, Perplexity and Copilot.

Record, for each: whether you appear, in what position within the answer, whether the description is accurate, and which competitors appear alongside you.

Step 3: Log it weekly in a simple spreadsheet

Tracking date, query, tool, appearance (yes/no/partial), position, accuracy, and competitors mentioned.

Step 4: Review monthly for patterns

Not week-to-week noise. AI outputs vary run to run, so a single test proves nothing — the trend over a month is what matters.

Perplexity is worth prioritising in this process because it shows its source links directly in the answer, giving you the clearest read of whether it's actually pulling from your site.

Closing the Attribution Gap: Connecting AI Visibility to Revenue

Citation tracking tells you whether AI tools mention you. It doesn't tell you whether that turns into business. That's a separate — and arguably more important — measurement problem, and it's solved through attribution, not analytics alone.

1. Add an "AI search" option to your lead source question. Whether that's a form field, a booking flow dropdown, or simply a question your sales team asks on the first call, capture "ChatGPT," "Google AI Overview," or "AI search" as explicit options. Most CRM misattribution isn't a tooling failure — it's that nobody asked.

2. Watch landing page mix, not just volume. If your FAQ pages, comparison pages, or in-depth guides start seeing disproportionate growth in "Direct" or "unassigned" traffic relative to your homepage, that's a reasonable indicator of AI-referred visits, since AI tools tend to link to specific answer-shaped pages rather than homepages.

3. Check engagement quality on those pages. Visitors arriving via an AI summary tend to already understand roughly what you do — they're further down the funnel than a cold organic click. Longer time-on-page and lower bounce rates on your key service pages, alongside rising unattributed traffic, is a supporting signal worth cross-referencing.

4. Track conversion rate on that traffic segment, not just visit volume. Anecdotally and across the client work we've seen, AI-referred visitors often convert at a comparable or higher rate than average organic traffic, because the AI has effectively pre-qualified them before they land.

5. Train your sales/enquiry team to ask and log it. A single CRM field — "how did you hear about us, AI tool included as an option" — closes more of the attribution gap than any analytics workaround.

None of these five signals is conclusive alone. Together, over a full sales cycle, they build a credible picture of AI-driven commercial impact.

What "Good" AI Search Performance Actually Looks Like

Set realistic benchmarks rather than chasing a "position one" mindset that doesn't apply here. For a service business, reasonable markers of strong AI visibility are:

• Appearing in roughly half or more of your relevant recommendation-style queries across major AI tools
• Being described accurately when you do appear
• Appearing alongside or ahead of your direct competitors, rather than being omitted
• Being cited as a linked source, not just mentioned by name

If you're starting from zero, expect the early wins to come from fixing fundamentals — consistent NAP (name, address, phone) and business description across the web, clearer service page structure, and content that directly answers the questions your customers are asking, rather than from any tracking tool itself.

The Tools Landscape (and Why We're Not Overselling It)

There's no single platform yet that comprehensively tracks AI citations the way a rank tracker tracks Google positions — and we'd rather tell you that plainly than sell you a dashboard that overpromises. What's currently useful:

• Perplexity's own source citations — the most transparent view available of what an AI tool is actually pulling from
• Google Search Console — for the indirect CTR/impression signals above
• Behavioural analytics tools (like Hotjar) — for engagement quality on AI-relevant landing pages
• Your CRM — the single most underused tool in this entire process, and the one that actually connects visibility to revenue
• Emerging enterprise AI-visibility platforms — worth watching, though most are currently priced and built for larger organisations rather than SMEs

The Bottom Line

Nobody, including the platforms themselves, has fully solved AI search measurement yet. Treat that as expected rather than a gap in your setup. The businesses getting real value from AI search optimisation right now aren't the ones with the fanciest tracking stack — they're the ones running consistent manual citation checks, asking every new lead how they found the business, and reviewing the data monthly instead of chasing a perfect dashboard that doesn't exist yet.

Start small: pick your core queries, test them weekly, add one CRM field, and review monthly. That process will still be valid regardless of which tools mature over the next year.

---