If you own the website, this has probably already landed on you. Your marketing lead forwards an article about AI search and asks how you show up in AI. Fair question. When I went looking for that answer on a real site, it was not where I expected. Analytics barely registered it. The SEO tool handed me a number with no method behind it. The AI platforms themselves told me almost nothing about the site. The data exists, but it is scattered across half a dozen sources, and one piece of it you cannot buy at any price.
I ran a full AI-visibility audit to sort out where each piece actually lives. Here is what I found, what the current research backs up, and what I would do about it. This is the technical half of a piece we wrote for marketing leaders on why being cited is not the same as being chosen. If your CMO sent you here, that one is the why. This is the how.
Four questions, and each has a different source
AI visibility is not one metric, and that trips people up. It breaks into four questions, each answered from a different place: are AI crawlers reading your site (server or CDN logs), are you cited and for which prompts (AI-visibility tools and direct testing), are you described accurately (the answers themselves and the sources they cite), and is AI sending anyone to you (analytics). Running the audit, I had to answer each on a different budget.
- Are AI crawlers reading my site? That is ingestion. It comes from server or CDN logs.
- Am I cited, and for which questions? That is visibility. It comes from AI-visibility tools and direct prompt testing.
- Am I described accurately? That is identity. It comes from reading the actual answers and the sources they cite.
- Is AI sending anyone to me? That is referral. It comes from analytics, and the number is a floor.
If you cannot answer all four, you do not know your AI visibility. You are guessing. Here is where I found each answer.
Start with what you already own
Two of the four you can answer right now without new spend, so that is where I started.
Ingestion is the easiest to see. If you are on a CDN, its bot analytics show it directly. Cloudflare’s AI Crawl Control lists which AI services are hitting your site and how hard, broken out by crawler (Cloudflare), so check what your plan includes. If you are not on a CDN, or you want the unfiltered picture, pull your server logs and filter for known bot user agents. That path costs nothing but your time. Either way, I could see exactly which AI crawlers were reading the site and how hard. Every major assistant was crawling, and one platform accounted for most of the volume, which is worth knowing before you assume the traffic is evenly spread. The logs also turned up a problem I was not looking for: thousands of requests to retired URLs from an old site structure, mostly redirecting fine, but a reminder that crawlers hold an outdated map of your site long after you have moved on.
Referral is in your analytics. If you run GA4, it now reports visitors arriving from AI answers, though I had to build a custom channel to catch every assistant’s referrer (chatgpt.com, claude.ai, perplexity.ai, and the rest). If you are still sorting out GA4, we wrote a primer on the change and why it matters. One caveat rides along with that number. It only counts people who click a live citation through to your site, and most never do. In the first four months of 2026, 68% of US Google searches ended without any click at all, and AI answers push that further (SparkToro). When an assistant recommends a brand, the person usually types the name or searches for it and lands in direct or organic. The AI referral figure is a floor. Report it without that caveat and someone upstairs will read it as the whole story.
Then there are the search platforms’ own consoles, also free, and most teams have not switched them on. Google Search Console now reports how often you appear in AI Overviews and AI Mode, though impressions only, no clicks or queries yet, and it is still rolling out by region (Google). On the Microsoft side, Bing Webmaster Tools and the free Clarity tool go further. Clarity shows the actual Copilot and Bing citations your pages earned and the retrieval queries behind them, from Microsoft’s own grounding logs rather than sampled test prompts (Search Engine Journal).
What the AI platforms will not give you
The AI-native platforms give site owners almost nothing, and the paid tools are not plugged into them either. Set aside Microsoft’s Bing tools, covered above, and OpenAI, Anthropic, and Perplexity offer no first-party citation reporting at all.
That leaves the thing everyone actually wants, citation share across ChatGPT, Gemini, and Perplexity, as the one piece you cannot get for free. That is what the paid tools sell, and it is worth knowing what you are buying. They do not get a feed from the platforms. They build a large library of prompts and run them at scale to see who gets mentioned and cited (SEMrush). It is estimation, and it has holes. Claude sits outside most of this tooling entirely, so I had to pick it up through referral data and manual prompting.
Crawled is not referred
One pattern reframed how I read all of it. Cloudflare, which sits in front of a large share of the web, tracks how often each AI platform crawls a site against how often it sends a visitor back. Across platforms, AI crawlers request orders of magnitude more pages than they return visits, and the gap keeps widening (Cloudflare). Traditional search sits far closer to an even exchange. Read the published ratios as directional rather than precise, because they move week to week and the figures circulating secondhand rarely match the source. The point holds either way: heavy crawl volume is not success. It is the cost of being training and grounding material. Whether that turns into a citation, let alone a referral, your logs cannot tell you.
Two things to protect when this goes upstairs
Both of these get flattened into a single confident number, and both matter.
First, different tools measure different things, and their numbers do not reconcile. On the site I audited, SEMrush reported share of voice and Microsoft Clarity reported share of authority. They agreed on direction and disagreed on the figure, because they are not measuring the same thing. One samples prompts at scale, the other reads Microsoft’s own grounding logs. Give the range and name the method. Do not blend them into one precise-looking metric.
Second, all of it is directional. AI answers are non-deterministic, there is no public dataset of real prompts, and most prompts are unique. Track trends, not decimal points. A dashboard that implies precision here is lying to you politely.
Where to start, and where to bring us in
You can begin this yourself, and you should. Pull your logs or your CDN’s bot analytics and your GA4 AI channel, turn on Microsoft Clarity, and check whether the Search Console AI report has reached your account. That is a first read on ingestion, referral, and Microsoft-side citations at no new cost, and it already puts you ahead of most of your competitors.
The hard part is everything after that first read. Assembling five sources into one honest picture, running prompt panels representative enough to trust, separating real signal from non-deterministic noise, and deciding what is worth paying to track continuously. That is where this gets complicated, where a confident-looking number can turn into a wrong decision, and where it helps to have people who do this for a living. It does not have to sit on your plate.
This is the work our AI services and development teams do. COLAB runs the full audit across every surface and source on your actual site, then hands back a picture you can take to leadership with the blind spots named rather than papered over. More to the point, we turn the findings into fixes: the site structure and machine-readable identity that decide whether AI describes you accurately, the technical layer that sits above your CMS and outside day-to-day content work. You get a straight answer to the question your CMO keeps asking, and a team that makes you look capable delivering it.
If AI visibility is on your roadmap, or just on your desk because someone upstairs put it there, let’s talk. Forward this to whoever owns the website, and bring us in when you want experts on your side of the table.
