> ## Content Index
> Fetch the complete content index at: https://www.bushletter.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Three ways to check if AI tools mention your business
- URL: https://www.bushletter.com/three-ways-to-check-if-ai-tools-mention-your-business/
- Published: 2026-08-18T23:00:00.000Z
- Updated: 2026-08-18T22:59:59.000Z
- Description: Your Google Analytics referral report almost certainly has a number in it that is wrong, not because the tool is broken, but because it is measuring the wrong thing.
- Author: Editor
- Tags: Marketing, AI, Business

![Diana Trent](https://res.cloudinary.com/dz77sb7j1/image/upload/v1774262576/bushletter/authors/diana-trent.png)

By **Diana Trent** · 2026-08-04

TLDR

AI assistants crawl websites voraciously but send back almost no visitors, with Claude fetching nearly 71,000 pages for every single referral. A Comscore study found ChatGPT generates outbound clicks in just 5.2 per cent of sessions. Three repeatable checks give business owners a workable picture of their AI visibility.

KEY TAKEAWAYS

01Claude made nearly 71,000 HTML page requests for every single referral it sent back in June 2025.

02ChatGPT produces outbound clicks in only 5.2 per cent of conversation sessions, far below traditional search.

03Mistral bucked the trend, sending ten referrals for every crawl request, a ratio of 0.1 to one.

04Referral counts alone badly understate AI mentions because assistants answer without sending users to source sites.

05Three checks, direct prompting, referral audits and crawler verification, form a repeatable baseline for owners.

## The number in your analytics is not what you think

Your Google Analytics referral report almost certainly has a number in it that is wrong, not because the tool is broken, but because it is measuring the wrong thing. When an AI assistant mentions your business, your product or your advice, it almost never sends the person reading that answer to your website. The referral count stays at zero. The mention happened anyway.

The scale of that gap has now been measured in concrete terms. For the week of 19 to 26 June 2025, Anthropic's Claude made nearly 71,000 HTML page requests for every single HTML page referral it sent back to publishers.[\[1\]](https://blog.cloudflare.com/ai-search-crawl-refer-ratio-on-radar/?ref=bushletter.com) Cloudflare published those figures using its Radar data platform. The methodology is straightforward: divide the total HTML page requests made by an AI crawler's user agent by the number of those requests that included a Referer header pointing back to the platform. Cloudflare flags one caveat worth knowing: visits sent by Claude's own app carry no Referer header at all, so the real gap is narrower than 71,000 to one, though still enormous.[\[1\]](https://blog.cloudflare.com/ai-search-crawl-refer-ratio-on-radar/?ref=bushletter.com)

The problem compounds inside the chat window itself. A Comscore clickstream study found that ChatGPT produces outbound clicks in only 5.2 per cent of conversation sessions, far below the referral rates of traditional search engines.[\[3\]](https://arxiv.org/abs/2607.07652?ref=bushletter.com) When someone asks ChatGPT about accounting software or local plumbers, the assistant answers and the conversation moves on. The user does not click through. Your referral count stays flat. Your brand was surfaced regardless.

Not every AI platform behaves this way. Mistral is the clear outlier: over the same June measurement period, Mistral's AI crawler sent ten times as many referrals as it made crawl requests, a crawl-to-referral ratio of 0.1 to one.[\[1\]](https://blog.cloudflare.com/ai-search-crawl-refer-ratio-on-radar/?ref=bushletter.com) That is a fundamentally different commercial relationship with publishers than what Claude or ChatGPT currently offer, and it matters when you are reading your own traffic data.

## Check one: ask the assistants the buying questions

The most direct audit starts with the tools themselves. Open ChatGPT, Claude and Google's Gemini in separate signed-out or private sessions, signed-out removes personalisation that can skew what the assistant surfaces. Frame your prompts the way a customer would, not the way you would search for yourself. "Who are the best \[your category\] businesses in \[your city\]?" and "What should I look for when choosing a \[your product type\]?" are more revealing than a direct brand-name query.

Run each question at least three times across different sessions. AI assistants draw on probabilistic outputs, so answers vary. A single run that includes your brand tells you almost nothing; three runs that consistently exclude you tells you something worth acting on. Note which competitors appear, which categories the assistant uses to frame your industry, and whether the language it reaches for matches the language on your own pages. Assistants are more likely to surface businesses whose content uses the same vocabulary as the question being asked.

Repeat the prompts monthly. The training data and retrieval behaviour of these platforms shifts, and what the assistant said about your category in June may not reflect what it says in September. Keep a simple log: date, platform, prompt, result. Three months of that data is more useful than one thorough audit that is never repeated.

## Check two: find AI referrals in your own analytics

In Google Analytics 4, referral sessions are identified when a user arrives via a link from an external third-party domain, detected through the HTTP Referer header and classified under the Referral channel.[\[2\]](https://support.google.com/analytics/answer/15567068?ref=bushletter.com) Navigate to Reports, then Acquisition, then Traffic Acquisition, and filter the channel grouping to Referral. You are looking for domains including claude.ai, chat.openai.com, chatgpt.com, gemini.google.com, perplexity.ai and mistral.ai.

The numbers you find will almost certainly be small. That is expected given the Comscore finding on ChatGPT's 5.2 per cent click-out rate. The more useful exercise is to track them over time rather than benchmark against traditional search. Set a monthly reminder and record the raw session count from each AI domain. A doubling over three months is a meaningful signal; a flat line is not cause for alarm given how rarely these platforms send clicks at all.

