AI traffic and sales: how to assess the value of visits

AI answers can reduce the number of visits to a website, but some visitors arrive after getting a general answer and choosing a suitable solution. For a business, the question is not about the click itself: does this visit bring a qualified inquiry, payment, and profit?

One client case shows why this hypothesis is worth testing. Its figures look convincing, but they do not yet provide a reliable forecast for other websites: the observation period, conversion definitions, and attribution rules are unknown.

What the figures from one case study show

An SEO project analysis compared about 20 000 organic visits and 359 conversions with 180 visits from ChatGPT and 44 conversions. The authors also estimated that only around 80–100 of the 359 organic inquiries were actually qualified. These are one company’s figures for its own funnel, not an industry benchmark.

Dividing the stated conversions by the visits gives ChatGPT a share about 13.6 times higher. The case study itself claimed five times the value. Without the original table, it is impossible to know exactly what was compared: different conversions, periods, or funnel stages. So neither the 5-fold nor the 13.6-fold difference can be considered a proven market coefficient. The base of 180 visits is also too small to support a confident conclusion.

The same account also noted that AI visitors spent two to three times longer on the site, and that some visits could have been counted as direct. According to the company, one client paid for a service after a single conversation and said they had found it through AI in Google. These are valuable signals to measure again, but the account does not specify the observation period or the number of such clients. Google also says that visitors who click through from AI Overviews are more likely to spend longer on the site; this does not mean that every such visit will end in an inquiry or sale (Google Search Central documentation).

Track funnel progression, not just forms

A submitted form is not yet a lead, and a lead is not a customer. In the case mentioned, around 80–100 of 359 inquiries were considered qualified: the rest may not have been a good fit for the service or price. If your report ends at the form, it shows only the top of the funnel and overstates the channel’s commercial value.

Use the same stages and definitions for each source:

• submitted inquiry, excluding spam and duplicates;
• qualified lead that fits the service and budget;
• sales proposal or consultation;
• paid order, its revenue, and margin.

Compare sources at each stage. For example, an AI channel may produce more qualified inquiries per 100 visits but a lower average order value. If a deal takes several months, wait for the cohort to mature: an early comparison will favor channels with short sales cycles.

Tag sources and account for lost attribution

Start with the Traffic acquisition report in GA4. The Session source and Session source / medium dimensions show the source of the current session. To track business results, pass the source to your CRM with the inquiry and store information about both the first touch and the last visit separately. In GA4, session-based and event-based reports use different attribution contexts, and the selected model affects key events (GA4 traffic acquisition help, help on scopes and attribution).

Visits from ChatGPT Search can be identified using the UTM tag utm_source=chatgpt.com, if the site allows the OAI-SearchBot crawler to access it. For other services, check the source domains and don’t combine all visitors into a single “AI” row. In the specific case, inquiries were attributed to ChatGPT, Gemini, Perplexity, and Claude; Claude generated the most inquiries only in this dataset, with no sample size or comparison period given. This result does not rank the platforms for the whole market. If the referrer is lost, a visit may be classified as Direct. Add the question “How did you hear about us?” to your form and compare the answer with your web analytics. GA4 classifies a session as direct if no referrer data is available; GA4 help on direct traffic describes this rule. The ChatGPT Search tag and OAI-SearchBot requirements are provided in the OpenAI guidance for publishers.

As of August 31, 2026, the Generative AI Performance report has been rolled out in Search Console for AI Overviews and AI Mode. It shows link impressions and lets you view data by page, country, and device; if there are few AI impressions, the report may not appear. This is a way to track visibility in Google Search, not to measure inquiries, payments, or lead quality (Search Console report description).

Don’t attribute every traffic drop to AI Overviews

If organic visits decline, compare like-for-like periods and page groups: clicks and impressions in Search Console, sessions in GA4, and inquiries and sales in your CRM. Check seasonality, demand for branded queries, page crawlability, changes in search results, and whether the form is working. A drop coinciding with the appearance of AI answers does not, by itself, prove a causal link.

It is also useful to review inquiries by cohort. Evaluate services with long sales cycles only after leads have had time to go through the usual sales stages. Record the source, date of first contact, qualification, deal amount, and payment status. This will show whether a higher share of qualified leads offsets the drop in overall visits.

Improve the page for the reader, not for a keyword percentage

A concise answer, clear comparison, and first-party data are more useful than a long text repeating general advice. A table helps when you need to compare prices or features; a list works for a sequence of steps. Add facts and examples from your own work, but explain the period and sample they relate to.

Check the technical basics: whether the page is accessible to crawlers, indexability, correct title and H1, canonical, internal links, image alt text, and whether structured data matches the page content. These are standard SEO and usability fundamentals, not a guarantee of being cited. Google says that AI Overviews and AI Mode do not require special markup or separate optimization; a page should meet the usual search requirements, but visibility is still not guaranteed (Google’s official guidance).

There is no confirmed percentage for a fixed “ideal keyword density” that will get you included in an AI answer. Use a language model as an editor: ask it to find unclear passages, unsupported claims, and missing comparisons. Its estimate of the chance of being cited does not replace Search Console, analytics, and actual inquiries.

Measurement plan for the coming month

1. Record current figures for organic search and each notable AI source: visits, forms, qualified leads, and payments.
2. Check that the “conversion” event in analytics means the same thing for all channels and excludes tests and duplicates.
3. Pass the visit source to your CRM. Also save the client’s answer about where they heard about the company.
4. Compare quality and revenue across equivalent cohorts, taking sales cycle length into account.
5. Once a month, compare trends with Search Console and note changes to pages, search results, and demand that could have affected the outcome.

This will show whether AI visits are indeed fewer but more likely to bring in suitable customers. Until then, a high conversion rate in a small sample is a reason to set up measurement, not a promise of sales growth.