Nobody can see AI search in one place. Your citations live in chat answers, your impressions live in Google's beta reports, your referral traffic hides in GA4's Direct bucket, and the crawlers reading your site show up in logs nobody opens. The measurement layer for AI search is real now, but it's fragmented across half a dozen surfaces that don't talk to each other.
The fragmentation has a price. Otterly.ai published a stark example this month: Claude drives 10.6% of their signups while Google Analytics credits it with 0.1%. A hundred-to-one attribution gap, at a company whose entire business is AI search monitoring. If they're flying that blind on one layer, what does the average SaaS dashboard miss?
And the volume being mismeasured is no longer a rounding error. Backlinko reports LLM-driven traffic to their site grew 800% year over year, and every engine keeps shipping new surfaces to measure. Meanwhile citation positions drift 40-60% month over month across platforms, per Profound's tracking, which is why a quarterly audit tells you where you were, never where you are.
This is the stack that closes the gaps: seven tools, one per measurement layer, chosen to complement rather than duplicate each other. Prices verified this week.
Key Takeaways
- AI search measurement splits into seven layers: citations, Google AI impressions, referral traffic, the Microsoft surface, crawler behavior, brand mentions, and prompt research. No single tool covers them all.
- Roughly 70% of AI referral traffic arrives with no referrer header, so GA4 alone dramatically undercounts; one vendor measured a 100x gap between real and attributed AI signups.
- The measurement floor is free: Google Search Console shipped a dedicated Generative AI report in June 2026, and Bing Webmaster Tools added AI Performance tracking in February.
- Paid tools earn their keep on diagnosis and action. A visibility score alone, in one practitioner's words, is selling opacity, not measurement.
- Backlinko's analysis found 91% of cited URLs appear in only one LLM, so single-platform tracking misses most of the picture.
Before the List: The Three Jobs Test
OK so I skipped over something important: how to evaluate any tool in this category, including ours. The most useful framework came out of a practitioner discussion about which AEO tools are worth paying for, and it splits the work into three jobs: tracking visibility, diagnosing why, and driving action. Most tools blur them, and the blur is where budgets die.
The thread's sharpest line, lightly paraphrased: a score you can't reproduce is selling opacity, not measurement. Skepticism about AEO tooling is earned; answers vary between runs, most tools run the same prompt-sampling methodology under different branding, and nobody should pay enterprise money for a number without the why underneath it.
The skeptics have a serviceable analogy too, which one commenter delivered as a small masterpiece: measuring LLM citations with prompt sampling is like wanting to know how often your friends mention you at lunch, so instead of listening at lunch, you write ten questions you think might come up and ask a magic 8-ball each one, ten times. Harsh, and not entirely wrong. Sampling is an estimate, every vendor's estimate uses different prompts and cadence, and anyone claiming to tie visibility scores directly to pipeline is, in the words of another practitioner who sat through six sales pitches in a week, most likely guessing:
The answer to the critique isn't abandoning measurement. It's a frozen, representative prompt set, repeated sampling so trends mean something, and first-party data wherever a platform offers it. The stack below is built with exactly that skepticism in mind: free first-party sources for the facts, paid tools only where they add diagnosis or action.
1. RankControl: The Citation Layer, Plus the Fix
Full stack disclosure up front: this is our product, and we built it because the three-jobs problem is real. Tracking without action is a dashboard you feel guilty about.
RankControl samples your tracked queries continuously across ChatGPT, Perplexity, Claude, Gemini, Copilot, and Google AI Mode, and logs citations with the source-level detail that answers "why them, not us." Then the same platform acts on the diagnosis: the content engine generates and publishes the pages the gaps call for, and lead capture attributes the visitors that arrive. One loop, measurement to fix.
Pricing: $1,900/mo, everything included. It's the most expensive item on this list because it's the only one that does the third job. If you only want the tracking layer, tools further down this list start at $99, and our ranked comparison of citation trackers covers those head-to-head.
