Check whether AI can see your business
Run a free audit
You open GA4 and see a sharp drop in traffic from chatgpt.com. Your first thought: AI has stopped recommending the business, or your GEO work has become less effective. Before changing the budget, check whether visits have decreased or only their source attribution has changed.
In mid-October 2026, public observations suggested that some ChatGPT links were not passing the information needed to identify visits. That is a reason to check analytics. It does not prove that every decline has the same cause.
What was observed in October 2026?
OpenAI announced GPT-6 with Intelligent UI on October 7, 2026. The announcement confirms the rollout of interactive answers. It does not, by itself, confirm deliberate removal of visit-source information. Sources in this article were checked on October 15, 2026. Source: OpenAI announcement.
How can a ChatGPT visit end up in Direct?
Attribution means assigning a visit or action to a source. A website can identify a visit from ChatGPT through URL parameters such as utm_source=chatgpt.com, or through information about the referring page, known as the referrer.
A missing UTM parameter is not enough to classify a visit as Direct. If the browser passes a usable referrer, the source can still be identified. The problem arises when there is no campaign tag, referrer or other available signal.
According to MDN, using noreferrer when opening a link omits the Referer header. The noopener option limits the new tab's access to the window that opened it; it does not itself remove the referrer. Source: Window.open documentation.
| Signals available on arrival | What can be established |
|---|---|
| UTM identifying ChatGPT | The URL contains a source tag; also check the medium and channel rules. |
| No UTM, usable referrer | The source may still be identified without the parameter. |
| Neither UTM nor referrer | The visit may appear as Direct; missing signals alone do not reveal its origin. |
| Source identified, channel assigned differently | Check classification. Identifying the source and assigning a channel are separate issues. |
The table explains the mechanism, rather than test results. The final record also depends on the configuration and attribution history in the report.
Google explains that (direct) / (none) represents traffic without a clearly identified source. It can include visits from documents, untagged links or redirects. Direct is therefore not a counter of AI visits. Source: GA4 documentation on Direct.
How to investigate the drop in GA4, step by step
1. Establish exactly what decreased
Start with the Traffic acquisition report and the Session source / medium dimension. Check combinations containing chatgpt.com and the historical chat.openai.com. Do not limit the analysis to Referral: that channel label may not include every identified visit. Source: GA4 Traffic acquisition report.
Compare sessions with sessions. Do not compare First user source with Session source. Record the date of the drop and check for changes to analytics, consent banners, redirects or reporting rules. Dates from public analyses are clues to investigate, not a condition for a certain diagnosis.
2. Compare the same landing pages
Select URLs that received identified ChatGPT visits before the decline. In an exploration, combine landing page with Session source / medium. For the same group of URLs, compare identified ChatGPT traffic, Direct and all sessions.
Use periods of equal length with a similar weekday pattern. Separate country, device and page type. A Direct increase on the homepage says little about missing attribution on a previously cited guide.
3. Test a visit from an actual link
Click a link to your website directly in a ChatGPT answer. Record the date, device, environment and link type. Ask the analytics owner to check the arrival URL, redirects and referrer passed with the first request and analytics event.
Test a regular answer link separately from a link in an additional panel, if present. Pasting the URL into the address bar does not reproduce a ChatGPT click. One test confirms only the conditions tested. Label test visits so they are not counted as marketing results.
4. Check alternative causes
Direct may increase because of an untagged newsletter, an offline campaign, consent changes or a redirect problem. Also check the analytics tag and purchase tracking. If all sessions and orders decline, missing source attribution alone does not explain the entire change.
Sample data: a source declines while visits remain similar
Suppose you analyse the same group of product guides across two comparable weeks. The sample data explains the method and does not represent results for any brand.
| Sessions on the same pages | Before | After | Difference |
|---|---|---|---|
| Identified ChatGPT source | 420 | 150 | -270 |
| Direct | 780 | 1,020 | +240 |
| Other sources | 2,800 | 2,800 | 0 |
| All sessions | 4,000 | 3,970 | -30 |
Identified ChatGPT traffic fell by 270 ÷ 420 × 100 = 64.3%. All sessions decreased by 30 ÷ 4,000 × 100 = 0.75%. The 240-session Direct increase is close to the decline in identified ChatGPT traffic. That is a good reason to investigate missing source information.
You cannot simply assign those 240 sessions to ChatGPT and claim that measurement has been recovered. The table does not link individual visits. Direct could have increased for other reasons, while some actual ChatGPT traffic could have declined.
How to report results when the source is less clear
Report identified ChatGPT sessions, the Direct change on the pages examined, all sessions and business actions separately. Note a suspected break in measurement continuity and the test results. If you produce an estimate, state its assumptions and keep it separate from measured data.
ChatGPT plus Direct can provide a secondary reference for selected pages. Do not label the sum as AI traffic. It cannot show which visits were actual referrals.
Watch for the reverse situation too. Restoring tags can increase visits attributed to ChatGPT without increasing total visits. Record measurement changes so improved source identification is not presented as newly acquired traffic.
AI visibility and visits answer different questions
A brand appearing in an answer, a page citation, a visit and a purchase are four different events. Answer monitoring checks whether AI considers the business for selected customer needs. Website analytics records visits and actions. Neither measurement alone reconstructs the entire purchase journey.
Semly offers brand monitoring in AI answers and analysis of cited sources. You can compare these observations with website analytics. Monitoring does not restore information the browser never passed, and it does not provide access to customers' private conversations. Public description of Semly features.
Learn more about separating mentions, recommendations and citations: How to measure brand visibility in AI.
Where to start today
Select a few pages that previously received ChatGPT traffic. Check whether Direct increased when identified visits declined, and test an actual click. The result will show whether there is a basis for investigating missing attribution or whether you need to look for a real decline in visits.
Base decisions about GEO changes on brand presence, content quality and business outcomes as well. A chart of traffic attributed to chatgpt.com alone is not enough to assess all the work.
Sources and scope
This article draws on public observations and documentation. October's observations are not treated as confirmation of a permanent OpenAI policy. The diagnostic guide and sample data are editorial material. Sources were checked on October 15, 2026.
Share:
