The Search Engine Nobody Searches: What My Own Traffic Data Says About AI Search Visibility
- Nico Dudli
- 14 minutes ago
- 7 min read
A case study in AI search visibility — and what it takes to rank in AI Overviews and AI Mode without losing SEO.
TL;DR (Personal Data Analysis, Backed by TechNovice's July 2026 Traffic Logs)
In July 2026, TechNovice logged 26,587 requests from OpenAI's ChatGPT-User agent — one of several AI bots tracked; combined across OpenAI, Perplexity, and Claude, total AI bot traffic reached 29,395 hits. OpenAI describes ChatGPT-User as a user-triggered fetcher used for certain actions in ChatGPT and Custom GPTs — not as its automatic search crawler. Human visitors that same month: 763 according to Wix Analytics.
Google Search Console's dedicated Generative AI performance report recorded 59,200 impressions for TechNovice in AI Overviews and AI Mode in July. Google defines these impressions as occasions when links from the site were shown to users within its generative AI search features — a measure of AI-search visibility, not visits and not necessarily citations inside the generated text.
SEO still has a place — Google says its generative AI features continue to run on core Search ranking and quality systems. But for a growing share of queries, ranking well no longer means a human ever sees the page — it means an AI system retrieves it, evaluates it, and answers on the user's behalf. That additional layer has a name: GEO (Generative Engine Optimization).
This is a single-site case study, not a universal benchmark — but the data is real, and the pattern is one every content owner should check for themselves.
I Started Writing for Humans. My Own Logs Told Me Otherwise.
I built TechNovice as a blog for people. Real readers, real clicks, real time-on-page. That was the goal for the first year.
Then I looked at the actual server logs instead of the pageview counter — and the numbers stopped fitting the old framework.
Research Box: Tester: Nico Dudli / TechNovice. Data sources: Wix bot access logs, Wix Analytics traffic report, Google Search Console Generative AI performance report, and Microsoft Clarity session data. Period: July 1–31, 2026. Self-funded; no client sponsored this analysis.
In July, TechNovice logged 26,587 HTTP requests carrying OpenAI's ChatGPT-User agent. OpenAI describes ChatGPT-User as a user-triggered fetcher that may access a page during certain actions in ChatGPT or Custom GPTs. It is different from OAI-SearchBot, which OpenAI uses for surfacing websites in ChatGPT Search, and from GPTBot, which may crawl content for model training.
That distinction matters: 26,587 ChatGPT-User requests are not 26,587 unique ChatGPT users, questions, or citations — they're server requests logged in TechNovice's access log.
The 26,587 figure covers only one bot type. Across all tracked AI bots — OpenAI's user and search bots, Perplexity's user and training bots, and Claude's training bot — TechNovice logged 29,395 combined hits in July.
Human traffic for the same 31 days was 848 pageviews and 763 unique visitors according to Wix Analytics. Microsoft Clarity reported 306 unique users for the same period. Because these tools use different measurement and filtering methods, I treat them as separate baselines rather than assuming that either represents a perfect count of human readers.
Numerically, that's roughly 35 ChatGPT-User requests for every Wix unique visitor, or about 87 for every unique user Clarity reported — a scale of machine-initiated access, not an AI-to-human audience ratio.
Separately, Google's Generative AI performance report recorded 59,200 impressions for TechNovice in AI Overviews and AI Mode during July. Unlike crawler requests, this metric represents actual visibility in a user-facing AI search product.
My Thesis: AI Search Visibility Isn't the Same as a Click
Recent clickstream research illustrates the shift. SparkToro, using Similarweb data, reported that 68.01% of US Google searches in the first four months of 2026 ended without a click — a figure covering the whole modern results page, not only AI.
What has changed is that a website can now create value without receiving the visit itself. A user may see information in a featured result, an AI Overview, AI Mode, or another search interface, while the original publisher gets an impression but no click.
SEO therefore still matters. Google's own guidance says its generative AI features continue to rely on its core Search ranking and quality systems. Crawling, indexing, useful content, technical accessibility, and clear information remain fundamental.
But publishers now have an additional visibility layer to measure: whether their content is surfaced, referenced, or used to ground answers inside AI-driven search experiences. My July data suggests that this layer has already become significant for TechNovice.
AI Trust Is the New Backlink
Backlinks haven't stopped mattering — Google's generative AI features still run on its existing Search ranking and quality systems, so established SEO signals stay in the picture.
But something broader now decides whether a system confidently uses your page: can it verify what you're claiming?
