Discussion starter: an SEO measurement question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- SEO Analytics
Apps and in-app browsers strip referrers from AI traffic. Has any method for estimating this dark AI traffic held up for you?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
No method gives you a precise figure for dark AI traffic, and I would be suspicious of any that claims to. What holds up is triangulation: combine what you can identify directly (referrers, and parameters some assistants add to links), patterns in direct traffic that line up with AI citations, user-triggered fetches in your server logs, and self-reported attribution. Together they give you a range and a direction. That is enough to decide whether AI matters for your site and whether it is growing; it is not enough to calculate a return on investment to the pound.
Where the traffic goes dark
- App and desktop clients that open links without passing a referrer.
- In-app browsers that strip or replace referrers.
- Copied URLs, where a user copies a link from an answer into a browser.
- Google’s own AI features. Clicks from AI Mode arrive as ordinary Google organic traffic, and GA4’s AI Assistant channel excludes AI Overviews and AI Mode.
- Influence without a click, where someone reads an answer and later searches your brand or types your URL. This is not traffic loss from referrers; it is influence you cannot attribute at all.
Methods, and how much weight each deserves
Method What it captures Reliability Referrers and assistant-added parameters Identifiable AI sessions. ChatGPT, for example, adds a utm_source=chatgpt.comparameter to many linksHigh for what it captures, but a floor, not a total Direct traffic to deep pages Sessions landing on article or product URLs with no referrer, which people rarely type Moderate; also includes bookmarks, email clients and other dark sources Cited versus uncited page comparison Extra direct traffic on pages that AI answers cite Moderate, if you have a good citation list User-triggered fetches in logs Requests from agents fetching a page because a user asked, such as OpenAI’s ChatGPT-User Good signal of AI interest; not a count of human visits Self-reported attribution Customers saying they found you through an AI assistant Good for direction; biased by memory and form design A triangulation approach that has held up
- Establish the identifiable floor. Build a channel definition that combines GA4’s AI Assistant channel with your own rules for known referrers and parameters. Track sessions, conversions and landing pages.
- Build a citation list. Use Bing Webmaster Tools’ AI Performance report, which shows cited pages for Copilot and Bing’s AI summaries, and your own prompt tracking, to list pages known to be cited by assistants.
- Compare direct traffic on cited and uncited pages. Take deep pages of similar type and traffic. If direct sessions on cited pages rise faster than on uncited pages after they start being cited, the difference is a reasonable estimate of dark AI traffic to those pages.
- Use logs as a leading signal. A rising number of user-triggered fetches of a page usually precedes, or accompanies, human visits from that assistant. Track the ratio of identifiable AI sessions to user-triggered fetches on pages where both appear, and use it cautiously to estimate visits where only fetches are visible.
- Ask customers. Add an AI assistant option to “How did you hear about us?” and compare its share with the identifiable AI share in analytics. The gap between the two is one estimate of how much is hidden.
- Report a range. Present identifiable AI traffic as the floor and your triangulated estimate as the upper bound, with the method stated.
Hypothetical example:
A software company sees 1,200 identifiable AI sessions a month. Its “How did you hear about us?” field shows 6% of new sign-ups choosing an AI assistant, while identifiable AI sessions account for 2% of sign-up sessions. Direct traffic to its forty most-cited documentation pages has risen by about 40% over six months, against 10% for comparable uncited pages. The team reports AI-influenced traffic as “at least 1,200 sessions a month, likely two to three times that”, with sign-up share between 2% and 6%, and explains each number’s source. Leadership gets a defensible range instead of a false precision.
What not to do
- Do not reassign all direct traffic growth to AI. Other sources of dark traffic, such as messaging apps, email clients and privacy tools, also grow.
- Do not multiply identifiable AI traffic by a factor someone published for another site. The ratio depends on your audience’s devices, apps and the assistants they use.
- Do not mix Google’s AI Mode into the AI Assistant channel estimate. It sits in organic search, and Search Console’s generative AI report shows impressions, not clicks, so it needs separate treatment.
Improve what you can capture
Some darkness is avoidable. Make sure your analytics keep UTM parameters on landing, including through redirects. Check that consent handling does not drop the landing page and referrer for users who consent after the first page view. Keep URLs stable so that links in older answers still land on the right page, and make sure redirects preserve query strings. Each of these recovers a little of the identifiable share.
Watch for this:
User-triggered fetchers generally ignore robots.txt, because a user requested the fetch. If you block AI crawlers in robots.txt, you may still see these fetches in logs. Do not read them as crawling, and do not count them as visits.
Identify what you can, estimate the rest with two or three independent methods, report a range with the method attached, and focus on the direction of change. If every method points up, AI matters to you and is growing, and that is the decision-relevant finding.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.