2. Google Search Console: Your AI Overviews Ledger
The biggest measurement news of the year arrived quietly in June: Google shipped a dedicated Generative AI performance report in Search Console, currently rolling out in beta.
It shows impressions and pages for your appearances in AI Overviews and AI Mode, first-party data straight from the source, free. That makes it the canonical record for one narrow question: how often Google's AI features surface your pages. What it cannot show is the bigger AEO picture: unlinked brand mentions (most of them), other engines, or why a competitor got the citation. Treat it as the ledger, not the strategy.
Setup: nothing. If GSC is verified, the report appears under Performance as it rolls out to your property.
3. GA4 with a Custom AI Channel Group: The Referral Layer
GA4 is where AI traffic goes to get mislabeled. The fix is a custom channel group matching AI sources, and honest expectations about what it can't see. Google's own channel groups documentation now includes an AI assistants example, which tells you how mainstream this setup has become.
| Detail | |
|---|---|
| What to build | Custom channel group with source regex: chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com, grok.x.com |
| What it shows | Sessions, conversions, landing pages, and engagement for AI-referred visits that carry a referrer |
| The catch | About 70.6% of AI referral traffic arrives with no referrer header and lands in Direct, per Loamly's attribution research |
| Price | Free |
Worth noting: GA4 has begun rolling out a native AI Assistant channel, and practitioners report it landing unevenly, catching some assistant referrals and missing plenty. Treat it as a bonus, keep the custom channel group as the workhorse, and read both as floors rather than totals.
That catch deserves a rhetorical question: if two-thirds of the channel is invisible, why bother? Because the visible third is your calibration set. The conversion rates on attributed AI traffic (consistently multiples of organic, as we covered in our AI referral stats roundup) tell you what the invisible portion is probably worth, and a swelling Direct bucket after a citation win stops being a mystery.
We'll show you exactly where your brand stands in AI search.
No commitment. No credit card. See how ChatGPT, Perplexity, Claude, and Gemini talk about your brand today.
Get your free report→
4. Bing Webmaster Tools: The Microsoft Surface
The perennially skipped tool earns its slot twice over. Bing's index is what ChatGPT browses, and as of February, Bing Webmaster Tools ships an AI Performance report in public preview showing when your site is cited across Copilot, Bing's AI summaries, and partner integrations, with the URLs being referenced.
Worth being precise about the boundary: it reports Microsoft's AI surfaces, not ChatGPT's answers directly. But the same submission and indexing hygiene serves both, since ChatGPT's live retrieval reads Bing's index. Verify the domain, submit the sitemap, and check the AI Performance report monthly. Free, fifteen minutes, and almost none of your competitors have done it.
5. Cloudflare Attribution Business Insights: The Crawler Layer
Everything above measures outputs. This layer measures inputs: which AI bots read your site, what they take, and what they send back. Cloudflare's Attribution Business Insights, launched July 1, classifies AI bot traffic by purpose (training, search, agent), and reports crawl-to-referral ratios per operator.
The crawl-to-referral ratio is the metric to watch; it tells you whether a given AI company is a trading partner or a strip miner, and whether your llms.txt and robots.txt decisions are working. Fair warning though: this sits behind Cloudflare's paid Bot Management tier, so check current pricing against your plan. Teams not on Cloudflare have two free approximations: server-log analysis, or Microsoft Clarity's bot analytics, which practitioners have been leaning on lately since it began surfacing non-human traffic and even robots.txt violations at no cost. Cruder views, but any answer to "what did the machines read this week" beats none.
6. Profound: Brand Mentions at Competitive Depth
For dedicated share-of-voice monitoring beyond your own tracked queries, Profound covers nine AI platforms (ChatGPT, Perplexity, Gemini, Copilot, Google AI Overviews, Meta AI, Grok, DeepSeek, Claude) with agent analytics layered on top.
