How to Tell If Your Website Shows Up in AI Search Results

Yes, you can check, and the useful checks are free. There are four: Google Search Console's generative AI performance report, AI referral traffic in your analytics, your crawler logs, and asking the assistants yourself. Each tells you something real, and each has a genuine hole in it. Run them in that order and you'll have a reasonable picture of your AI search visibility within an afternoon. What you won't have is a complete one, because a meaningful part of this isn't measurable by anyone at any price right now, and the tools charging monthly to solve that are mostly selling you a proxy.

This is the fifth piece in our AI search optimization series. The first covered the AEO and GEO hype generally, the second looked at whether an llms.txt file is worth building, the third walked through how AI search actually crawls and cites content, and the fourth asked whether AI Overviews are really going to cost you traffic. This one closes the loop: how do you know whether any of it is happening to your site?

Check 1: Search Console's generative AI performance report

Start here, because it's the only first-party data that exists. Google announced a dedicated Search Generative AI performance report in Search Console on June 3, 2026, and it's the first time AI feature visibility has been separable from everything else.

Open Search Console, go to the Performance section, and look for the generative AI report. It shows impressions in AI Overviews and AI Mode, which Google defines as how many times links to your site were shown to someone inside a generative AI feature on Search, broken down by page, country, device, and date. A matching report covers generative AI features in Discover.

Three things to understand before reading anything into the numbers:

• It's impressions only. No clicks, no click-through rate, no position, and no query data. You can see that you appeared. You cannot see whether anyone came, or what they'd asked to find you.
• It's a breakout, not new data. Google notes the report draws on data already inside the Web search type of your main Performance report. Those impressions were always in your totals. They just weren't separable until now, which is worth remembering if you're tempted to read the new line as new traffic.
• Not everyone has it yet. Google's own wording is that "not all properties have access to the report, as we're rolling out over time." Your site also needs enough impressions in AI features to show anything at all, so an empty report can mean either "you don't have the feature yet" or "you're not appearing," and you can't always tell which.

If you have access, look at it now even if you don't plan to do anything with it yet. You can't analyze a period you weren't watching, and this is the baseline every later comparison depends on.

One note on interpreting a decline, since this is where owners tend to spiral. Impressions in AI features rising while clicks fall is the pattern the whole industry is arguing about, and we went through what the research actually shows in the traffic piece earlier in this series. Google's own position, for whatever it's worth coming from Google, is that clicks from search pages with AI Overviews tend to be higher quality, with visitors more likely to spend real time on the site.

Check 2: AI referral traffic in your analytics

This is the check most articles lead with. It belongs second, because the number it gives you is wrong in a specific and predictable direction.

In Google Analytics 4, go to Reports, then Acquisition, then Traffic acquisition, and switch the dimension to session source or session source/medium. Look for chatgpt.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com. To make it a standing view rather than a one-off dig, build a custom channel group under Admin, Data display, Channel groups matching those domains, and order it above Referral so AI visits stop getting mixed into general referral traffic.

Now the problem. Several assistants strip the referrer, so those visits land in your analytics as Direct and you never learn they came from AI. Ahrefs tested this hands-on, checking what each assistant actually passes when you click a link out of it. ChatGPT strips it on in-content links for paid accounts using a noreferrer attribute, though free accounts and list-style links pass it. Gemini AI Mode was stripping it as a confirmed bug. Grok, DeepSeek in most cases, and Copilot on Windows also strip it. Claude, Mistral, Meta AI, and Perplexity on the web pass it through.

So your AI referral number is a floor, not a measurement. You'll see confident percentages online claiming exactly how much AI traffic hides in Direct. We won't repeat one, because the figures circulating don't trace back to a methodology anyone can check. The mechanism is real and documented. The size of the gap isn't known. Ahrefs' own caveat applies too: traffic has always leaked into Direct for ordinary reasons, and this isn't unique to AI.

Treat the trend as more informative than the absolute number, and don't conclude you're invisible in AI search because this report reads near zero.

Check 3: Whether AI crawlers can reach you at all

Nobody talks about this one, and it separates two completely different problems. "AI systems can't access my site" and "AI systems access my site and don't cite it" look identical from the outside and need opposite fixes.

If your site is on Cloudflare, you already have this. Open your site in the Cloudflare dashboard and go to AI Crawl Control, previously called AI Audit. Its Overview, Crawlers, and Metrics tabs show which AI services have been fetching your content and how often. It's included on every plan including the free one, where it identifies crawlers by user agent, so it catches the well-known bots that identify themselves honestly. If you're not on Cloudflare, your host's raw server logs give the same answer with more effort: search them for the bot names.

Which bots matters, because they aren't interchangeable. Training crawlers and search crawlers are separate and separately controllable, which we broke down in the piece on how AI search finds and cites content. If OAI-SearchBot and Claude-SearchBot show up regularly, your content is reachable and citation is a content question. If nothing shows up, you have a technical problem, and no amount of content work fixes it until the door is open.

Check your Cloudflare bot settings while you're in there. Defaults in this area have been changing, and a bot you assumed had access may not.

Check 4: Ask the assistants, carefully

Do this last, and hold it loosely. It's the check everyone starts with and the least reliable one you'll run, for reasons worth understanding before you trust the result.