That means publishing claims that are specific, internally consistent, and easy to check. Original measurements, named sources, transparent methodology, clear authorship, and stated limitations make a page more useful — to human readers and to systems retrieving information alike.
That's the GEO opportunity I care about. Not a new ranking trick, but removing the ambiguity a retrieval system has to resolve before it can use your facts.
For product and shopping queries, that means product information, availability, specifications, testing evidence, and business details need to be easy to identify and current. No publisher controls exactly why an AI system picks one source over another — but a publisher can control how easy that source is to trust.
How to Rank in AI Overviews and AI Mode — Without Losing SEO
If you run a website — from a solo craftsman to a SaaS company — I would focus on five things:
Think beyond keywords and include the actual questions your audience needs answered. Traditional search terms still matter, but clear answers to real user problems make content useful across classic search and conversational interfaces alike.
Keep structured data — but don't treat it as an AI-ranking hack. JSON-LD is still useful for describing entities and making pages eligible for supported search features. Google's current guidance explicitly states that structured data is not required for AI Overviews or AI Mode, and that there is no special schema.org markup for generative AI search. The visible page still needs to contain the facts you want systems and users to understand.
Treat different AI ecosystems separately. Google AI Overviews and AI Mode draw from Google's Search systems. Microsoft reports AI citations from Bing and Copilot inside Bing Webmaster Tools. ChatGPT has its own OAI-SearchBot for Search, while ChatGPT-User represents certain user-triggered page accesses. Don't lump all AI crawlers and search products into one channel.
Make important facts explicit. Business names, authors, dates, product specifications, contact information, test methodology, and primary evidence should be stated clearly in visible text instead of hidden inside images or vague marketing copy. That improves both human comprehension and machine retrieval.
Build pages that work for both humans and machines. Good visual design still matters, but important information should also survive outside the visual layout: clean HTML, crawlable text, descriptive headings, internal links, and consistent entity information give search and retrieval systems something reliable to work with.
The Obvious Pushback — And Why Part of It Is Right
The strongest objection to GEO is that many of its recommendations sound suspiciously like good SEO: create useful content, make pages crawlable, use clear structure, establish expertise, and keep information accurate.
That criticism is valid. Google itself says the same SEO fundamentals continue to apply to AI Overviews and AI Mode, and that publishers don't need special AI markup or a separate way of writing for generative search.
Where I think something has genuinely changed is measurement and distribution. A publisher can now appear thousands of times inside an AI-generated search experience without those impressions producing traditional pageviews. At the same time, user-triggered AI systems can fetch individual pages directly without behaving like a conventional search crawler.
GEO doesn't replace SEO. It's the layer on top: understanding how your existing content gets discovered, interpreted, retrieved, and represented now that the interface between the user and the web is, increasingly, an AI-generated answer. My July logs are the proof that layer already exists for TechNovice — whether or not the SEO industry has fully priced it in yet.
FAQ
Is ChatGPT-User the same as OpenAI's search crawler?
No. ChatGPT-User fires on user-triggered actions inside ChatGPT and Custom GPTs. OAI-SearchBot handles ChatGPT Search; GPTBot handles training.
Do 26,587 ChatGPT-User requests mean 26,587 people read my site?
No — that's raw server requests, not readers, questions, or citations.
Do the 59,200 Google impressions mean the content was quoted inside an AI answer?
No. It means the link was shown. Visibility, not citation.
Does structured data (JSON-LD) improve AI Overview or AI Mode visibility?
No dedicated markup exists for it, and Google says structured data isn't required for AI Overviews or AI Mode. It still earns you other Search features — just not this one.
Is SEO becoming irrelevant because of GEO?
No. AI Overviews and AI Mode run on the same core Search ranking systems. GEO adds a layer; it doesn't remove the foundation.
Where does Microsoft/Bing fit into this?
Bing Webmaster Tools now reports AI citations from Bing and Microsoft Copilot separately from Google's report. Publishers should treat Google, Microsoft, and OpenAI's systems as three distinct measurement surfaces rather than one combined "AI channel."
I write about AI search behavior and traffic patterns regularly — more of my analysis is on HackerNoon.
Methodology note: TechNovice-specific figures in this article come from July 2026 data collected through Wix server bot logs, Wix Analytics, Google Search Console, and Microsoft Clarity. No third-party GEO tracking tool was used for these site-level figures. HTTP bot requests, Search Console AI impressions, and analytics users are different measurement units and should not be interpreted as equivalent audiences, searches, or citations.
Comments