Starter runs $99/mo for ChatGPT only; Growth at $399/mo covers three platforms; enterprise is custom with SOC2 and SAML. The platform-count math matters more than it looks: Backlinko's tool analysis found 91% of cited URLs appear in just one LLM, so a single-platform tracker sees a sliver of your actual footprint. The engines don't even source alike; practitioners tracking citations across models report Perplexity leaning heavily on forums while Claude favors documentation pages, unverified as a study but consistent with what our own sampling shows. Cross-model variance is the norm, which is the whole argument for monitoring more than one. Alternatives worth a look: Ahrefs Brand Radar if you want social channels in the same view (pricier, from ~€358/mo), or Otterly.ai's $29 Lite tier for a budget entry.
Know exactly what AI says about your competitors.
RankControl's Recon Agent monitors competitor citations across ChatGPT, Perplexity, Claude, and Gemini. See where they show up and you don't.
See how it works→
7. Semrush AI Visibility Toolkit: The Prompt Research Layer
The last layer answers the question every other tool assumes: which prompts should you even track? Sample the wrong twenty prompts and every score upstream is noise. Semrush's AI toolkit pairs prompt tracking with its search volume data, and the free AI Search Visibility Checker is a genuinely useful zero-cost starting point: it surfaces the prompts already driving mentions for a domain.
Paid plans start at $165/mo (annual) with 50 daily tracked prompts and competitor gap analysis. Build the prompt library the way practitioners keep recommending: a frozen set of 20-30 prompts mirroring how buyers actually research your category, changed rarely, so trend lines mean something. Our take on sourcing those prompts from real buyer language is the prompt research playbook.
What Not to Buy
The anti-list matters as much as the list, so three fast rules from the practitioner trenches. Skip any tool selling "LLM search volume" as a hard number; the platforms don't publish prompt volumes, so every figure is modeled, and vendors rarely say how. Skip per-prompt or per-brand billing if you're an agency or multi-product team; the pricing model that feels cheap at one brand compounds brutally at ten. And skip anything that outputs a proprietary score without showing the underlying answers and cited URLs, because a number you can't audit is a number you can't act on. Every dollar in this stack should buy either a fact you can verify or an action you'd otherwise do by hand.
Wiring It Together
Seven tools, but the workflow is one loop, weekly:
- Prompt library (Semrush layer) defines what you measure.
- Citation tracking (RankControl) samples those prompts across engines and flags movement, with share of voice as the headline metric.
- GSC and Bing WMT confirm the impression side on Google and Microsoft surfaces.
- GA4's AI channel group catches the referred visitors; the Direct-bucket delta hints at the unattributed rest.
- Crawler analytics verify the machines can still read you after every deploy.
Total spend for a lean version: $0 on layers 2 through 5's free tiers, plus one paid monitoring tool. Total time by hand: the free layers take a monthly hour each, but the citation sampling in step 2 is the recurring sink, 3-4 hours weekly done manually, which is exactly the layer worth automating first.
The stack's real output is the ability to answer, in one meeting, the question fragmented tooling can't: are we more or less visible to AI buyers than last month, why, and what are we doing about it?
15+ content types. Published on your domain. Matched to your brand.
Guides, comparisons, listicles, case studies, and more. RankControl generates content that gets cited by ChatGPT, Claude, Perplexity, and more.
Start free→
Start With the Free Floor
If this list is overwhelming, sequence it. Week one: verify GSC and Bing WMT, build the GA4 channel group, run the free Semrush checker to draft your prompt library, and turn on whatever bot analytics your host already offers. That's the whole measurement floor for $0. Week two: add the paid layer that matches your bottleneck, monitoring if you're blind, action if you're stuck.
You can operate all seven surfaces yourself and reconcile them in a spreadsheet every Friday. Or RankControl can run the citation layer continuously, publish the fixes the data calls for, and hand you the reconciled picture, while you attend exactly zero of the meetings where someone asks what the visibility score means.