These models generate text probabilistically, so the same prompt genuinely returns different answers on different runs, in different sessions, for different people. Personalization, what the model retrieved that moment, and even the earlier questions in the same conversation all shift the result. One query proves nothing in either direction.

That's the problem our free AI Visibility Grader was built to work around. You give it your website and confirm the question a customer would actually type, and it asks Claude with web search enabled, three separate times, then shows you all three answers: who got named, whether your site was cited, and where your page fell short. It also scores five things that decide whether an assistant can name you at all, which are business clarity, content extractability, question coverage, structured data, and crawlability.

Two honest limits on that, which we put in the tool itself rather than the fine print. It asks Claude, not ChatGPT or Gemini or Perplexity, because those are separate systems with separate indexes that will answer differently, and pretending otherwise would make a normal difference between two models look like a broken tool. And three runs is a small sample. Read the pattern across the three, not any single answer. The report prints the exact question it asked so you can run it yourself and check our work.

To check the other assistants by hand, the same rules apply:

• Ask buyer questions, not brand questions. "Who's a good web designer in Orangeville" tells you something. "Tell me about Fheba Design" tells you almost nothing, because you've handed the model the answer.
• Log out, or use a private window. Your own logged-in ChatGPT account has memory and months of your chat history behind it. It is the single most biased test environment available to you.
• Run several phrasings. Perplexity is the easiest place to start because it lists numbered citations openly. Then repeat in ChatGPT, Gemini, and Google's AI Overviews.
• Write down what you find. A spreadsheet with the prompt, the tool, the date, and whether you appeared beats the memory of one good result.

Ten queries across a few tools, repeated monthly, is a signal. One is an anecdote.

What you can't measure, at any price

Being clear about the limits saves money. Here's what nobody can currently tell you:

• Which questions you appeared for. Search Console gives AI impressions with no query data attached. There's no AI equivalent of a keyword report.
• What an AI answer said about you when it didn't link you. Being described without being cited is invisible, and it's probably common.
• Whether an AI answer influenced a customer who arrived some other way. Someone reads an AI summary mentioning you, searches your name later, and lands via Google. That's an AI-influenced conversion recorded as branded search.
• A stable, repeatable "AI ranking." There isn't one. Position doesn't exist as a concept here the way it does in a list of ten blue links.

Any tool claiming to solve these is estimating, usually by running prompts on a schedule and reporting how often you turned up. That's a reasonable proxy. It isn't the same thing, and it's worth knowing which you're buying.

Should you pay for an AI visibility tracker?

For most small businesses right now, no. Here's the arithmetic.

Dedicated AI visibility tools generally start around $99 USD per month for a single domain and one assistant, with Perplexity and Google AI Overviews coverage typically landing on a higher tier closer to $399 USD per month. Some, like Ahrefs' Brand Radar, come bundled if you already pay for the parent tool. These are approximate market ranges rather than standardized rates, the category is repricing constantly, and advertised prices commonly exclude per-engine coverage, extra seats, and API overages. As a Canadian buyer you're absorbing the exchange rate on top.

Compare that to what's being measured. Across 81,947 sites Ahrefs studied in mid-2025, AI traffic averaged about 0.25% of total site traffic, having grown roughly 9.7 times year over year. An earlier study of 3,000 sites found 63% received at least one AI visit, at around 0.17% of visitors. Both are 2025 datasets on a steep growth curve, so treat the percentages as directional rather than current, and both sets of authors stress the figures are undercounts for the referrer reasons above.

One real counterweight. When Ahrefs looked at its own 30 days of data, AI search was 0.5% of visitors but 12.1% of signups, converting far better than organic search visitors. That's a single-company study of a software product with a technical audience, and the same analysis found AI search users click links far less often overall, so don't assume it transfers to a local service business. It does suggest the traffic can be worth more per visit than its share implies.

A threshold falls out of that. A monthly tracker earns its cost when AI referrals are already a visible line in your analytics rather than a rounding error, when you're in a category where people genuinely ask assistants for recommendations before buying, or when a competitor keeps surfacing in answers you don't and you need to know whether that's changing. Below that, the subscription costs more than the traffic it measures is worth, and the free checks above tell you the same thing.

If what you actually want is the thing those tools lead with, a real answer from a real assistant with your weak spots named, that's what our AI Visibility Grader does at no cost and without an account.

What to actually do with what you find

If the checks come back thin, resist the urge to buy a solution to a measurement problem. The fix is almost never AI-specific. It's the same work we landed on in the llms.txt piece and everywhere else in this series: make sure the crawlers can reach you, answer real questions clearly and specifically on pages built for that purpose, and get the SEO fundamentals genuinely right rather than approximately right. AI search draws on the same crawlable, indexed content everything else does, and Google states plainly that there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode.

If you'd rather hand that work to someone, it's what our AI search visibility service covers, and it's the same list either way.

Check quarterly, not weekly. The numbers are small and noisy enough that watching them closely will tell you nothing except that they're small and noisy.

---

Want to skip straight to the answer? Our free AI Visibility Grader asks Claude the question your customers would ask, three times, and shows you exactly what came back along with the five things holding your page down. No account, no credit card.

And because most AI visibility problems turn out to be ordinary website problems wearing a new label, our Website Grader checks the fundamentals underneath.