If your site runs through Cloudflare, the Radar dashboard gives you crawl-to-referral data broken down by AI agent. Tyler Denk, co-founder and CEO of publishing platform beehiiv, said that as AI changes how people find and consume content, publishers need real leverage, and that their partnership with Cloudflare gives creators the data and controls they need to either maximize discovery and distribution, or protect their writing and dictate their own terms.[\[4\]](https://www.cloudflare.com/press/press-releases/2026/cloudflare-and-beehiiv-introduce-ai-crawl-controls-to-help-independent-publishers-navigate-the-ai-era/?ref=bushletter.com)

## Check three: confirm AI crawlers are reaching your pages

A page that cannot be crawled cannot be cited. Before worrying about whether AI assistants are mentioning your business, confirm that their crawlers are actually getting through to your content. Check your server logs or, if you are on Cloudflare, its bot analytics panel, which separates AI crawler traffic by user agent.

The user agents to look for include ClaudeBot (Anthropic), GPTBot (OpenAI), Google-Extended (Google's AI training crawler), PerplexityBot and MistralBot. If none of those agents appear in your logs over a two-week window, your site may be blocking them through your robots.txt file, a blanket bot-blocking rule, or a firewall setting applied without realising its scope. Check your robots.txt directly by visiting yourdomain.com/robots.txt and looking for Disallow rules targeting these agents or using a wildcard covering all bots.

If you want to be found by AI assistants, those agents need read access to your content pages. Matthew Prince, co-founder and CEO of Cloudflare, said that Cloudflare is dedicated to protecting and enabling content creators, from independent bloggers to the world's largest publishers, and that as the Internet evolves, the commitment remains the same: ensuring creators have the tools they need to thrive.[\[4\]](https://www.cloudflare.com/press/press-releases/2026/cloudflare-and-beehiiv-introduce-ai-crawl-controls-to-help-independent-publishers-navigate-the-ai-era/?ref=bushletter.com) Crawl access is now a visibility decision, not just a technical default.

## How often to repeat this and what to do with the results

Run the prompt audit monthly, the referral check monthly, and the crawler access review quarterly, unless something changes on your site. AI platforms are updating their retrieval and crawl behaviour faster than most webmasters update their robots.txt files, so tie the quarterly crawler check to any significant site migration, security update or hosting change rather than treating it as a calendar formality.

Two temptations are worth resisting when you read the results. The first is treating a single positive mention as proof of broad AI visibility. The second is treating a flat referral count as evidence you are invisible to AI entirely. Given that ChatGPT generates outbound clicks in only 5.2 per cent of sessions, referral data is a floor, not a ceiling, on how often your brand actually appears in AI-generated answers.[\[3\]](https://arxiv.org/abs/2607.07652?ref=bushletter.com) The prompt audit is the more reliable signal, even if it is the less precise one.

What these three checks give you is not a perfect picture of AI visibility. They give you a repeatable baseline, and a baseline is what lets you see change over time. Claude's crawl-to-referral ratio of nearly 71,000 to one, measured for the week of 19 to 26 June 2025, is the clearest illustration yet of why referral data alone cannot tell you whether AI is working for or against your business.[\[1\]](https://blog.cloudflare.com/ai-search-crawl-refer-ratio-on-radar/?ref=bushletter.com)

SOURCES & CITATIONS

1. [AI, search, and the crawl-refer ratio, Cloudflare Radar](https://blog.cloudflare.com/ai-search-crawl-refer-ratio-on-radar/?ref=bushletter.com)
2. [How Google Analytics 4 classifies referral traffic, Google Support](https://support.google.com/analytics/answer/15567068?ref=bushletter.com)
3. [Comscore clickstream study: ChatGPT outbound click behaviour, arXiv](https://arxiv.org/abs/2607.07652?ref=bushletter.com)
4. [Cloudflare and beehiiv introduce AI crawl controls, Cloudflare press release](https://www.cloudflare.com/press/press-releases/2026/cloudflare-and-beehiiv-introduce-ai-crawl-controls-to-help-independent-publishers-navigate-the-ai-era/?ref=bushletter.com)

FREQUENTLY ASKED QUESTIONS

Why does my referral count from AI platforms stay near zero even if they mention my business?

AI assistants answer queries directly inside their own interfaces. Users rarely need to click through to a source website to get the answer they wanted. A Comscore study found ChatGPT generates outbound clicks in only 5.2 per cent of conversation sessions, meaning the vast majority of mentions produce no referral at all.

Which AI crawlers should I look for in my server logs?

The main user agents to identify are ClaudeBot (Anthropic), GPTBot (OpenAI), Google-Extended (Google's AI training crawler), PerplexityBot and MistralBot. If none appear in a two-week window, your robots.txt or firewall rules may be blocking them.

How is the crawl-to-referral ratio calculated?

Cloudflare calculates it by dividing the total number of HTML page requests made by an AI crawler's user agent by the number of those requests that included a Referer header from that platform. Claude's ratio of nearly 71,000 to one means it fetches around 71,000 pages for every one page a user actually navigated to via Claude.

Is Mistral really different from Claude and ChatGPT in how it sends traffic?

Yes. For the week of 19 to 26 June 2025, Mistral sent ten referrals for every one crawl request, a ratio of 0.1 to one, which is the inverse of what Claude and ChatGPT produce. That makes Mistral a meaningful source of referral traffic relative to its crawl activity.

![Diana Trent](https://res.cloudinary.com/dz77sb7j1/image/upload/v1774262576/bushletter/authors/diana-trent.png)

[Diana Trent](https://bushletter.com/author/diana-trent/?ref=bushletter.com)

Diana Trent writes about regulation, competition and the law as it meets technology. She reads the judgments and the regulator filings that most people skip, and finds the story in them.