ChatGPT is sending people to websites. Is your business visible when AI recommends something?

A customer can ask an AI assistant for a recommendation, see your business mentioned and still never visit your website. They can click through and leave a few seconds later. Or they can arrive, find what they need and contact you. Those are different outcomes, even though they all begin with the same AI answer.

That distinction matters because recent figures about ChatGPT sending people to websites have attracted rather larger conclusions than the data supports.

What BrightEdge's 95.1% figure actually says

BrightEdge, an SEO and analytics company, reported on 24 September 2026 that ChatGPT accounted for 95.1% of AI-platform referral traffic in its measured August dataset. It also said its measured ChatGPT referrals had risen 101% between January and August, with a 36% month-on-month rise in August. Those are BrightEdge's observations of traffic referred from AI platforms in its own dataset. They are not ChatGPT's share of search, internet traffic, AI users or AI conversations.

BrightEdge's 24 September 2026 referral-data announcement →

The denominator is the important part. In a separate chart, BrightEdge says AI search accounted for 1.9% of total search referrals in August, while Google accounted for 86.9%. These are values from BrightEdge's own measurement, not a census of every website. The two figures also describe different groups: ChatGPT's 95.1% is within AI-platform referrals; the 1.9% is AI search within total search referrals.

BrightEdge does not publish enough detail on those pages to establish the sample's size, industry mix, geography or exact referral-attribution rules. It also notes that it revised historical data in July 2026 after changes in analytics reporting affected its ChatGPT figures. The numbers are interesting, but they should be read as one company's measured dataset, not as a forecast for your own website.

BrightEdge AI Market Pulse and its referral-share chart →

Being mentioned is not the same as being visited

An AI answer might name a company without showing a link. It might show a link that nobody clicks. A click might become a visit, but the visitor may not read the page or make an enquiry. Analytics can help with the clicked visit; they cannot, on their own, tell you every time an AI system mentioned your business without sending someone over.

A useful way to keep the stages separate is:

Visibility → Visit → Engagement → Enquiry → Customer

Each step needs its own evidence. A citation or mention is a visibility signal. A tagged referral in analytics is evidence of a visit. Time on a useful page, a form submission, a phone call or a sale are later steps. Do not report one as if it proves the next.

Matomo's guide to identifying AI-chatbot referrals in web analytics →

Where to look in your analytics

Start with the traffic-acquisition or referrals report in the analytics service you already use. Look at the referrer, source and medium fields for visits from AI assistants. Depending on the product and how a visitor opened a link, an assistant's website or app domain may appear as the source. The names can change, and a visit without a usable referrer may instead be grouped as direct or unattributed traffic.

Record what you find over time rather than treating one week's handful of visits as a trend. For each source, compare the landing page, engaged visits and enquiries with the rest of your website traffic. A small business might see a large percentage rise from a very small starting number; the count and the outcome matter more than the percentage on its own.

If AI referrals are not visible, that does not prove that an assistant has never mentioned your business. Someone may have seen the recommendation and searched for your company separately, or the visit may not carry an identifiable referrer. Analytics is useful for measuring clicks that reach your site, not every influence that happened before them.

Make the website useful when somebody arrives

There is no special paragraph that guarantees an AI will recommend your business. The practical basics remain refreshingly unglamorous: explain what you do, who you help, where you work and how customers can take the next step.

• Keep service descriptions, locations, opening hours and contact details accurate.
• Answer real customer questions in clear language, with enough detail to be useful.
• Make pricing, availability and the limits of an offer easy to understand where you can.
• Check that important pages work well on a phone and make it obvious how to enquire.
• Continue looking after conventional search, local listings and the sources customers already use.

If an AI referral lands on a service page, that page should explain the service rather than make the visitor work backwards from a vague homepage. The same is true of a referral from Google, a directory or a recommendation from a colleague.

How Can YouTube Help Your Business Appear in AI Search? →

Google Maps Is Becoming an AI Agent — Why Local Businesses Should Care →

Measure the journey, not the headline

AI assistants are becoming another way people discover information, and it is sensible to check whether they send useful visitors to your website. But the BrightEdge figures do not mean Google has stopped mattering or that a small business needs to buy an AI-ranking package. No reputable provider can guarantee that an independent AI system will recommend a particular company.

Keep the measurement chain in view: Visibility → Visit → Engagement → Enquiry → Customer. Improve the parts you control—accurate information, useful pages and a clear route to contact you—then use your own analytics and enquiries to decide whether AI referrals are becoming meaningful for your business. There is no need to abandon search work that is already helping customers find you.

Sources and further reading

BrightEdge: ChatGPT referral traffic and its August 2026 dataset →

BrightEdge AI Market Pulse →

Matomo: AI chatbot traffic guide for web analytics in 2026 →

Found this useful? Forward it to someone who might too.